Methods and systems for predicting the probability of venous thromboembolism in patients with sepsis

By using a computer system to assess coagulation activation bias and treatment exposure in sepsis patients, and combining this with genetic factors, the probability of venous thromboembolism can be predicted. This solves the problem of the inability to individually assess risk in existing technologies, and enables reasonable treatment frequency and accuracy.

CN123120331APending Publication Date: 2026-11-17THE PEOPLES HOSPITAL SHAANXI PROV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611636622.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing risk assessment methods for venous thromboembolism cannot accurately distinguish individual risks among sepsis patients in intensive care units, leading to inappropriate treatment frequencies, including both long-term overtreatment and some patients who were not given early warnings.

Method used

By obtaining the measured coagulation activation amount and current status test quantity of sepsis patients, the activation bias is calculated and divided by the bias scale to obtain the unexpected activation residual. Combined with treatment exposure records and genetic factors, the probability of venous thromboembolism is predicted, and a reasonable treatment frequency is set.

Benefits of technology

It enables individualized assessment of venous thromboembolism risk among sepsis patients in the intensive care unit, reducing false alarms and missed alarms, and allowing for appropriate triggering frequencies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120331A_ABST
    Figure CN123120331A_ABST
Patent Text Reader

Abstract

This application discloses a method and system for predicting the probability of venous thromboembolism in sepsis patients, relating to the field of medical data processing technology. It addresses the problem of inconsistent thrombosis risk assessment and the inability to pre-determine the frequency of triggered interventions in sepsis patients within the intensive care unit. The method, executed by a computer, includes: acquiring the measured coagulation activation level and current status test value of the sepsis patient; obtaining the expected coagulation activation level from the current status test value; obtaining the activation bias by subtracting the logarithm of the expected coagulation activation level from the logarithm of the measured coagulation activation level; obtaining the bias scale from various detection variations, coefficient estimation uncertainty, inter-patient dispersion, and intra-patient biological fluctuations; dividing the activation bias by the bias scale to obtain the unexpected activation residual; and obtaining and outputting the probability of the patient developing venous thromboembolism within a preset scope from the unexpected activation residual. This application can be used for thrombosis monitoring and tiered treatment of sepsis patients.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of medical data processing technology, and more specifically, to a method and system for predicting the probability of venous thromboembolism in patients with sepsis. Background Technology

[0002] Patients with sepsis have a higher rate of venous thromboembolism during their hospitalization in the intensive care unit than the general hospitalized population. Early-stage sepsis lacks specific clinical signs, and diagnosis relies on imaging examinations. Since imaging examinations cannot be performed on all hospitalized patients daily, clinical risk assessments are conducted first to determine the intensity of prevention and monitoring, and limited testing resources are allocated to high-risk individuals.

[0003] Current risk assessments for venous thromboembolism (VT) often combine several clinical items and routine laboratory results with fixed scores or cutoff points for interpretation. These scores and cutoff points are typically determined for the general inpatient population. When applied to sepsis patients in intensive care units, the vast majority of inpatients in the same ward are grouped into the same category, leading to a high degree of consistency in interpretations within this population. However, subsequent comparisons of treatments implemented based on these assessments with actual outcomes reveal a range of cases: some patients were treated as high-risk for an extended period without developing VT, while others were not alerted to the risk prior to detection.

[0004] Therefore, how to provide hourly, differentiated probabilities of venous thromboembolism among sepsis patients in the intensive care unit and pre-set the frequency of interventions is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a method and system for predicting the probability of venous thromboembolism in sepsis patients, at least solving the technical problem of how to provide hourly, distinct probabilities of venous thromboembolism among sepsis patients in intensive care units, and how to pre-set the frequency of interventions.

[0006] In a first aspect, embodiments of this application provide a method for predicting the probability of venous thromboembolism in patients with sepsis, the method being executed by a computer, the method comprising:

[0007] Obtain the measured coagulation activation volume and current status test results of sepsis patients;

[0008] The expected coagulation activation amount is obtained from the current state test quantity;

[0009] The activation deviation is obtained by subtracting the logarithm of the expected coagulation activation amount from the logarithm of the measured coagulation activation amount.

[0010] The deviation scale is obtained from the detection variation of the measured coagulation activation quantity and the current state test quantity, the estimation uncertainty of each coefficient used to obtain the expected coagulation activation quantity, the inter-patient dispersion, and the intra-patient biological fluctuation. The deviation scale increases with the increase of any part of the uncertainty corresponding to the detection variation, the estimation uncertainty, the inter-patient dispersion, and the intra-patient biological fluctuation.

[0011] Dividing the activation bias by the bias scale yields the unexpected activation residual;

[0012] The probability of occurrence is obtained from the unexpected activation residual and output.

[0013] The probability of occurrence is the probability that the sepsis patient will develop venous thromboembolism within a preset field of vision.

[0014] Optionally, the method further includes: obtaining treatment exposure records of the sepsis patient, and obtaining the cumulative exposure dose of each treatment exposure from the treatment exposure records; the expected coagulation activation dose increases with the increase of the cumulative exposure dose of treatment exposures that increase coagulation activation via the coagulation activation channel, and decreases with the increase of the cumulative exposure dose of treatment exposures that inhibit coagulation activation via the coagulation activation channel.

[0015] In the above scheme, a therapeutic exposure is converted into a continuous amount that increases with its duration of presence and gradually decreases after withdrawal. This amount enters the expected coagulation activation amount in both directions of increase and inhibition. The rise and fall caused by the therapeutic action in the measured record is thus removed before subtraction, so that the expected value moves with the rise and fall of the therapeutic effect in the body and does not jump with the start and end times on the medical record.

[0016] Optionally, the cumulative exposure dose increases with the continuation of the treatment exposure and decreases with the withdrawal of the treatment exposure, and the rate of increase and decrease is taken to minimize the sum of the absolute values ​​of the mean changes in the unexpected activation residuals before and after the treatment exposure is initiated and before and after its withdrawal.

[0017] In the above scheme, the rate at which the cumulative exposure dose is established and decreases is determined by the mean change in the unexpected activation residual before and after the treatment exposure is initiated and withdrawn. The amount on which this is based can be directly measured in the calibration cohort without waiting for the outcome. The rate at which the dose is counted is therefore synchronized with the process of the exposure taking effect and receding in vivo, so that the start and end of the treatment action itself no longer cause a systematic shift in the judgment value.

[0018] Optionally, obtaining the expected coagulation activation amount from the current state test quantity includes: the current state test quantity includes an inflammation intensity test quantity, a clearance capacity test quantity, and an inhibitory reserve test quantity; the clearance capacity test quantity characterizes the ability of the measured coagulation activation amount to be cleared from plasma; the inhibitory reserve test quantity characterizes the amount of thrombin generation inhibitors; the expected coagulation activation amount increases with the increase of inflammation intensity characterized by the inflammation intensity test quantity, increases with the decrease of clearance capacity characterized by the clearance capacity test quantity, and increases with the decrease of the amount of inhibitory reserve test quantity.

[0019] In the above scheme, inflammation intensity, clearance capacity, and inhibitory reserve each play a driving role and two corrective roles: the driving term raises the expected value according to inflammation intensity, the clearance correction term includes the accumulation caused by clearance obstruction in the expected value, and the inhibitory reserve correction term includes the increased thrombin generation after inhibitor depletion in the expected value; the increase outside of these three factors is left in the activation bias, thus enabling the same test result to give different biases among patients with different organ function levels.

[0020] Optionally, obtaining and outputting the occurrence probability from the unexpected activation residual includes: repeatedly obtaining the unexpected activation residual over a preset time period; weighting the unexpected activation residual of each time period and the time periods preceding that time period to obtain the duration level of that time period; obtaining the time period occurrence rate of that time period from the duration level of that time period; and synthesizing and outputting the occurrence probability from the time period occurrence rates of each time period within the preset horizon.

[0021] In the above scheme, the residual is not directly entered into the interpretation based on the value of a single time period. Instead, the persistence level is first synthesized from the values ​​of each time period within the window. The impact of an isolated detection jump is weakened by the values ​​of the other time periods within the same window. The persistence level is converted into the occurrence rate of the time period and then synthesized segment by segment along the horizon. This achieves the output quantity covering the entire horizon instead of the value at a single moment. The available time margin for processing is determined by the length of the horizon.

[0022] Optionally, the occurrence rate of the time period is obtained by taking at least one of the current state test values; the effect of the same value taken in two places on the expected coagulation activation amount and its effect on the occurrence rate of the time period are given by independent coefficients.

[0023] In the above scheme, the same test quantity is included by a coefficient in two places: at the expected coagulation activation level, it accounts for the effect of the test value through the coagulation activation channel; at the time period incidence level, it accounts for the part of the effect that does not go through this channel. The former is subtracted first, so the two pathways can be distinguished in the data, and the two effects from the same source are each given a coefficient that can be estimated separately, instead of being combined into a single quantity with an unclear direction.

[0024] Optionally, the weight of each of the unexpected activation residuals in the weighting of the duration level is taken as the autocorrelation of the unexpected activation residuals over the interval between the time period in which they are located.

[0025] In the above scheme, the weights are taken from the measured values ​​of the residual autocorrelation at that interval. The larger the interval and the lower the correlation, the smaller the share of the weights. The weights are determined by the data itself, which realizes that the weighting does not depend on any assumptions about the decay of correlation with interval. The same set of rules provides a window that is appropriate for each application scenario with different rates of residual correlation decay.

[0026] Optionally, the method further includes: obtaining the minor allele doses of the sepsis patient at each functional locus that is effective in the measured coagulation activation level; obtaining a fitness gain from the minor allele dose and the effect of the corresponding functional locus on the measured coagulation activation level; the fitness gain increasing as the product of the minor allele dose and the corresponding effect increases; and the expected coagulation activation level increasing as the fitness gain increases.

[0027] In the above scheme, the minor allele doses of each functional locus are combined into a dimensionless constitutional gain according to their respective effect sizes. This gain is then incorporated into the expected coagulation activation level. The effect size is estimated using the measured coagulation activation level as the dependent variable, and it is on the same logarithmic scale as the subtracted amount, so the subtraction between the two does not require conversion. The patient's congenital activation level is thus removed before the residual is formed, ensuring that the individual's baseline activation level is no longer interpreted as an abnormal increase in the current period.

[0028] Optionally, the method further includes: comparing the occurrence probability with a preset level threshold, and outputting the venous thromboembolism monitoring level for the sepsis patient according to the comparison result; the level threshold is the minimum value that ensures the frequency of patients who have not experienced venous thromboembolism triggering the venous thromboembolism monitoring level does not exceed a preset upper limit.

[0029] In the above scheme, the threshold is not specified by experience, but is obtained by inverse solving on the calibration queue based on the false alarm frequency that the user can tolerate. Taking the minimum value that meets the upper limit maximizes the number of time periods that are alerted under the same tolerance. This allows the frequency of triggering the action to be determined before deployment according to the value given by the user. When the value is changed or the basic risk of the population changes, the same rule can be re-solved.

[0030] Secondly, embodiments of this application provide a prediction system for the probability of venous thromboembolism in sepsis patients, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the prediction method for the probability of venous thromboembolism in sepsis patients as described in the first aspect or any optional method of the first aspect.

[0031] In the above aspects, the measured coagulation activation amount is subtracted from the amount that the patient should have in the current state in the logarithmic domain. The resulting activation deviation only includes the increase from sources other than those listed. This deviation is then divided by the deviation scale synthesized from its own uncertainties, and the resulting unexpected activation residual, with its own uncertainty as a unit, has a known zero distribution. This residual is then window-weighted to become a persistence level, and then synthesized segment by segment along the horizon to obtain the probability of occurrence falling on the probability scale. The probability of occurrence is compared with the threshold obtained by inverse solving the tolerable false alarm frequency to give the venous thromboembolism monitoring level, thereby providing hourly venous thromboembolism occurrence probabilities that are distinct among sepsis patients in the intensive care unit, and allowing the frequency of triggering treatment to be preset. Attached Figure Description

[0032] Figure 1 A flowchart illustrating the method for predicting the probability of venous thromboembolism in sepsis patients provided in this application embodiment;

[0033] Figure 2 This is a schematic diagram illustrating the generation relationship between various data collection quantities and occurrence probabilities provided in the embodiments of this application;

[0034] Figure 3 This is a schematic diagram of the structure of a prediction system for the probability of venous thromboembolism in sepsis patients provided in an embodiment of this application. Detailed Implementation

[0035] The technical solutions in the embodiments of this application will be further described below with reference to the accompanying drawings. The described embodiments are some, but not all, embodiments of this application; other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are all within the scope of protection of this application.

