Physiological index active confirmation sampling strategy generation method and device, and electronic device

Abstract

This invention provides a method, apparatus, and electronic device for generating an active confirmation sampling strategy for physiological indicators, relating to the field of computer data processing technology. The method includes: acquiring current and historical detection data of at least one physiological indicator of a target object; determining a processable indicator from the at least one physiological indicator; generating a monitoring state probability distribution of the target object based on the current and historical detection data of the processable indicator; performing an uncertainty assessment on the target object to obtain an uncertainty vector; generating a set of candidate sampling actions for the target object based on the monitoring state probability distribution and the uncertainty vector; predicting the action benefit of executing each candidate sampling action based on the monitoring state probability distribution and the uncertainty vector; selecting a target candidate sampling action from multiple candidate sampling actions based on the action benefit of each candidate sampling action, and outputting a confirmation sampling strategy for the corresponding target candidate sampling action.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, specifically to a method and apparatus for generating an active confirmation sampling strategy for physiological indicators, and an electronic device. Background Technology

[0002] Continuous monitoring data is commonly found in home testing, wearable recording, remote follow-up records, and self-testing records. This type of data typically includes the measured value, collection time, sample type, data quality, and life events. Some indicators have recognized thresholds, reference ranges, or grading boundaries. Examples include urine albumin-to-creatinine ratio, urine total protein-to-creatinine ratio, urine pH, blood pressure, fasting blood glucose, random blood glucose, continuous blood glucose statistics, glycated hemoglobin, and body mass index.

[0003] Current processing methods often employ fixed thresholds and fixed retest frequencies. This approach struggles to distinguish between transient disturbances, boundary fluctuations, insufficient baselines, and continuous changes. A single detection exceeding the threshold may also stem from sample conditions, recent activity, lack of sleep, fever, measurement errors, or missing crucial context. Directly increasing the detection frequency would increase the burden on users; directly updating the long-term baseline could allow transiently disturbed samples to alter subsequent judgment criteria. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device, and electronic device for generating active confirmation sampling strategies for physiological indicators. Based on the probability distribution and uncertainty of the current monitoring state of the target object, the invention can select target candidate sampling actions with higher action benefits, thereby providing a more suitable confirmation sampling strategy for the target object and offering more reasonable retesting suggestions for the target object.

[0005] The method for generating active confirmation sampling strategies for physiological indicators provided by this invention includes: Obtain indicator-related data for at least one physiological indicator of the target object, wherein the indicator-related data includes: current detection data and historical detection data; A manageable indicator is determined from the at least one of the physiological indicators; Based on the current and historical detection data of the processable indicators, a monitoring status probability distribution of the target object is generated, wherein the monitoring status probability distribution indicates the conditional probability of multiple preset monitoring states. An uncertainty assessment is performed on the target object to obtain its uncertainty vector. Based on the monitoring state probability distribution and the uncertainty vector, a candidate sampling action set for the target object is generated, the candidate sampling action set including: multiple candidate sampling actions; Based on the monitoring state probability distribution and the uncertainty vector, predict the action benefit of performing each of the candidate sampling actions; Based on the action benefits of each of the candidate sampling actions, a target candidate sampling action is selected from the plurality of candidate sampling actions, and a confirmation sampling strategy corresponding to the target candidate sampling action is output. The confirmation sampling strategy includes at least two of the following: sampling time, sample type, number of samplings, and verification type.

[0006] The present invention also provides an active confirmation sampling strategy generation device, the device comprising: The data acquisition unit is used to acquire indicator-related data of the physiological indicators of the target object, including: current detection data and historical detection data; A manageable indicator determination unit is used to determine a manageable indicator from at least one physiological indicator; A state probability calculation unit is used to generate a monitoring state probability distribution of the target object corresponding to the physiological indicators based on the current detection data and the historical detection data. The monitoring state probability distribution indicates the conditional probability of multiple preset monitoring states. An uncertainty calculation unit is used to perform uncertainty assessment on the physiological indicators and obtain the uncertainty vector of the physiological indicators; A candidate action generation unit is used to generate a set of candidate sampling actions for the physiological indicator based on the monitoring state probability distribution and the uncertainty vector. The set of candidate sampling actions includes multiple candidate sampling actions. An action benefit calculation unit is used to predict the action benefit of performing each of the candidate sampling actions based on the monitoring state probability distribution and the uncertainty vector. The strategy output unit is used to select a target candidate sampling action from the plurality of candidate sampling actions based on the action benefits of each candidate sampling action, and output a confirmation sampling strategy corresponding to the target candidate sampling action. The confirmation sampling strategy includes at least two of the following: sampling time, sample type, number of samplings, and verification type.