[0036] In this application, sepsis patients refer to patients admitted to the intensive care unit due to host response dysregulation caused by infection; coagulation activation refers to the process by which thrombin is generated and fibrinogen is converted into fibrin, which is then degraded by the fibrinolytic system, and its intensity is characterized by the concentration of the corresponding markers in the plasma; venous thromboembolism is a collective term for deep vein thrombosis and pulmonary embolism; diffuse activation refers to coagulation activation that is widespread throughout the systemic microcirculation without local thrombus formation, in contrast to the generation of fibrinolytic products on the surface of local thrombi; calibration cohort refers to a historical cohort with outcome annotations, on which all parameters of this method are calibrated offline; deployment cohort refers to a cohort of similar patients in the location where this method is to be deployed; at-risk periods refer to the time periods during which patients are still in the intensive care unit and have not yet developed venous thromboembolism, nor have they died or been transferred; the set of time periods obtained by defining the first step of the calibration section below in the at-risk periods of the fitting sample index calibration cohort; and the validation cohort refers to a cohort that is of the same origin as the calibration cohort, does not participate in the calibration, and is used to evaluate the operating characteristics after the parameters are fixed.

[0037] Current risk assessment methods for venous thromboembolism in hospitalized patients are divided into two categories based on the construction of their diagnostic metrics. The first category uses an itemized scoring system, assigning a fixed score to each item such as immobilization, acute infection, history of thrombosis, and malignancy. These scores are then accumulated and compared to a fixed threshold to determine whether prophylactic medication should be administered. The mechanism is that a true result for each item is treated as a constant risk increment, and it is assumed that the increments of each item are additive. The second category uses a coagulation status scoring system, employing three parameters: the international normalized ratio, platelet count, and organ failure assessment score. Each parameter is assigned a score at a fixed cutoff point and then accumulated to determine the degree of coagulopathy in sepsis patients. The mechanism is that the severity of consumptive coagulopathy is used as a proxy for thrombotic risk.

[0038] When the above two approaches are applied to sepsis patients in the intensive care unit, the diagnostic criteria lose their discriminatory power for three reasons: First, the items are saturated. Bed rest and acute infection are both applicable to almost all patients in this population. The sum of these two conditions alone exceeds the intervention threshold, and the variance of the diagnostic criteria can only come from a few unsaturated items. The value distribution is compressed into a narrow range above the threshold, resulting in a degradation of the ranking within the population. Second, the weights are fixed. The risk corresponding to the same item is not the same under different coagulation states, and the fixed score masks this difference. Third, the pathway is not specific. The three items used in the coagulation state scoring are all measures of organ function and consumptive coagulopathy. There is no specific pathway between them and local thrombosis. Their elevation is in the same direction in both disseminated intravascular coagulation and local deep vein thrombosis, so it is impossible to distinguish between the two.

[0039] In one alternative implementation, the measured concentration of the coagulation activation marker or its binarized result is directly used as the judgment value, and interpreted according to a fixed concentration threshold. The limitation of this implementation in this population is that the concentration in sepsis patients is almost constantly elevated, originating from both diffuse activation and fibrinolytic products on the surface of local thrombi; these two factors are superimposed in the same detection value and cannot be distinguished. For the same concentration value, the corresponding risk of local thrombosis differs by more than an order of magnitude in two patients with different levels of inflammation, renal clearance capacity, antithrombin reserves, and treatment exposure. Therefore, the fixed concentration threshold cannot determine the false alarm frequency or the unit of deviation from the individual's normal range.

[0040] Based on the above problems, this application changes the judgment quantity from the measured concentration itself to the deviation of the measured concentration from the concentration that should be present in the patient's current state, and normalizes the deviation according to its own uncertainty so that the judgment quantity obtains a known zero distribution on a unit scale; at the same time, each listed source is included in two places, and its effect on the detection value through the coagulation activation channel is deducted first, and its effect on the thrombosis process itself is included in the incidence rate with an independent coefficient.

[0041] Before proceeding with each step, the correspondence between the perturbations and their effects on which this protocol is based is first presented. Each quantity used in subsequent steps is traced back to one of these perturbations. When inflammation intensity increases, tissue factor expression and thrombin production increase, leading to a higher rate of fibrin degradation product formation in plasma, resulting in an overall upward shift in measured coagulation activation levels. When renal clearance capacity decreases, the rate of fibrin degradation product clearance from plasma decreases, and at the same formation rate, its plasma concentration increases, while thrombin production does not increase. When the reserves of thrombin inhibitors such as antithrombin decrease, thrombin production increases at the same driving force, and activation markers increase accordingly. The foreign body interface of the central venous catheter generates diffuse activation via the contact activation pathway, the intensity of which accumulates with the duration of indwelling and subsides after removal due to intimal repair. Mechanical ventilation generates a weaker similar activation through endothelial stretching, and the increased intrathoracic pressure and immobilization cause venous return obstruction and muscle pump failure. Prophylactic anticoagulation inhibits thrombin production, leading to a decrease in activation markers. Different minor allele doses at various functional sites along the coagulation pathway result in different basal activation levels in individuals, and this difference remains constant throughout life. Once a local thrombus nucleates and grows, the rate of fibrinolytic product formation on its surface increases with the area of ​​the thrombus exposed to plasmin, manifesting as an additional elevation above the levels explained by the aforementioned sources. This protocol aims to identify precisely this additional elevation.

[0042] The values ​​below are those used in this embodiment. For other implementation methods, implementers can set them according to their actual situation.

[0043] See Figure 1The method for predicting the probability of venous thromboembolism in sepsis patients provided in this embodiment is executed by a computer and includes steps S101 to S106.

[0044] S101, obtain the measured coagulation activation amount and current status test amount of sepsis patients.

[0045] S102, the expected coagulation activation amount is obtained from the current state test quantity.

[0046] S103, the activation deviation is obtained by subtracting the logarithm of the expected coagulation activation amount from the logarithm of the measured coagulation activation amount.

[0047] S104, the deviation scale is obtained from the detection variation of the measured coagulation activation amount and the current state test amount, the estimation uncertainty of each coefficient used to obtain the expected coagulation activation amount, the inter-patient dispersion, and the intra-patient biological fluctuations.

[0048] S105, divide the activation deviation by the deviation scale to obtain the unexpected activation residual.

[0049] S106, The occurrence probability is obtained from the unexpected activation residual and output, wherein the occurrence probability is the probability that the sepsis patient will develop venous thromboembolism within the preset field of view.

[0050] See Figure 2 The generation relationship between the above quantities is as follows: minor allele doses synthesize physical gain, treatment exposure records are synthesized into cumulative exposure dose through the forgetting nucleus, and the two, together with the current state test, synthesize the expected coagulation activation amount; the difference between the measured value and the expected value gives the activation deviation, which is divided by the deviation scale to obtain the unexpected activation residual; the residual is weighted by time period to obtain the persistence level, and the persistence level, physical gain, inflammation intensity test, clearance capacity test, and cumulative exposure dose are used to obtain the time period incidence rate. The time period incidence rate of each time period within the field of view is synthesized into the occurrence probability, and the monitoring level is given by the occurrence probability and the number of consecutive time periods exceeding the limit. Figure 2 The same physical gain, the same inflammation intensity test, the same clearance capacity test, and the same cumulative exposure dose each have two destinations: one is the expected coagulation activation amount, and the other is the time period incidence. The effect of each destination is carried by a coefficient, and the two coefficients are independent of each other.

[0051] In step S101, a fixed time period is first used. Divide the hospital stay into time slots, and record the time slots as follows: , No. Each time period is self- to For a period of time, in this embodiment Use a 12-hour period. All inputs and outputs in subsequent steps will be aligned to the specified time period. If the same item has multiple test results within the same time period, use the last result within that time period.

[0052] Measured coagulation activation quantity refers to the measured test quantity characterizing the degree of coagulation activation, denoted as . This includes, but is not limited to, plasma D-dimer concentration, thrombin-antithrombin complex concentration, and soluble fibrin monomer concentration; the activation level in the name refers to the coagulation activation being characterized, and is not limited to a single biomarker. In this embodiment, plasma D-dimer concentration, in mg / L FEU, was measured by immunoturbidimetry, and the data was obtained from the laboratory information system, collected once per time period.

[0053] Current state test quantities refer to the total number of test quantities that reflect the patient's current pathological state and are used to estimate the amount of coagulation activation that should be present in that state. These include, but are not limited to, inflammation intensity test quantities, clearance capacity test quantities, inhibitory reserve test quantities, and test quantities characterizing blood flow stasis or vascular endothelial activation. The latter are entered into formula (2) using a logarithmic ratio of the same type, and their reference values ​​and indices are determined according to the same rules as the first three categories. Inflammation intensity test quantities refer to test quantities characterizing inflammation intensity, denoted as... This includes, but is not limited to, C-reactive protein concentration, interleukin-6 concentration, and procalcitonin concentration. In this embodiment, C-reactive protein concentration, in mg / L, is measured by immunoturbidimetry. The clearance capacity test quantity refers to a test quantity characterizing the ability of the measured activated coagulation amount to be cleared from plasma, denoted as... This includes, but is not limited to, serum creatinine concentration, plasma cystatin C concentration, and estimated glomerular filtration rate (GFR). The estimated GFR increases with increasing clearance capacity, and its sign in formula (2) must be redefined based on its relationship with clearance capacity. In this embodiment, serum creatinine concentration, in μmol / L, is used and determined by enzymatic method. The inhibitory reserve test quantity refers to the test quantity characterizing the reserve of inhibitors for thrombin formation, denoted as... These include, but are not limited to, antithrombin activity, plasma fibrinogen concentration, and platelet count. In this embodiment, antithrombin activity is measured as a percentage (%) using a chromogenic substrate method. All three parameters are obtained from the laboratory information system, with a sampling frequency of once per time period, regardless of the data level.

[0054] Step S101 also involves acquiring the treatment exposure records of the sepsis patient. Treatment exposure records refer to records documenting the start and end times of various treatment exposures, including but not limited to records of central venous catheter placement, mechanical ventilation, and prophylactic anticoagulation administration. They may also include records of parenteral nutrition, extracorporeal circulation tubing for blood purification, and administration of vasoconstrictors. In this embodiment, the first three items are used, with data sources including medical orders, nursing records, and ventilator records, converted to exposure in-place indicators by time period. Take 0 or 1. Number the exposure item and select one of the following: central venous catheterization, mechanical ventilation, or prophylactic anticoagulation.

[0055] Peripheral blood genotyping was performed once upon admission. The genotyping method used was the TaqMan probe method, which is a well-known technique and will not be described in detail here. As other implementation methods, the implementer may use other existing methods, such as time-of-flight mass spectrometry, microarray chip methods, etc. This embodiment does not impose any special restrictions on this. The genotyping results were obtained from a genotyping database.

[0056] The logarithmic coefficient of variation for each test item is converted from the relative standard deviation of the corresponding test item's internal quality control data in that laboratory. The conversion formula is: square the relative standard deviation, add one, take the natural logarithm, and then take the square root. When the relative standard deviation does not exceed 0.30, the difference between the converted value and the relative standard deviation does not exceed 4.5% of the original value, and the original relative standard deviation can be directly substituted. For test items exceeding 0.30, the conversion formula described above is used. In this embodiment, we take... For 0.10, 0.08 0.05 The coefficient of variation is 0.07. The four coefficients are, in order, the logarithmic scale coefficients of variation for the measured coagulation activation quantity, inflammation intensity test quantity, clearance capacity test quantity, and inhibitory reserve test quantity. The above conversion makes the variation of each test item additive on the logarithmic scale, so that they can be included in the synthesis of the bias scale on the same scale.

[0057] The procedures for missing measurements of each quantity are as follows. For the period when the measured coagulation activation quantity is missing, the unexpected activation residual is taken according to formula (6). For the period when the inflammation intensity test quantity, clearance capacity test quantity, or inhibition reserve test quantity is missing, the patient's most recent result is substituted, and the logarithm of the quantity is extrapolated to the missing period according to its forward predictive coefficient. The above substitution ensures that the expected coagulation activation quantity can still be calculated for the missing period, and the determination is not interrupted due to the missed detection of a single test. For patients whose inhibition reserve test quantity has never been measured, the median of the calibration cohort is substituted, and the unmeasured indicator is set. The value is set to 1, and the detection variation of this item is replaced with the input discreteness in the bias scale. This treatment allows patients who have never measured this item to be interpreted on the same scale as those who have measured it, and the bias scale is widened according to the dispersion of the input value itself. If any collection quantity is missing for two consecutive time periods, an insufficient data flag is output, and the patient's level in the current period is not lower than level 1. The number of consecutive missing time periods is determined by the lag when the autocorrelation coefficient first falls below 0.40. This lag is taken as the boundary so that the correlation on which the extrapolation is based has not fallen to an unusable level. In this embodiment, this lag is 2. The above missing value treatment ensures that missing values ​​are neither treated as normal values ​​nor interrupt the judgment.

[0058] Step S101 takes two types of inputs: the measured coagulation activation amount and the current status test amount. This ensures that subsequent steps have both the quantity to be judged and the measurement of the state of that quantity, allowing the judgment to be performed after aligning the two within the same time period. Each quantity is aligned according to a fixed time period length, so that data with different collection frequencies enter the same time base, and the inputs of subsequent steps do not change with the frequency of collection.

[0059] In step S102, the constitution gain is first obtained from the genotyping results, then the cumulative dose of each exposure is obtained from the treatment exposure record, and finally combined with the three types of current status test quantities to form the expected coagulation activation quantity.

[0060] Functional sites refer to sites that are effective in the measured coagulation activation level, including but not limited to sites whose functional effects act on coagulation factors, anticoagulant proteins, fibrinolytic regulatory proteins, or vascular endothelial functional proteins. Sites are selected based on three criteria: their functional effect acts on one of the above four types of proteins; their minor allele frequency in the calibration cohort is not less than 1%; and their logarithmic effect estimate on the measured coagulation activation level in the fitted sample is not zero after multiple comparison correction with an error discovery rate not exceeding 0.05. The Benjamini-Hochberg method is a well-known technique and will not be elaborated upon here. As other implementation methods, practitioners may use other existing methods, such as the Bonferroni method, etc. This embodiment does not impose any special restrictions on this; any correction level not less than 0.01 and not more than 0.05 is acceptable. The coagulation factor V Leiden and prothrombin G20210A loci have extremely low frequencies in some populations, and are usually not selected based on frequency alone. Therefore, the locus set must be re-determined within the deployment cohort according to the above three criteria, rather than being transplanted from an existing locus table of other populations. The candidate loci in this embodiment are the following 10, in order: ABO gene rs8176719, FGG gene rs2066865, F11 gene rs2036914, THBD gene rs16984852, PROC gene rs146922325 and rs199469469, SERPINE1 gene rs1799762, APOH gene rs8178847, NOS3 gene rs1799983, and MTHFR gene rs1801133. The selected loci constitute the functional locus set based on the above three criteria, and the number of these loci is denoted as [missing information]. The summation below is performed only on the selected sites.

[0061] The site number is denoted as , No. The minor allele dose at each functional locus is denoted as The value is 0, 1, or 2; the allele effect size of this locus on the measured coagulation activation amount is denoted as . In this embodiment, physical fitness gain Take each point and The accumulation of From 1 to The summation is performed by multiplying each product point by point and then adding them together, resulting in a dimensionless logarithmic scale. Alternatively, the mass gain can be the sum of the products transformed by a monotonically increasing function, such as the hyperbolic tangent, so that the mass gain tends to saturate when the number of sites is large; as long as the synthesis result increases with each product, the subsequent steps remain unchanged.

[0062] The meaning must be consistent with its purpose: In this method It refers to the logarithmic increase in measured coagulation activation for each additional minor allele, i.e., the effect size on coagulation activation, not the effect size obtained by taking the logarithm of the prevalence ratio of venous thromboembolism. The two have different scales and are not proportional. The effect size is estimated directly within the deployment cohort, without conversion from the disease odds ratio. When locus selection and effect size estimation are performed on the same sample, the effect size of the selected loci is systematically overestimated, and its covariance is systematically underestimated. Therefore, the calibration cohort is randomly divided into half of the patients: the first half is calculated using the three selected loci mentioned above, and the second half is calculated using the set of loci. After the two halves are interchanged, the average of the coefficients and covariance is taken. For loci selected only in one half, the effect size is taken from the estimated value of that half, and the other half is averaged using the same method after being marked as 0. This division is the estimation procedure used in the second step of the calibration segment.

[0063] Constitutional gain separates the lifelong, unchanging genetic component, allowing it to be subtracted from the expected coagulation activation amount beforehand. The portion of the measured value that can be explained by genotype is therefore no longer included in the unexpected activation residual. The effect size is estimated using the measured coagulation activation amount as the dependent variable, placing it on the same logarithmic scale as the expected coagulation activation amount, so that no scale conversion is needed when subtracting the two.

[0064] The cumulative exposure dose for each treatment exposure is extrapolated over time according to the forgetting nucleus, as shown in formula (1):

[0065] Formula (1);

[0066] In the formula, For the first Treatment exposure in The cumulative exposure dose at the end of each time period, dimensionless, with a value not less than 0 and not greater than 1, initial value. Take 0, The cumulative dose of exposure to the same treatment at the end of its previous time period; The time-period coefficient for this exposure is retained and is dimensionless; For this exposure in The exposure presence indicator for each time period is either 0 or 1. The duration is expressed in hours. Let be the time constant of the forgetting nucleus for this exposure, in hours. This formula is a first-order exponentially weighted moving average: [Exposure duration...] Monotonic approaching 1, after removal press The decay occurs over time periods and the values ​​are bounded, so the recursion does not amplify the error. Depend on and As given in the above formula, and The units are both hours, their ratio is dimensionless, and it remains dimensionless after taking the exponent, so the dimensions on both sides of the equal sign are consistent.

[0067] The cumulative exposure dose increases with the continuation of the treatment exposure and decreases with the withdrawal of the treatment exposure. The rate of increase and decrease is taken to minimize the sum of the absolute values ​​of the mean changes in the unexpected activation residuals before and after the treatment exposure is initiated and withdrawn. The selection rules are as follows. The quantity based on these rules is the unexpected activation residual defined in this method, which can be directly measured within the calibration cohort without outcome labeling. The search is performed with the pathophysiological duration of the exposure's effect on coagulation activation as the initial value, within a range not less than 0.5 times and not more than 4 times the initial value; any value within this range is feasible. In this embodiment, the initial value is 96 hours for central venous catheterization, 60 hours for mechanical ventilation, and 6 hours for prophylactic anticoagulation; after the search... The values ​​were taken at 96 hours, 60 hours, and 24 hours, respectively. The value for anticoagulation was higher than its pharmacokinetic half-life because the expected response rate of coagulation activation to this exposure must match the rate at which the biomarker itself is cleared from the plasma, rather than the rate at which the drug plasma concentration decreases.

[0068] As another implementation, the cumulative exposure dose can also be the ratio of the number of time periods in the most recent fixed-length window to the number of time periods in that window, as long as the amount increases with the continuation of exposure and decreases with the removal of exposure; alternatively, a time constant can be set for both establishment and attenuation, and the two constants can be determined by minimizing the change in the mean residual value before and after establishment and before and after removal, respectively, while the rest of the formula remains unchanged. This implementation allows the establishment end and the removal end to be aligned respectively.

[0069] The cumulative exposure dose changes the therapeutic exposure from an on / off state to a bounded dose that accumulates with the duration of indwelling and gradually diminishes after withdrawal. This ensures that a withdrawn exposure is still counted as a residual dose for the period before its effects have subsided. Consequently, the expected coagulation activation dose varies with the actual in vivo effect of the exposure, rather than abruptly changing with the start and end times recorded in the medical order. The activation component of catheterization and ventilation is recorded with a positive sign, and the inhibition component of anticoagulation is recorded with a negative sign. This allows the same therapeutic exposure record to change the expected coagulation activation dose in two directions, and the portion of the measured value that can be explained by the therapeutic action is therefore deducted in the direction indicated by the therapeutic action.

[0070] The forgetting kernel time constant is determined based on the criterion that the residual does not undergo systematic shift before and after the change of exposure status, so that the rate of inclusion is aligned with the actual rate of onset and resolution of the treatment exposure; the initiation and withdrawal of the exposure therefore do not cause systematic shift of the judgment quantity, and the magnitude of its zero point shift with the treatment action is minimized within the search range.

[0071] The expected coagulation activation level is the coagulation activation level that the patient should have in the current state, denoted as . Its logarithm is shown in formula (2):

[0072] Formula (2);

[0073] In the formula, Reference activation level, i.e., coagulation activation level under reference conditions, unit: Same, the reference state is defined as equal , equal , equal , 0, The values ​​are 0 and the cumulative dose of all three exposures is 0; , , The reference values ​​for inflammation intensity, clearance capacity, and inhibitory reserve are, in order, 50 mg / L, 80 μmol / L, and 90% respectively in this embodiment. The rules for determining these three values ​​are explained in the following paragraph. Inflammation-driven index To clear the correction index, To suppress the reserve correction index, To suppress the unmeasured bias term when the reserve inspection quantity is not measured, For unmeasured indications, take 0 or 1. All of the above quantities are dimensionless and obtained from calibration. and The activation coefficients of central venous catheterization and mechanical ventilation via the coagulation activation channel are, in order: the catheterization activation coefficient and the ventilation activation coefficient. The anticoagulation inhibition coefficient of prophylactic anticoagulation on coagulation activation is given. All three are dimensionless and obtained by calibration. , , , and If all values ​​are greater than 0, their signs are determined by their respective mechanisms of action. If the sign of the estimated value obtained by calibration is opposite to that of the value, the term is determined to be unidentifiable in the queue. The term is removed from the formula and the source is included in the period occurrence rate. The calibration step is repeated. , and The cumulative exposure doses for the three treatment exposures mentioned above are given by formula (1). In the implementation method that does not use genotyping results, In the implementation where treatment exposure records are not used, the cumulative dose of all three exposures is set to 0, and the corresponding components are removed from the independent variable vector. In this case, the expected coagulation activation is given only by the current state test value. Subsequent steps remain unchanged in both cases. Both sides of the equal sign represent the logarithmic value of coagulation activation. When a unit is substituted into a term, the independent variables of the remaining terms are either ratios of the same dimension or are dimensionless themselves. Therefore, all the terms are dimensionless, and the dimensionality is closed.

[0074] The effects of the terms on the right-hand side of the above equation are as follows. The inflammation-driving term takes the power-law form of the logarithmic ratio of inflammation intensity, and... Greater than 0 and less than 1: Activation of the coagulation cascade is limited by substrates and inhibitors; thrombin production tends to saturate when inflammation intensity increases, hence the index is less than 1; the same value range also makes... right The relative error is times The relative error converges during propagation. As another implementation, the inflammation-driven term can be in a saturated form, i.e., the sum of the inflammation intensity test value and a saturation constant is used as the denominator, and the test value as the numerator, the logarithm is taken, and the same form at the reference value is subtracted; the saturation constant is initially determined by the median of the inflammation intensity test values ​​in the cohort based on goodness of fit. The clearance correction term addresses the fact that the test value increases with decreasing clearance capacity at the same thrombin production level: fibrin degradation products are cleared by the kidneys and mononuclear phagocyte system; when clearance is impaired, these products accumulate in the plasma, while thrombin production does not increase, so this portion is included in the expected coagulation activation level. The inhibition reserve correction term addresses the fact that thrombin production increases after the consumption of inhibitors such as antithrombin, so the logarithm of the ratio of the reference value to the measured value is taken, making this term positive when the test value decreases. In summary, the expected coagulation activation amount increases with the increase of inflammation intensity represented by the inflammation intensity test quantity, increases with the decrease of clearance capacity represented by the clearance capacity test quantity, and increases with the decrease of the reserve represented by the inhibitory reserve test quantity. , , The only function of these three is to treat each logarithmic ratio as zero in the reference state; their changes are... The intercept absorption of the occurrence rate over a period of time can be arbitrarily selected from the commonly used range in this field, which is the rule for determining the three.