[0007] The present invention also provides an electronic device and a computer-readable storage medium. The electronic device and the computer-readable storage medium are used to perform the above-described method.

[0008] The technical effects of this invention are limited to the data processing level. This method can reduce the occurrence of transient perturbation samples entering the long-term baseline, select retest actions with higher state entropy reduction benefits with a limited number of confirmation samplings, and retain confirmation markers when the detected value is close to the reference boundary. This method outputs monitoring process suggestions and data processing markers, but does not output disease diagnosis conclusions or provide treatment suggestions.

[0009] In one embodiment, for each preset monitoring state in the preset monitoring state set S, the conditional probability of each preset monitoring state is calculated; the formula for calculating the conditional probability of the preset monitoring state is: b_t(s)=exp(r_t(s)) / Σ_{v∈S}exp(r_t(v)); r_t(s) = ω_s^TF_t + ξ_s; Wherein, b_t(s) represents the conditional probability that the target object is in the preset monitoring state s at time t. If the number of processable indicators is 1, then F_t includes the target indicator features of the processable indicator at time t; if the number of processable indicators is multiple, then F_t includes the aggregated features formed by the target indicator features of all the processable indicators at time t. The target indicator features of the processable indicators include at least two of the following: the standardized value, recent change, boundary distance, uncertainty component, and data quality score of the processable indicator; ω_s represents the weight vector of the preset monitoring state s, ξ_s represents the bias term of the preset monitoring state s, and v represents any preset monitoring state in the preset monitoring state set S.

[0010] In one embodiment, based on the monitoring state probability distribution and the uncertainty vector, predicting the action benefit of performing each of the candidate sampling actions includes: For each candidate sampling action, the action benefit of performing the candidate sampling action is calculated based on at least three of the following: uncertainty change, false alarm cost, false negative cost, acknowledgment delay cost, sampling burden cost, and resource consumption cost.

[0011] In one embodiment, the change in uncertainty of performing the candidate sampling action is represented by the change between the current state entropy and the expected state entropy after performing the candidate sampling action; the state entropy is used to measure the dispersion of the current monitoring state probability distribution, and the current state entropy is calculated as follows: E_t=-Σ_s b_t(s)log(b_t(s)+ε); Where E_t represents the state entropy at time t, s represents the preset monitoring state, b_t(s) represents the conditional probability of being in the preset monitoring state s at time t, log represents the natural logarithm function, and ε represents a small positive number used to avoid taking the logarithm of zero.

[0012] In one embodiment, the calculation formula for the action benefit of performing the candidate sampling action is: U(a)=λ_I•ΔE(a)-λ_F•C_false(a)-λ_M•C_miss(a)-λ_D•C_delay(a)-λ_B•C_burden(a)-λ_R•C_resource(a); Where U(a) represents the action benefit of performing candidate sampling action a, ΔE(a) represents the expected reduction in state entropy after performing candidate sampling action a, C_false(a) represents the false alarm cost, C_miss(a) represents the missed alarm cost, C_delay(a) represents the acknowledgment delay cost, C_burden(a) represents the sampling burden cost, C_resource(a) represents the resource consumption cost, λ_I represents the weight corresponding to the state entropy reduction benefit, λ_F represents the weight corresponding to the false alarm cost, λ_M represents the weight corresponding to the missed alarm cost, λ_D represents the weight corresponding to the acknowledgment delay cost, λ_B represents the weight corresponding to the sampling burden cost, and λ_R represents the weight corresponding to the resource consumption cost.