[0075] The activation and inhibition parameters are calculated according to the action classification rules. These rules state that any activation that can be explained by a listed source via the coagulation activation pathway is included in the expected coagulation activation value and deducted from the measured value. Direct effects from the same source on thrombus nucleation and growth are separately included in the time period incidence rate. The foreign body interface of catheter placement, endothelial stretching caused by ventilation, and the inhibitory effect of anticoagulants all act on the measured value along the coagulation activation pathway, and are therefore all included in this formula, with the anticoagulant term taking a negative sign. The expected coagulation activation value increases with the cumulative dose of therapeutic exposure that increases coagulation activation via the coagulation activation pathway, and decreases with the cumulative dose of therapeutic exposure that inhibits coagulation activation via the coagulation activation pathway.

[0076] The logarithm of the expected coagulation activation dose is taken as the right-hand side of the above formula and the lower limit of quantitation. The larger of the two logarithms. To determine the lower limit of quantitation (LOQ) for the coagulation activation factor (CEFF) detection method, this embodiment uses 0.22 mg / L FEU. Detection results below the LOQ are not quantitatively significant, and their logarithms cannot be used for subtraction; therefore, the logarithm of the LOQ is used as the lower bound. During the period when this limit is in effect, a flag indicating that the expected CEFF has exceeded the LOQ is set. This flag is set without further processing and is recorded by the log module. This limitation ensures that the expected CEFF value falls within the quantifiable range of the detection method, thus ensuring that subsequent subtractions are always performed between two quantitatively significant values.

[0077] The inflammation intensity test, clearance capacity test, and inhibitory reserve test serve as the driving force and two correction factors, respectively, making the same measured coagulation activation level comparable among patients with different inflammation levels, clearance capacity, and inhibitory reserve. The determination no longer depends on the patient's current organ function level. Specifically, the clearance correction factor deducts the increase in the measured value due to decreased clearance capacity according to the mechanism, and this source is no longer included in the unexpected activation residual. The inhibitory reserve correction factor includes the increased thrombin generation due to inhibitor consumption into the expected value, and the increase in activation under the same driving force is not considered unexpected.

[0078] Step S102 uses the current state test value to obtain the coagulation activation value that the patient should have in the current state, so that the minuend and subtrahend in the subsequent subtraction are under the same patient, the same time period, and the same state. Therefore, the judgment value is a deviation relative to the individual's current state rather than a deviation relative to the population. The physical gain, cumulative exposure dose, and the three types of current state test values ​​are included in this formula, so that the three explanatory sources of genetics, treatment, and pathology are deducted before entering the residual.

[0079] In step S103, the measured coagulation activation amount is first truncated to upper and lower limits: Less than the lower limit of quantitation Time to take Greater than the reportable limit Time to take If the value falls between the two or is exactly equal to one of them, it takes the original value. As the reportable upper limit of this detection method, 20 mg / L FEU is used in this embodiment. When both censoring points take effect, a threshold indicating whether the measured coagulation activation level exceeds the upper limit or the lower limit is set respectively.

[0080] After truncation, the activation deviation Pick natural logarithm minus The difference obtained from the natural logarithm is dimensionless. Subtraction is performed in the logarithmic domain: the sources of measured coagulation activation are multiplicative on the concentration scale. Genetic gain, inflammation-driven, clearance correction, inhibitory reserve correction, and exposure activation each amplify or reduce the concentration by a factor. After taking the logarithm, these factors become additive, and the right side of formula (2) can thus be written as the sum of the terms. The difference obtained from subtraction is... This is the logarithmic increment of the portion outside of all sources. This quantity and the deviation scale of the next step are both logarithmic, and the result of dividing the two is a dimensionless quantity.

[0081] Step S103 changes the construction of the judgment quantity from taking the measured value itself to taking the difference between the measured value and its conditional expectation, so that the same measured coagulation activation quantity gives different deviations in two patients with different expected values; the subtraction is performed in the logarithmic domain, making each multiplicative source additable, providing a dividend on the same scale for the subsequent step of synthesizing the divisor on the logarithmic scale.

[0082] In step S104, the deviation scale is synthesized from the four-part variance, as shown in formula (3):

[0083] Formula (3);

[0084] The measurement propagation term is given in formula (4):

[0085] Formula (4);

[0086] In the formula, For the first The deviation scale for each time period is dimensionless and is the standard deviation on a logarithmic scale. For the measurement propagation term, that is, the detection variation of each detection term is propagated to the activation deviation according to the coefficients of formula (2); To detect variations in activation levels, For detecting mutations in inflammation, To eliminate detected variants, To suppress the variation of reserve detection, the four are, in order, the logarithmic scale coefficients of variation of the measured coagulation activation amount, inflammation intensity test amount, clearance capacity test amount and the inhibition reserve test amount, and the values ​​are shown in step S101; , , and Same as that indicated in formula (2); This refers to the biological fluctuations within a patient, specifically the portion of the activation bias in the fitted sample at different time points that cannot be explained by the detected variations within the same patient. This refers to inter-patient dispersion, i.e., individual differences that cannot be explained by the expected amount of coagulation activation. The independent variable vector consists of the constant term 1, the minor allele dose at each point, the three log ratios, the undetected indicator and the three cumulative exposure doses, which are the independent variables substituted into the terms of formula (2). The coefficient covariance matrix is ​​the covariance matrix estimated by each coefficient in formula (2), and robustly estimated according to patient clustering; the superscript T indicates transpose. This is a quadratic form, and its value represents the variance of the expected coagulation activation amount during this period due to the uncertainty of the coefficient estimate. The four additive terms are all variances on a logarithmic scale, with consistent dimensions; their sum, taken as the square root, is then multiplied by the variance. Same scale.

[0087] In formula (4), To substitute for the discrete value, i.e., the standard deviation of the logarithm of the suppression reserve test value in the calibration cohort. This is when the suppression reserve test value has been measured. If the value is 0, this item is included based on its detected variation and correction index; if no measurement is performed... If the value is 1, this item is changed to be included according to the dispersion of the substituted value, that is, the dispersion of the substituted value itself is used instead of the detection variation. The substituted dispersion is greater than the detection variation, so the bias scale of the unmeasured patients is widened according to the dispersion of the substituted value, and the interpretation is not strict because an unmeasured value is substituted. The square of each coefficient is multiplied by the square of the corresponding coefficient of variation because each logarithmic ratio in formula (2) is entered in a linear manner, and its coefficient is the sensitivity of error propagation.

[0088] From formula (3), we can see that, Not less than , and The square root of the sum of the squares of the three, where A constant greater than 0 and Since the values ​​are not less than 0, and all three are independent of the time period, the divisor has a lower bound greater than zero, and the division will not be amplified by the divisor being close to zero. As another implementation method, the deviation scale can also be the sum of the standard deviations of the above four parts, or a weighted sum of the maximum value among the four and the sum of the remaining terms, as long as the quantity increases with the increase of the uncertainty of each part; the formula for the square root of the sum of squares is based on the premise that the parts are uncorrelated, and the other synthesis methods correspond to the conservative approach when the parts are correlated.

[0089] Step S104 changes the divisor from a fixed constant to a composite quantity that varies over time, ensuring that the four sources—detection variability, coefficient estimation uncertainty, inter-patient dispersion, and intra-patient biological fluctuations—each contribute to the activation bias according to their actual contribution. The divisor includes a constant term greater than 0 that does not change over time, providing a definite lower bound for division. The four parts are added together by their variances and the square root is taken, so that the parts are synthesized on the same scale. The resulting divisor and activation bias are both on a logarithmic scale.

[0090] In step S105, the unexpected activation residual Take activation deviation Divide by the deviation scale The resulting quotient is dimensionless. The numerator of this quotient is the logarithmic increment of activation that cannot be explained by the listed sources, and the denominator is the uncertainty of this increment itself. Therefore, its null distribution is on a unit scale: in the absence of local thrombosis, The expected value is 0 and the standard deviation is 1, so the judgment threshold can be set directly in this unit. One unit is one unit of its own uncertainty.

[0091] When upper and lower limit truncation takes effect, it will be handled as follows: For time periods that are truncated due to exceeding the reportable upper limit, Take / press The calculated value is the lower bound of the true residual, therefore the grade for this period should not be lower than grade 1; periods that are truncated because they are below the quantitative lower limit, Take / press The calculated value, since the expected coagulation activation amount has been limited to the same lower limit, is at this point... If the value is not positive, it will not cause the gear to be shifted up, so only a flag is set and no further action is taken. The time period other than the two truncation points is taken according to the value given by the formula.

[0092] Step S105 divides the quantities on the two logarithmic scales to obtain a dimensionless decision quantity with zero distribution on the unit scale, so that the threshold can be set by multiples instead of by concentration; the same decision quantity still uses its own uncertainty as the unit after changes in detection method, reference value and population baseline level, so a fixed correspondence is obtained between the threshold and the false alarm frequency.

[0093] In step S106, the aforementioned steps are repeated according to a preset time period to obtain the unexpected activation residual for each time period; then, the duration level of the time period is obtained by weighting the unexpected activation residual of each time period and the time periods before that time period; the time period occurrence rate of the time period is obtained from the duration level; and the occurrence probability is synthesized from the time period occurrence rates of each time period within the preset horizon and output.

[0094] The residuals, after deducting all listed sources, still contain a patient-level constant component, namely the individual variability not explained by the expected coagulation activation; this component is biased against the individual baseline. The characterization serves as a benchmark for three aspects: missing measurement handling, forward look, and over-limit counting. The arithmetic mean and contraction coefficient of the unexpected activation residuals for each time period within the anchoring window of the patient with measured coagulation activation results are taken. The accumulated amount, anchored at the window after check-in Each period, This is used to set the length of the anchor window. The selection rule is as follows: take the maximum number of time periods that ensure the anchoring window does not overlap with the formation period of hospital-acquired thrombosis, which is the 1% quantile of the distribution of the earliest detection time of venous thromboembolism in the calibration cohort divided by 1 / 2. And round down, in this embodiment Take 2. Time period number is less than hour, Calculations are performed according to the above rules for each time period up to the current time period, based on the results already available. Take the number of valid anchoring periods up to this time period; if there are no measured coagulation activation results before this time period. Set the value to 0 and mark it as unanchored. The shrinkage coefficient is given in formula (5):

[0095] Formula (5);

[0096] In the formula, The contraction coefficient is dimensionless and takes a value that is not less than 0 and less than 1. The number of effective time periods to be anchored, i.e. the number of time periods within the anchoring window that contain measured coagulation activation results; The discrete share, i.e., the proportion of the variance in the unexpected activation residuals that is discrete among patients, is taken as... The square of the term is divided by the sum of the square of the mean of the measured propagation term on the fitted sample, the square of the biological fluctuation within the patient, and the square of the discreteness between patients, and the value of the entire cohort is taken, without changing from time to time. For the remaining shares, take 1 minus The fewer the effective anchoring time periods, the better. The smaller the value, the higher the estimated value. As the contraction approaches zero, patients with sparse results within the window will not be assigned an excessively large baseline due to accidental values ​​in a few time periods.

[0097] There are two criteria for abandoning anchoring: either there are no measured coagulation activation results within the anchoring window, or the final value of the residual within the window has increased by one unit compared to the initial value. The former cannot be estimated; the latter indicates that an increase in activation has already occurred within the window, and continuing anchoring would treat this process as baseline absorption. Either criterion must be met. If 0 is set and an unanchored flag is set, and neither is true, the value is determined according to the above rules. If 0 is set as the value after abandoning anchoring, the patient's assessment reverts to a calculation method without individual baseline, and the assessment can still be performed without interruption.

[0098] For the period when the measured coagulation activation amount was missing, the unexpected activation residual is shown in formula (6):

[0099] Formula (6);

[0100] In the formula, The lag is the number of time intervals between this time interval and the most recent time interval with a result; The residual is the one obtained from the period with the most recent result. The autocorrelation coefficient is the residual after patient-centeredness and lag. The measured autocorrelation on the baseline is dimensionless, and its value is determined by calibration. This formula causes the value during the missing period to contract from the most recent measured value towards the patient's own baseline, rather than contracting towards 0; the rate of contraction is determined by the measured autocorrelation, without assuming any decay mechanism. Using the individual baseline bias as the contraction target, it ensures that patients with already high baselines are not considered to have fallen back during the missing period, and patients with low baselines are not considered to have risen.

[0101] The persistence level is the amount obtained by weighting the unexpected activation residuals in each time period according to autocorrelation, denoted as See formula (7):

[0102] Formula (7);

[0103] In the formula, The window length is expressed in units of time periods. The lag within the window; Same as formula (6), it also serves as the weight in this formula; For the first The unexpected activation residuals for each time period are taken as the value of formula (6) when the measurement is missing for that period. The denominator is the sum of the weights, therefore and At the same scale, it remains a dimensionless quantity. The rule for determining the window length is: take a value that makes... The first instance of a hysteresis count below 0.10 means the window only covers hysteresis that is not yet negligible. Take 6; Insufficient number of hospitalization periods Patients were calculated using a weighted approach based on existing time periods.

[0104] As another implementation method, the weights can also be chosen in other ways that do not increase monotonically with lag, such as equal weights for each time period, or decaying by a fixed proportion for each time period; in this case, the window length is determined according to the weights decreasing to a negligible level, while the rest of the formula remains unchanged. Using measured autocorrelation as the weight, the residuals at each lag participate in the weighting according to their actual correlation with the current period's residuals, and the weights are not given by a presumed decay form; in applications where the correlation decreases slowly with lag, the window automatically lengthens, and in applications where the correlation decreases quickly, the window automatically shortens.

[0105] After calculation, the range is limited by the interval: it falls within the lower bound of the continuous level. With the upper bound of the persistence level When the value is within the specified range, the original value is used. When the value is exactly equal to one of the boundaries, the original value is also used. When the value is below the lower boundary, the lower boundary is used. When the value is above the upper boundary, the upper boundary is used. When the value exceeds the boundary, an out-of-range flag is set. This flag is only recorded and does not trigger further action. The interval is taken as the 0.1% and 99.9% quantiles of the quantity in the calibration queue over all risk periods, that is, the range of values ​​covered by the estimated coefficients of the period occurrence rate. The two logarithmic ratios in the period occurrence rate are limited according to the same rule. Limiting ensures that each coefficient is not linearly extrapolated outside its estimated range.

[0106] The number of consecutive periods exceeding the limit is denoted as Take up to the number For each time period, the residual relative to the individual baseline bias continuously satisfies Greater than the threshold The number of time periods; exactly equal to The time period is not included, and the count is reset to zero for the missing period. The standard deviation of the zero distribution of the residuals, constructed according to step S105, is 1, so it is set to 1 unit. The individual baseline bias count is subtracted from the residuals because the criterion is to judge the extent of deviation from the individual's normality. Patients with a consistently high baseline have a deviation of zero. When counting by the residuals themselves, patients with an individual baseline bias greater than 1 will meet the criterion in the long term.

[0107] The unexpected activation residuals are repeatedly obtained by repeating the process over a preset time period. This allows the residuals to enter subsequent steps at a sustained level rather than as a single point value. The impact of a jump caused by a single anomaly detection is weakened by the values ​​of the remaining time periods within the window. The sustained level is a weighted average of the residuals within the window, ensuring that a sustained period of high values ​​differs from an isolated high value in that quantity. The occurrence rate of a time period is obtained from the sustained level, and then the occurrence rate of each time period within the field of view is used to synthesize the probability of occurrence. This ensures that the output quantity is the cumulative probability over a period of time rather than the value at a single moment in the current period. Therefore, the time margin for processing is directly given by the length of the field of view.

[0108] The occurrence rate of the time period is the conditional probability of venous thromboembolism occurring within a given time period, denoted as . The linear predictions are given by connecting the complementary repeated logarithms. The linear predictions are given in formula (8):

[0109] Formula (8);

[0110] The incidence rate during a given time period is shown in formula (9):

[0111] Formula (9);

[0112] In the formula, These are linear predictions, dimensionless. The incidence intercept is the logarithm of the product of the time period length and the incidence rate per unit time under the reference condition. For the continuous level coefficient, This is the body mass gain coefficient. The direct coefficient of inflammation. To remove direct coefficients, , and The direct coefficients are, in order, tube placement direct coefficient, ventilation direct coefficient, and anticoagulation direct coefficient. Each coefficient is dimensionless and obtained through calibration. As given by formula (7), , , , , The cumulative dose of the three exposures is the same as that referred to in formula (2); Dimensionless, with values ​​greater than 0 and less than 1. There are two reasons for choosing complementary log-joins: First, the periodic incidence rate is the probability of an event occurring at least once within a period, and the relationship between this join and the incidence rate per unit time is precisely what this join gives. Therefore, the coefficients can still be compared after the period length changes, and the actual parameter of the exponent is dimensionless. Second, this join makes the effect of each item on the incidence rate multiplicative; the absolute risk increment from the same exposure changes synchronously with the patient's persistence level and the other items, rather than being a constant increment.

[0113] The physical gain, inflammation intensity test, and clearance capacity test appear for the second time in formula (8). The roles they play in the two appearances are different: in formula (2), the three are used to subtract the components of the measured coagulation activation that can be explained by the listed sources; in this formula, the three play a role in nucleation and growth, namely, hypercoagulable state, endothelial activation and leukocyte adhesion caused by inflammation, and hemodynamic and platelet function changes accompanied by decreased clearance capacity. The roles in the two places are given by independent coefficients, and the same independent variable therefore has one coefficient in each formula. The three exposures in this formula also play a role that does not pass through the coagulation activation channel: the mechanical foreign body interface of the central venous catheter and local blood flow disturbance, venous return obstruction caused by increased intrathoracic pressure during mechanical ventilation and muscle pump failure caused by immobilization, and the direct inhibition of nucleation by prophylactic anticoagulation.

[0114] The same source is included by one coefficient in each of the two formulas, so that the effect of the coagulation activation pathway and the effect of the other pathway can be estimated separately; the deduction makes the two pathways distinguishable in data, and the coefficient of the direct pathway thus takes a sign and magnitude consistent with the mechanism. With two independent coefficients, the coefficients of the two formulas can also be re-estimated separately: the dependent variable of each coefficient in formula (2) is the measured coagulation activation amount, which can be directly measured in any new application; the dependent variable of each coefficient in formula (8) is the outcome, which can only be re-estimated after the outcome is obtained.

[0115] This method excludes both death and transfer out of the intensive care unit, collectively termed competing events: death is a medically significant competing event, while transfer is an administrative censoring; although the patient may still contract the disease after transfer, they are no longer monitored by this method. The competing event time period rate is denoted as... The linear prediction value is given in the same form as formulas (8) and (9): the linear prediction value is obtained by taking an intercept, adding a coefficient to each of the logarithmic ratio of inflammation intensity and the logarithmic ratio of clearance capacity, and then connecting them by complementary multiple logarithms. The intercept and two coefficients mentioned above are obtained from calibration. After subtracting both, the quantity output by this method is the probability of venous thromboembolism being observed within the intensive care unit and within the preset field of view. All thresholds of this method are set according to this definition; this definition is consistent with the treatment scenario, because each treatment level is an action within the intensive care unit.

[0116] The number of time periods within the preset field of view is denoted as In this embodiment, 4 is used, corresponding to 48 hours. The values ​​of each quantity in each forward-looking time period within the field of view are taken in the following manner. Forward sight numbering, from 0 to . When the foreseeable residual is 0, the unexpected activation residual for that period is taken directly, without consulting a table; From 1 to At that time, the foresight residuals are taken from the individual baseline bias and conditional mean foresight table in lag. The current deviation is the sum of the table values ​​at the stratum to which it belongs. The current deviation refers to the difference between the current residual and the individual baseline bias. The conditional mean forward table is obtained by statistically analyzing the event-free periods of the calibration cohort: the current deviation is divided into seven strata according to its 20%, 40%, 60%, 80%, 95%, and 99% quantiles in that batch of periods, and the subsequent strata are statistically analyzed layer by layer and lag by lag. The average deviation over a given time period; when the number of samples for a certain lag in a certain stratum is not less than 200, the table value is taken as the measured average of that stratum; when it is less than 200, the average is taken as... The product of the product with the boundary of the layer, except for the highest layer which takes its lower boundary, and the other layers which take its upper boundary. The forward-looking persistence level is calculated by substituting the forward-looking residuals and the residuals of each observed period into formula (7), and then limiting the amplitude according to the same interval. The forward-looking values ​​of the two logarithmic ratios are taken as the least mean square linear prediction of the independent variable with its current value. The coefficient is called the forward-looking linear prediction coefficient. This coefficient and the intercept are obtained by lag in the calibration queue. The linear regression is given. The cumulative dose of forward exposure evolves deterministically according to formula (1), and the exposure in-situ indication takes the continuation of the current state as the boundary condition; the scheduled extubation or weaning time can be directly substituted into its future value, and the output at this time is the probability of following the plan.

[0117] The conditional mean of the forward-looking residual is taken instead of the product of the autocorrelation coefficient and the current deviation. The reason is that the autocorrelation coefficient is measured on the sequence after deducting the mean of each patient's own residual, while the deviation caused by forward-looking is the individual baseline bias after shrinkage of the residual. The shrinkage coefficient is less than 1, and a part of the patient-level constant component is still retained in this deviation. Its lag correlation is higher than that of the autocorrelation coefficient. Using the autocorrelation coefficient as the forward-looking coefficient will systematically underestimate the persistence. The amount of underestimation is greater at the two ends of the deviation than in the middle. The persistence level is a linear functional of the residual sequence. Therefore, its conditional expectation is equal to substituting the conditional expectation of the residuals in each time period into the same functional. In this scheme, the forward-looking value of the residual is given first, and then the persistence level is recalculated according to formula (7). The result is the conditional expectation itself, without any functional form assumption.

[0118] The original cumulative probability is synthesized using the discrete-time cumulative occurrence function, as shown in formula (10):

[0119] Formula (10);

[0120] In the formula, The original cumulative probability is the uncalibrated estimate of the cumulative probability given by the definition described above, and it is dimensionless. This represents the number of time periods within the field of view. Forward view number, The product number is used for the consecutive product numbers. The incidence rate during the forward vision period, i.e., the incidence rate during the forward vision period. The occurrence rate of each time period is obtained by substituting the aforementioned forward-looking quantities into formulas (8) and (9); and For the same incidence rate during the forward visual period, its second subscript follows the product number. Values; The competition event period rate is determined by taking the current period value within the field of view; When the result is 0, the product within the square brackets is empty and takes the value 1; if any factor in the product is less than 0, that factor takes the value 0. The result within the square brackets represents the product of the first product. The probability that neither type of event has occurred in the preceding time period (i.e., neither venous thromboembolism nor competing events) is multiplied by the occurrence rate of that time period to obtain the probability that the event occurs exactly within that time period. The sum of the probabilities for each time period is the cumulative probability within the horizon. The discrete-time cumulative occurrence function and competing risk reduction are well-known techniques and will not be elaborated upon here. As other implementation methods, implementers can use other methods from existing technologies, such as a reduction formula based on an exponent of cumulative risk. This embodiment does not impose any special restrictions on this.

[0121] The probability of occurrence The original cumulative probability is taken after calibration mapping transformation. This mapping is monotonically non-decreasing and is determined by a set of anchor points. The anchor points are determined as follows: the original cumulative probability is divided into seven segments according to the 20%, 40%, 60%, 80%, 95%, and 99% quantiles of the calibration queue during the at-risk period (i.e., the same group of quantiles used for the stratification of the forward look-ahead table). One anchor point is taken for each segment, with its x-axis representing the mean of the original cumulative probability within that segment and its y-axis representing the mean of the original cumulative probability during the at-risk period within that segment. The number of time periods in which events occur according to the aforementioned definition is divided by the number of at-risk time periods within that time period. Events within that time period with fewer than 5 time periods are weighted and merged with adjacent time periods. If the ordinates of two adjacent time periods violate monotonically, they are weighted and merged using the pooling adjacent violator method until monotonically achievable. The pooling adjacent violator method is a well-known technique and will not be elaborated upon here. As other implementation methods, implementers can use other methods from existing technologies, such as other methods for order-preserving regression. This embodiment does not impose any special restrictions on this. The mapping linearly interpolates the logarithmic abscissa and logarithmic ordinate between adjacent anchor points. When the abscissa is lower than the first anchor point, it is extrapolated according to the slope of the first segment; when it is higher than the last anchor point, the ordinate of the last anchor point is taken. Therefore, the upper bound of the output of this method is the ordinate of the last anchor point. The rationale for this mapping is: the original cumulative probability is... The incidence rate for each time period is derived by reducing the competitive risk and compounding the risk. The compounding amplifies the deviation at each step, making it more conservative in the segment with the highest value. This mapping is monotonically non-decreasing, so it does not reverse the order of any two time periods, but only changes the probability scale.

[0122] As another implementation method, the probability of occurrence can be directly derived from the unexpected activation residual of a single time period through a monotonically increasing mapping, without repeating the calculation over time periods. In this case, the preset horizon is the length of one time period, while the rest of the formula remains unchanged. The probability of occurrence is obtained from the unexpected activation residual and output, so that a dimensionless deviation is converted into a probability scale within the preset horizon that can be compared with the clinically tolerable false alarm frequency. The time periodization and calibration mapping work in sequence to ensure that the output covers the entire horizon and falls on the probability scale measured in the calibration queue.

[0123] After obtaining the probability of occurrence, it is compared with a preset threshold, and the patient's venous thromboembolism monitoring level is output according to the comparison result. This embodiment has three levels: Level 0 is the routine level, where mechanical and pharmacological prophylaxis are performed according to standard ward procedures; Level 1 is the monitoring level, where lower limb circumference, skin temperature, and clinical signs are assessed each shift, and the implementation of mechanical and pharmacological prophylaxis measures is verified; Level 2 is the confirmation level, which includes lower limb compression ultrasound examination. Entry conditions are: Not lower than the gear 2 threshold and No less than the minimum number of overdue periods At that time, the original gear was gear 2; the combination conditions were not met. Not lower than the gear 1 threshold When the above conditions are not met, the original gear is gear 1; when neither of the above conditions is met, the original gear is gear 0. The gear obtained from the above comparison is the original gear. The venous thromboembolism monitoring gear output by this method is obtained from the original gear according to the gear adjustment rules below. The criteria for the two gears are different: gear 1 is based on the probability level, while gear 2 is based on the combined evidence that the probability reaches its threshold and the residual continuously exceeds the limit. The selection rule is: take the smallest integer that makes a single abnormal detection not trigger the confirmation file on its own. In this embodiment, it is 2.

[0124] The gear threshold The minimum value is set such that the frequency at which this setting is triggered by patients who have not experienced venous thromboembolism does not exceed the preset upper limit. The tier number is 1 or 2. The trigger frequency is defined as the number of time periods during which a patient without events meets the entry criteria for that tier, divided by the number of patient weeks equivalent to their observation duration at risk; the upper limit for false alarm frequency is... The number of false alarms per patient per week is determined by the user. Numerical inverse solution is obtained on the calibration queue. This embodiment takes... Two time periods per patient per week The threshold is 0.5 time period / patient / week. The two thresholds are solved inversely, therefore... It can be lower than A patient can directly enter level 2 without going through level 1. When users set tolerance based on the number of alarms rather than the number of time periods, the tolerance is first calculated by the ratio of the number of times the current queue has been upgraded to that level to the number of time periods in that level.

[0125] The threshold level is derived by inverse solving the tolerable false alarm frequency, so that the threshold is no longer a probability value given by experience, but is directly determined by the number of false alarm periods per week that the user can tolerate. The same rule can be inversely solved again after changes in the field of view length or the basic risk of the population, so that the false alarm frequency returns to the set value. The minimum value that satisfies the upper limit of this frequency is taken to maximize the number of time periods identified under this frequency constraint.

[0126] Upshifting and downshifting are performed as follows: Upshifts are performed immediately upon meeting the entry conditions. Downshifts occur only one gear at a time, and the entry conditions for the current gear must not be met for two consecutive time periods, with a relaxed probability threshold and hysteresis bandwidth. That is, if the entry condition for category 2 is not met for two consecutive time periods, the probability threshold will be changed. reduce For the subsequent combination conditions, level 1 must not be satisfied for two consecutive time periods with a probability of occurrence of not less than [a certain value]. reduce This condition; if any one of these conditions is met, the gear will not be downgraded. After downgrading, the count is reset to zero, and two consecutive time periods must be counted again before downgrading to the next gear; the refractory period count and the number of consecutive time periods exceeding the limit are not affected by downgrading. The exit condition for each gear is consistent with the composition of its entry condition. Gear 2 contains both probability and persistence, while gear 1 contains only probability; downgrading requires two consecutive time periods in addition to the above criteria. Calculate the standard deviation of the persistence level change during the event-free period for each category. The median of the change in the probability of occurrence caused by this The standard deviation of the difference in persistence levels between two adjacent time periods: shift the residual series as a whole. The remaining independent variables are taken as the actual values ​​of the time period. The probability of occurrence is recalculated and its increment is taken. The median is taken in the no-event time period where the probability of occurrence falls within the neighborhood of the threshold. The neighborhood is a double-sided interval centered on the threshold and expanded at the same ratio above and below. Its relative width is the minimum value that makes the number of time periods in the interval not less than 500. If the value obtained according to the above rules is greater than the threshold of that gear, then half of the threshold of that gear shall be taken.

[0127] After triggering and performing imaging examinations at level 2, a positive result confirms the diagnosis, and the patient is transferred to anticoagulation therapy and exits the tiering process of this method. If the result is negative, the tier is lowered to level 1 and the patient enters the refractory period, the length of which is specified in the original text. The number of lag periods at which the autocorrelation coefficient first falls below 0.10 is taken; in this embodiment, 6 lag periods are used. The system will not enter level 2 during the refractory period; the refractory period is immediately lifted when the sustained level rises by 1 from the imaging time, and thereafter, the system is judged according to the conventional entry conditions. If any data acquisition is missing for two consecutive lag periods, the system will be at least at level 1 and marked as having insufficient data.

[0128] The classification determination of this method is based solely on the data collection quantity and its derivatives listed in step S101, without taking any information other than the above quantities; its output is the venous thromboembolism monitoring level and the triggering reason of that level, for medical staff to verify, without directly executing any drug administration, drug discontinuation or other treatment actions, and the clinical treatment corresponding to the level is determined and implemented by medical staff according to the ward procedures.

[0129] Each parameter is calibrated offline on the calibration queue. After calibration, the parameters are fixed and will not be updated with patients during deployment. The hyperparameter values ​​and convergence criteria are as follows. Step 1: Define the fitting sample: Select samples from the calibration queue that are at least a certain number of time intervals from the detection time of venous thromboembolism. Furthermore, for all at-risk periods for which actual coagulation activation levels were measured, patients who did not experience events met the criteria for all at-risk periods for which actual coagulation activation levels were measured. The upper bound of the time required for a local thrombus to go from nucleation to detection is taken as the initial value. The standard time from detection to clinical diagnosis of deep vein thrombosis in this exposed population is used as the initial value. In this embodiment, the initial value is taken as 6 time periods. After completing steps two to four, the number of consecutive time periods exceeding the limit for the event patients is taken as the first time not less than The 90th quantile of the time interval distribution from the time of event detection to the time of event detection is used as the new If the difference from the previous round is no more than one time period, the calculation will stop and converge. Otherwise, the calculation will return to the current step and be repeated. The number of recalculations will not exceed three rounds. If the convergence criterion is not met after three rounds, the calculation will stop. The number of sample time periods used for the third round of calculation and the parameters obtained from the second to fourth steps will be retained and the subsequent calibration will continue. The second step estimates the coefficients and site effect quantities of formula (2): The logarithm of the measured coagulation activation quantity is used as the dependent variable. The calibration queue is divided in half according to the body gain section. The least squares estimation is performed by taking the site set selected in the other half for each half. The design matrix takes the constant term, the dose series of each point and the independent variables of the other terms on the right side of formula (2). The covariance is robustly estimated according to the patient clustering. After the two halves are interchanged, the average is taken according to the body gain section. Least squares estimation is a well-known technique and will not be described in detail here. As other implementation methods, implementers may use other methods of the prior art, such as weighted least squares, ridge regression, etc. This embodiment does not impose any special restrictions on this. When the estimated value is not greater than 0 or not less than 1, the saturated form of the inflammation driving term is used and this step is repeated. The third step is to perform variance decomposition: decompose the activation bias of the patient to each time period of the fitted sample. The variance within the patient is the weighted average of the variance within each patient with the number of time periods minus one. The square of the dispersion between patients is the product of the variance of the patient mean minus the average of the variance within the patient and the reciprocal of the number of time periods of each patient. The square of the biological fluctuation within the patient is the difference between the average of the variance within the patient and the square of the measurement propagation term. When both are negative, they are both taken as 0. The fourth step is to obtain the measured autocorrelation sequence: center the fitted sample by the patient first, and then calculate the autocorrelation coefficient for lags of 1 to 12 time periods. The window length, the lag for missing detection, the refractory period length and the reduction coefficient of the forward table are determined by this sequence. When none of the above lags are satisfied within 1 to 12, they are all taken as 12. In formula (6), when the lag is greater than 12, the autocorrelation coefficient is taken as 0. The unexpected activation residual of this time period is taken as the individual baseline bias. Step 5: Determine the amplitude limit interval: According to the rules described in step S106, take the upper and lower bounds of the duration level and the two log ratios for all periods in the calibration queue that are at risk. Step 6: Estimate the coefficients of formula (8) and the coefficients of the competing events: Construct discrete time period survival data with time period as the observation unit, the dependent variable is whether the event occurs in the time period, perform complementary relog regression, solve with Newton-Raphson iteration, the convergence criterion is that the difference in log likelihood between two adjacent iterations is less than the convergence limit, in this embodiment the convergence limit is 0.000001, when the number of iterations reaches 200 and still has not converged, take the parameters of the previous round and set the calibration non-convergence flag; among them, Newton-Raphson iteration is a well-known technology, and will not be described in detail here. As other implementation methods, implementers can use other methods of existing technology, such as iterative reweighted least squares, quasi-Newton method, etc., this embodiment does not impose special restrictions on this. Step 7: Calibrate the forget kernel time constant: Search according to the rules described in step S102. If the time constant changes, return to step 2 for a recalculation round. The number of recalculation rounds should not exceed three. If the time constant still needs to be updated after three rounds, stop the recalculation, retain the time constant used in the third recalculation, and retain the parameters obtained from steps 2 to 6, and continue with subsequent calibration. Step 8: Construct the conditional mean lookahead table and calibration mapping: According to the rules described in step S106, statistically analyze the layer boundaries of the lookahead table and the table values ​​of each layer and each lag in the calibration queue. Then, use this table to calculate the original cumulative probability of all at-risk periods and find the anchor point of the calibration mapping. Step 9: Inverse solution threshold and hysteresis bandwidth: Numerical inverse solution is performed on the calibration queue according to the rules described in step S106.

[0130] If any of the following four events occur: change of testing method or testing platform, change of ward preventive medication plan, change of site set, or change of sampling period length, recalibration must be performed according to steps one through nine above.

[0131] The following substitutions do not change the relationship between the above steps. When the suppression reserve test quantity is taken as plasma fibrinogen concentration or platelet count, the sign direction of this item in formula (2) must be re-determined according to the relationship between the substitution quantity and thrombin generation, while the rest of the formulas remain unchanged. The time period can be 6 hours or 24 hours. At this time, the forgetting kernel time constant, autocorrelation sequence, window length and incidence intercept must be recalibrated, while complementary logarithmic connections make the coefficients outside the incidence intercept still roughly comparable. The number of time periods of the horizon can be 2 or 8, and the threshold must be re-solved under this horizon according to its selection rules. In addition to the persistence level, a differential statistic such as the window slope can be added to the persistence level. At this time, anti-abnormal measures must be configured at the same time, because the weight of the differential operator on a single abnormal detection value is higher than that of the weighted average. Without anti-abnormal measures, a single abnormal detection will be amplified into a treatment. The time-period incidence rates given by formulas (8) and (9) can be replaced by a discrete time-period proportional hazards model or gradient boosting tree, as long as the inputs remain the duration level, physical gain, logarithmic ratio of the two test values, and cumulative dose of the three exposures. The inverse solution rules for the threshold remain unchanged after the replacement, while the amplitude limit tree model for the duration level can be omitted because the tree model does not perform linear extrapolation. The treatment exposure record can be supplemented with the remaining items as listed in step S101. For each new item, its forgetting kernel time constant must be determined simultaneously, and its component through the coagulation activation channel is classified into the expected coagulation activation amount, while the remaining components are classified into the time-period incidence rate, according to the aforementioned action classification rules.

[0132] This method identifies the additional increase in activation contributed by the formed thrombus. This contribution increases with the surface area accessible to plasmin; therefore, this method is suitable for detecting proximal or larger thrombi. For small, isolated distal thrombi, where the activation increase falls within the deviation scale, this method is not recommended for detection. This applicability aligns with the lower extremity compression ultrasound procedure used in confirmatory procedures, which is itself insensitive to small distal thrombi.

[0133] When this method is applied to postoperative patients without sepsis, inflammation is no longer the dominant driver of coagulation activation. The inflammation-driven term in formula (2) loses its driving force and must be reconstructed by the surgical trauma dose. At the same time, the baseline of measured coagulation activation in this population is not saturated, and the components distinguished by the residuals change accordingly. Therefore, when this method is applied to this population, the inflammation-driven term must be reconstructed as a whole, and the remaining steps are performed as described in this application.

[0134] The information and data collected in this application (including but not limited to the test results of D-dimer, C-reactive protein, serum creatinine and antithrombin activity, records of the three treatment exposures listed in this embodiment, and minor allele doses of coagulation pathway functional sites obtained from peripheral blood genotyping) are information and data obtained with the written informed consent of the subject or their legal representative, or with full authorization from all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with relevant laws, regulations, and standards, and necessary confidentiality measures have been taken. They do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entry points for users to choose to agree to or refuse the automated decision results; if the user chooses to refuse, the process proceeds to the expert decision-making process.

[0135] Genotyping is limited to the coagulation pathway functional sites listed in step S102, without whole-genome sequencing or inference of ancestry or population affiliation; written informed consent from the subject or their legal representative must be obtained before data collection. Genotyping results are linked to clinical data with random numbers, and these random numbers are stored separately from identity information. The original genotyping data is not provided to external parties along with the genotyping results.

[0136] See Figure 3This embodiment also provides a prediction system for the probability of venous thromboembolism in sepsis patients, deployed on a hospital intranet. The system's modules correspond one-to-one with the aforementioned steps, including ten modules: data access module, physical fitness gain module, exposure dose module, expected activation module, residual module, baseline and persistence module, probability module, decision module, calibration module, and log module. The inputs and outputs of each module are as follows: The data access module retrieves the test results from the laboratory information system, the start and end times of each treatment exposure from the medical orders and nursing records and ventilator records, and the site dose from the genotyping database. It aligns the data by time period and executes the missing test handling in step S101, outputting the quantities and missing test indicators listed in step S101; The physical gain module performs the synthesis of physical gain in step S102 and outputs the physical gain; The exposure dose module executes formula (1), outputs the cumulative dose of the three exposures and saves the recursive state; The expected activation module executes formula (2), outputs the expected coagulation activation amount and its lower limit indicator; The residual module executes steps S103 to S105, outputting the unexpected activation residual, the deviation scale and two truncation indicators; The baseline and persistence module executes... Formulas (5) to (7) and the limit count are used to output the individual baseline bias, duration level, number of consecutive limit-exceeding periods, unanchored flag, insufficient data flag, and flag indicating that the calibration range is exceeded. The probability module executes formulas (8) to (10) and calibration mapping to output the period occurrence rate, competing event period rate, original cumulative probability, and occurrence probability. The decision module performs grade determination, maintains the grade, downgrade count, refractory period count, and imaging time duration level for each patient, and outputs the patient's grade and triggering reason. The calibration module performs each calibration step offline, outputs and versiones all parameters, and records the calibration queue identifier and calibration time. The log module records the input, intermediate quantity, output, and all flags of each module in each period for retrospective purposes. The records are randomly numbered and do not contain identity information.

[0137] The connection relationship of the above ten modules is as follows: the output of the data access module is connected in series in three paths: one path goes through the constitution gain module, one path goes through the exposure dose module, and one path goes directly to the expected activation module. The outputs of the first two paths are connected in parallel with the third path at the expected activation module. The measured coagulation activation outputs of the expected activation module and the data access module are sent together to the residual module. Subsequently, the residual, baseline and persistence, probability, and decision modules are connected in series in sequence. The calibration module and the log module are connected in parallel to the main chain. The calibration module sends parameters to each calculation module, and each module reports its inputs, intermediate values, and outputs to the log module. The modules are synchronized in time periods. All calculations for a time period are triggered once after the test results are reported at the end of that time period.

[0138] The prediction system of this embodiment includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned method for predicting the probability of venous thromboembolism in sepsis patients. The memory includes, but is not limited to, random access memory, read-only memory, flash memory, and disk storage. The processor includes, but is not limited to, a central processing unit, a graphics processing unit, and an application-specific integrated circuit. Implementing the aforementioned method with a memory and a processor does not limit the implementation form; the same method can be implemented on a centralized server or distributed across multiple devices.

[0139] This application also provides a computer-readable storage medium storing a computer program that, when run by a processor, implements the aforementioned method for predicting the probability of venous thromboembolism in sepsis patients. This application also provides a computer program product including a computer program that, when run by a processor, implements the aforementioned method. This application also provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, which, when run by the processor, implements the aforementioned method.

[0140] The following is an experimental example of this method on a set of synthesis queues. The synthesis queues are generated by an independently defined mechanistic model at 1-hour intervals, and the equations and constants at the generation end are as follows.

[0141] Genotypes were extracted from allele frequencies at each locus according to Hardy-Weinberg equilibrium. The frequencies at the ten loci were 0.31, 0.27, 0.42, 0.09, 0.04, 0.03, 0.55, 0.12, 0.18, and 0.44, respectively. The logarithmic effect size per allele at the generating end was 0.225, 0.150, 0.128, 0.104, 0.250, 0.275, 0.068, 0.058, 0.046, and 0.026, respectively. An additional log-normal individual term was added, with a log-normal standard deviation of 0.22. The sum of the products of each locus dose and the above logarithmic effect size, and the sum of the individual term, was denoted as the individual logarithmic gain.

[0142] The trajectory of the inflammation intensity test was the sum of 8 mg / L and a decay term, which was the exponent of the peak value multiplied by the negative time divided by the time constant. The peak value was drawn from a log-normal distribution with a median of 120 mg / L and a log-standard deviation of 0.62. The time constant was drawn from a log-normal distribution with a median of 70 hours and a log-normal deviation of 0.42. For 26% of patients, a time point was drawn uniformly between 72 and 240 hours after admission, and a second increase was superimposed from that time point, with a median amplitude of 90 mg / L and a time constant of 40 hours. The baseline of the clearance capacity test was drawn from a median of 70 μmol / L. For 42% of patients, the peak value was drawn from a log-normal distribution with a peak value of 210 μmol / L and a log-standard deviation of 0.45, and the value decreased over a median time constant of 110 hours. For the remaining patients, the peak value was drawn from a median of 78 μmol / L and a log-standard deviation of 0.24. The inhibition reserve test value is calculated by subtracting 100% from the difference between 100% and the individual minimum value, multiplying by the severity, and then dividing by 0.75. The severity is calculated by dividing the inflammation intensity test value by the sum of the inflammatory intensity test value and 60 mg / L. The individual minimum value is extracted at 62% of the mean and 14% of the standard deviation and truncated at 28% to 100%.

[0143] 86% of patients had their central venous catheters inserted at uniformly distributed sampling times between 0 and 36 hours after admission, with the indwelling duration following a log-normal distribution of a median of 150 hours and a log-standard deviation of 0.55; 52% of patients started mechanical ventilation at uniformly distributed sampling times between 0 and 24 hours, with the duration following a median of 96 hours and a log-standard deviation of 0.6; and 68% of patients started prophylactic anticoagulation between 6 and 48 hours, with the initiation time determined by a uniformly distributed independent sampling method, independent of the patient's coagulation status, exposure status, and thrombotic status, and continued medication thereafter.

[0144] Catheter-related intimal injury (CADI) is calculated starting from 0 and progressing hourly. The hourly increment is 0.030 multiplied by the central venous catheter placement indicator, then multiplied by 1 and the difference from this increment, minus the increment, and divided by 96 hours. Immobilization-related blood flow obstruction is also calculated starting from 0 and progressing hourly. The hourly increment is 0.034 multiplied by the greater of the mechanical ventilation placement indicator and 0.55 times the central venous catheter placement indicator, then multiplied by 1 and the difference from this increment, minus the increment, and divided by 60 hours. The hypercoagulable component is calculated as the individual logarithmic gain exponent, multiplied by the severity, divided by 0.45, multiplied by the ratio of 90% to the inhibitory reserve test, and then multiplied by a factor of 0.58 when prophylactic anticoagulation is used and 1 when not. Thrombosis nucleation rate is calculated as... The volume of the thrombus after nucleation is multiplied by the sum of the immobilization-related blood flow obstruction and 0.18, the sum of the catheter-related intimal injury and 0.12, the hypercoagulable component, and then by a factor of 0.55 when prophylactic anticoagulation is used and 1 when not. The volume of the thrombus after nucleation increases hourly from 0, with the hourly increment being 0.055 multiplied by the hypercoagulable component, multiplied by 1, and the difference between this volume and the thrombus volume, minus 0.020 multiplied by the thrombus volume; an event is recorded when the volume exceeds 0.42.

[0145] The measured coagulation activation dose was accumulated hourly starting from 0.05 mg / L FEU. The hourly increase was calculated as follows: 0.125 multiplied by the hypercoagulable component, plus 0.085 multiplied by the catheter-related intimal injury, plus 0.030 multiplied by the immobilization-related blood flow obstruction, plus 0.280 multiplied by the thrombus volume, and then subtracted by the clearance rate constant multiplied by this amount. The clearance rate constant was calculated as the sum of the difference between 0.115 per hour divided by 1 and 0.55 times the following, and this difference was calculated as the clearance capacity test dose divided by 80 μmol / L, minus 1 and... The larger of the two; the lower limit of this amount is taken as 0.05 mg / L FEU. The hourly mortality rate is taken as... Multiply by the inflammatory intensity test dose divided by 100 mg / L to the power of 0.8, then multiply by the clearance capacity test dose divided by 100 μmol / L to the power of 0.6; the hourly rate of excretion is taken as... Multiply by 60 mg / L and divide by the smaller of the inflammatory intensity test result and 3.0, starting 48 hours after admission.

[0146] Each test quantity was analyzed using the coefficient of variation listed in step S101, plus log-normal measurement noise. The inhibition reserve test quantity was never measured in 34% of patients; the measured activation coagulation level was randomly omitted in 6% of time periods. The calibration cohort consisted of 2000 patients, with random seed 20260904 and measurement random seed 20260905; the validation cohort consisted of 1400 patients, with random seed 20261004 and measurement random seed 20261005. Both cohorts were followed up for a maximum of 28 time periods. The incidence of venous thromboembolism was 10.6% in the calibration cohort and 10.1% in the validation cohort. The fitted sample consisted of 42,771 time periods, with the incidence rate of each period calibrated across 46,883 at-risk time periods and 212 events. A total of 481 patients died or were transferred out of the calibration cohort; 34 of these deaths occurred after the detection of venous thromboembolism and were no longer at risk, while the remaining case occurred at the end of the follow-up period. Therefore, competing events were calibrated across 446 events in the same batch of time periods. The functional forms of the three steps—physical gain, inter-patient dispersion, and inhibitory reserve—are identical at the generation end and the method end. Therefore, the following values ​​only record the implementation results of the method's formulas at these three steps and are not used as evidence for the effects of genetic gain on coagulation activation, the existence of individual differences in a constant bias form, or the increase in activation under unit-driven inhibitor consumption.

[0147] The following is marked on the synthesis queue: It is 0.1768. It is 0.5501. It is 0.6650. It is 0.3435. It is 0.0171. It is 0.7718. It is 0.1394. The value is 0.3953; the value substituted when the inhibition reserve test quantity is not measured is 85.53%. The value was 0.1167 for the period during which the suppression reserve test quantity was measured. It is 0.1412. The effect size was 0.1918. The effect sizes of the ten sites were 0.1654, 0.1131, 0.0874, 0.0703, 0.1865, 0.1671, 0.0466, 0.0416, 0.0308, and 0.0247, respectively, and their robust standard errors for patient clustering were 0.0071, 0.0069, 0.0067, 0.0104, 0.0165, 0.0196, 0.0065, 0.0100, 0.0087, and 0.0064, respectively. for , It is 1.0435. It is 1.1220. It is 1.0324. It is 0.3418. It is 4.3836. It is 0.8952. for The sustained horizontal range is... The logarithmic ratio of inflammation intensity was 3.0937. Up to 2.1545, the logarithmic ratio of clearance capacity is Up to 1.8453. The intercept for the competition event is... The coefficients of its two logarithmic ratios are respectively: and The forward predictive coefficients of the log-ratio of inflammation intensity at lags of 1 to 4 were 0.9389, 0.8865, 0.8353, and 0.7864, respectively, with intercepts of [missing values]. , , , The coefficients of the logarithmic ratios of the scavenging capacity were 0.8842, 0.8343, 0.7874, and 0.7426, respectively, with intercepts of 0.0175, 0.0055, and 0.0055, respectively. , The seven boundary levels of the conditional mean forward table are as follows: , The values ​​are 0.2571, 0.7293, 1.4508, and 2.1268, with the number of samples at each lag level ranging from 374 to 7794. The six anchor points of the calibration mapping have the following x-coordinates: 0.001037, 0.006158, 0.014668, 0.037438, 0.103802, and 0.288844, and the following y-coordinates: 0.001706, 0.007145, 0.016745, 0.038965, 0.105011, and 0.162047.

[0148] One example was taken from the 16th time period of patient number 124 in the validation cohort, which ended 192 hours after admission. The site doses were 2 for the first site, 1 for the third site, 1 for the seventh site, 2 for the ninth site, and 2 for the tenth site, with the rest being 0; the measured coagulation activation dose was 2.315 mg / L FEU, the inflammation intensity test dose was 7.87 mg / L, the clearance capacity test dose was 74.05 μmol / L, and the inhibition reserve test dose was 86.78%. The central venous catheter was in place, the patient was on mechanical ventilation, and prophylactic anticoagulation had been used; the cumulative doses of the three exposures were 0.8647, 0.9592, and 0.9994, respectively. The fitness gain was 0.5758. The terms on the right side of formula (2) were 0.1768, 0.5758, and 0.5758, respectively. , 0.0125, 0.6674, 0.1337 The unmeasured bias item was not included because the inhibition reserve test was already measured in this case and was therefore zero; the total was 0.1026, with an expected coagulation activation level of 1.108 mg / L FEU. The bias scale was 0.2663, the activation bias was 0.7368, and the unexpected activation residual was 2.767. The individual baseline bias was 0.2252, the persistence level was 1.621, and the number of consecutive out-of-limit periods was 2. The linear predictive value was... The incidence rate during this period was 0.0208. The original cumulative probability was 0.1129, the occurrence probability was 0.1088, and the level was 2. The actual time of occurrence for this patient was 196 hours after admission.

[0149] Another example was taken from the 7th time period of patient number 141 in the same cohort, which ended 84 hours after admission. The site doses were 1 for the second, third and seventh sites, 2 for the tenth site, and 0 for the rest, with a fitness gain of 0.2965. The results of the four tests in this time period were: measured coagulation activation dose 5.662 mg / L FEU, inflammation intensity 146.70 mg / L, clearance capacity 246.09 μmol / L, and inhibitory reserve 77.40%; all three treatment exposures were in place, with cumulative doses of 0.5276, 0.7534 and 0.9179 respectively. The terms added to the right side of formula (2) were 0.1768, 0.2965, 0.5921, 0.7472, 0.0518, 0.4072, 0.1050, Unmeasured bias items were all 0; the total was 2.0138, with an expected coagulation activation level of 7.492 mg / L FEU; the bias scale for this period was 0.2658, and the activation bias was... The unexpected activation residual was Individual baseline bias is The sustained level is The number of consecutive time periods exceeding the limit is 0. The linear prediction value is... The incidence rate during this period was 0.0044. The original cumulative probability was 0.0191, the incidence rate was 0.0212, and the level was 0. This patient did not experience venous thromboembolism during hospitalization.

[0150] The mean of the unexpected activation residuals in the fitted sample of 42,771 event-free periods was 0.0000 and the standard deviation was 0.9853. When the detection variation of the measured coagulation activation amount was 0.10 as a fixed divisor, the standard deviation of this amount in the same batch of periods was 2.616.

[0151] During the event-free period of the validation queue, the linear correlations of the unexpected activation residuals with each listed channel are as follows: the correlation coefficient for cumulative catheter dose is 0.014, and for cumulative anticoagulation dose is... The logarithmic ratio for inflammation intensity was 0.046, and the logarithmic ratio for clearance capacity was... The gain on physical fitness is 0.031. When the exposure activation term is not included in formula (2), the correlation coefficient for cumulative dose after catheterization is 0.642; when the anticoagulation inhibition term is not included, the cumulative anticoagulation dose is... Without a clearing correction term, the logarithmic ratio of clearing ability is 0.681; without a constitution gain term, the constitution gain is 0.540.

[0152] Under the search results obtained according to the rules for determining the time constant of the forgotten nucleus, the mean changes in the unexpected activation residuals before and after the initiation or withdrawal of various treatment exposures are as follows: Catheter initiation 0.188, sample 769; Ventilation initiation 0.182, sample 339; Catheter withdrawal 0.411, sample 975; Ventilation withdrawal 0.102, sample 673; Anticoagulation initiation Sample size 750. When the anticoagulation inhibition term is not included in formula (2), the change before and after anticoagulation initiation is: , Sample size 750.

[0153] According to the autocorrelation sequence after patient-centeredness, the lags from 1 to 6 are 0.4371, 0.3546, 0.2673, 0.1924, 0.1203, and 0.0605, respectively.

[0154] The physical gain coefficient in the incidence rate over time is 1.1220 when the physical gain has been deducted in formula (2), with a standard error of 0.3843; and is [value missing] when the physical gain has not been deducted in formula (2). Its standard error is 0.4895.

[0155] The area under the receiver operating characteristic (ROC) curve, which measures the probability of occurrence over the next 48 hours, was 0.843 for the validation cohort (32,916 at-risk periods and 536 positive periods, i.e., periods where the event occurred within the next four periods). After removing the persistence level term from the period incidence rate and recalibrating, the ROC was 0.814 for the same cohort of periods. The mean predicted probability for all periods was 0.017318, and the mean measured incidence was 0.016284.

[0156] The upper limits of the false alarm frequency for the two levels are set at 2 and 0.5 time periods / patient / week, respectively. The results are obtained by inverse solving over 44,839 event-free time periods and 3,202.8 patient weeks in the calibrated cohort. It is 0.0298. The value was 0.0053; the actual trigger frequencies measured in the validation cohort of 31,544 event-free periods and 2,253.1 patient weeks were 2.055 and 0.474 periods / patient / week, respectively.

[0157] The comparison with the item-based scoring was taken from the same validation cohort time period: acute infection 1 point, bed rest 3 points, respiratory failure 1 point, and known thrombotic tendency 3 points. Acute infection and bed rest were counted as true for all time periods. Respiratory failure was counted as mechanical ventilation. Known thrombotic tendency was counted as a fitness gain not lower than the 90th percentile in the calibration cohort. Only four different scores were recorded, and all time periods had scores not lower than 4 points, with a mean of 4.475 and a standard deviation of 0.961. The area under ranking for the same time period based on this score was 0.600. The comparison with directly interpreting measured concentrations was taken from the event-free time period of the validation cohort: the standard deviation of the logarithmic value of measured coagulation activation was 0.782, with a sample size of 29,665. The ratio of this to the median of the deviation scale (0.2656) for the same time period was 2.944.

[0158] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions or improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting the probability of venous thromboembolism in patients with sepsis, characterized in that, The method is executed by a computer, and the method includes: Obtain the measured coagulation activation volume and current status test results of sepsis patients; The expected coagulation activation amount is obtained from the current state test quantity; The activation deviation is obtained by subtracting the logarithm of the expected coagulation activation amount from the logarithm of the measured coagulation activation amount. The deviation scale is obtained from the detection variation of the measured coagulation activation quantity and the current state test quantity, the estimation uncertainty of each coefficient used to obtain the expected coagulation activation quantity, the inter-patient dispersion, and the intra-patient biological fluctuation. The deviation scale increases with the increase of any part of the uncertainty corresponding to the detection variation, the estimation uncertainty, the inter-patient dispersion, and the intra-patient biological fluctuation. Dividing the activation bias by the bias scale yields the unexpected activation residual; The probability of occurrence is obtained from the unexpected activation residual and output. The probability of occurrence is the probability that the sepsis patient will develop venous thromboembolism within a preset field of vision.

2. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 1, characterized in that, The method further includes: obtaining treatment exposure records of the sepsis patient, and obtaining the cumulative exposure dose of each treatment exposure from the treatment exposure records; the expected coagulation activation dose increases with the increase of the cumulative exposure dose of treatment exposure that increases coagulation activation through the coagulation activation channel, and decreases with the increase of the cumulative exposure dose of treatment exposure that inhibits coagulation activation through the coagulation activation channel.

3. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 2, characterized in that, The cumulative exposure dose increases with the continuation of the treatment exposure and decreases with the withdrawal of the treatment exposure. The rate of increase and decrease is taken to minimize the sum of the absolute values ​​of the mean changes in the unexpected activation residuals before and after the treatment exposure is initiated and before and after its withdrawal.

4. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 1, characterized in that, The method of obtaining the expected coagulation activation amount from the current state test quantity includes: the current state test quantity includes an inflammation intensity test quantity, a clearance capacity test quantity, and an inhibitory reserve test quantity; the clearance capacity test quantity characterizes the ability of the measured coagulation activation amount to be cleared from the plasma; the inhibitory reserve test quantity characterizes the amount of thrombin generation inhibitors; the expected coagulation activation amount increases with the increase of inflammation intensity as represented by the inflammation intensity test quantity, increases with the decrease of clearance capacity as represented by the clearance capacity test quantity, and increases with the decrease of the amount of inhibitory reserve test quantity.

5. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 1, characterized in that, The step of obtaining and outputting the occurrence probability from the unexpected activation residual includes: repeatedly obtaining the unexpected activation residual over a preset time period; weighting the unexpected activation residual of each time period and the time periods preceding that time period to obtain the duration level of that time period; obtaining the time period occurrence rate of that time period from the duration level of that time period; and synthesizing and outputting the occurrence probability from the time period occurrence rates of each time period within the preset horizon.

6. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 5, characterized in that, The occurrence rate of the time period is obtained by taking at least one of the current state test values; the effect of the same term taken in two places on the expected coagulation activation amount and its effect on the occurrence rate of the time period are given by independent coefficients.

7. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 5, characterized in that, The weight of each unexpected activation residual in the weighting of the duration level is taken as the autocorrelation of the unexpected activation residual over the interval between its time period and the time period.

8. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 1, characterized in that, The method further includes: obtaining the minor allele doses of the sepsis patient at each functional locus that is effective in the measured coagulation activation level; obtaining a fitness gain from the minor allele dose and the effect of the corresponding functional locus on the measured coagulation activation level; the fitness gain increases as the product of the minor allele dose and the corresponding effect increases; and the expected coagulation activation level increases as the fitness gain increases.

9. The method for predicting the probability of venous thromboembolism in sepsis patients according to claim 1, characterized in that, The method further includes: comparing the occurrence probability with a preset level threshold, and outputting the venous thromboembolism monitoring level for the sepsis patient according to the comparison result; the level threshold is the minimum value that ensures the frequency of patients who have not experienced venous thromboembolism triggering the venous thromboembolism monitoring level does not exceed a preset upper limit.

10. A predictive system for the probability of venous thromboembolism in patients with sepsis, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method for predicting the probability of venous thromboembolism in sepsis patients as described in any one of claims 1 to 9.