[0013] In one embodiment, the expected change in state entropy ΔE(a) after performing candidate sampling action a is calculated as follows: ΔE(a)=E_t-Σ_oP(o|a,b_t,u_t)E_{t+1}(o,a); Where o represents the possible future observation result after performing candidate sampling action a, and the future observation result includes at least one of the following: future detection value, future standardized value, sample type or measurement condition, quality label, missing label, context completion result and external reference detection result; P(o|a,b_t,u_t) represents the probability of obtaining future observation result o under the conditions of the monitoring state probability distribution b_t and the uncertainty source vector u_t, and E_{t+1}(o,a) represents the state entropy at the next moment after obtaining future observation result o by performing candidate sampling action a.

[0014] In one embodiment, selecting a target candidate sampling action from the plurality of candidate sampling actions based on the action benefit of each candidate sampling action includes: From all the candidate sampling actions, select the candidate sampling action that satisfies the preset selection rules as the target candidate sampling action; the preset selection rules include at least one of the following: highest action benefit rule, parallel action priority rule, quality and safety rule, user authorization status rule, external reference detection availability rule, or coverage rule specified in the indicator configuration table.

[0015] In one embodiment, the indicator-related data further includes: indicator configuration data; the indicator configuration data includes: the indicator type, unit, reference boundary, transformation function, and quality requirements of the physiological indicator; The manageable indicator is determined from the at least one physiological indicator, including: For each of the physiological indicators, if the indicator configuration data of the physiological indicator indicates that the physiological indicator has a recognized threshold, reference range, grading boundary or positive / negative boundary, then the physiological indicator is determined to be a processable indicator. If the indicator configuration data of the physiological indicator does not indicate that the physiological indicator has a recognized threshold, reference range, grading boundary, or positive / negative boundary, then the physiological indicator is determined to be a non-processable indicator, and the physiological indicator is used as an auxiliary input in the context record data or the uncertainty source vector.

[0016] In one embodiment, if the number of processable indicators is one, the uncertainty source vector includes at least two of the following: uncertainty of sampling conditions, uncertainty of transient events, uncertainty of reference boundary distance, uncertainty of individual baseline reliability, uncertainty of measurement error, and uncertainty of missing data; wherein, the uncertainty of sampling conditions is determined by sample type, sampling time, or measurement condition field; the uncertainty of transient events is determined by event occurrence identifier, event severity, event time interval, and event record reliability; the uncertainty of reference boundary distance is determined by the distance between the standardized value and the reference boundary; the uncertainty of individual baseline reliability is determined by the number of valid baseline samples and the baseline interval width; the uncertainty of measurement error is determined by equipment error, repeated measurement coefficient of variation, or calibration status; and the uncertainty of missing data is determined by the proportion of missing necessary fields.

[0017] In one embodiment, if there are multiple processable indicators, the uncertainty source vector is obtained as follows: First, obtain the uncertainty source components of each of the processable indicators; According to the trigger priority, maximum value, weighted average, or rules corresponding to each of the processable indicators, the uncertainty source components of all the processable indicators are aggregated to obtain the uncertainty source vector.

[0018] In one embodiment, the physiological indicators include at least one of the following: urine albumin-to-creatinine ratio, urine total protein-to-creatinine ratio, urine pH, urine specific gravity, urine glucose, urine ketones, urine occult blood, urine red blood cell count, blood pressure, fasting blood glucose, random blood glucose, continuous blood glucose statistics, glycated hemoglobin, body mass index, and resting heart rate.

[0019] In one embodiment, the current detection data and the historical detection data include at least three of the following: detection value, detection time, sample type, data source, quality marker, and missing marker.

[0020] In one embodiment, the indicator-related data further includes: context recording data, which includes at least two of the following: exercise records, water intake records, sleep records, stress records, fever records, medication or supplement change records, device logs, and calibration status.

[0021] In one embodiment, the candidate sampling actions in the candidate sampling action set are selected from the candidate sampling action library, and the candidate sampling actions in the candidate sampling action library include at least three of the following: maintaining regular detection, retesting within a preset time window, retesting according to preferred sample conditions or measurement conditions, retesting after a brief event decay, paired sampling retesting, continuous sampling retesting, retesting after completing the key context, and external reference detection confirmation. Specifically, if the main source of uncertainty is the uncertainty of sampling conditions, then retesting under preferred sample conditions or measurement conditions should be prioritized; if the main source of uncertainty is the uncertainty of transient events, then retesting after the transient events have decayed should be prioritized; if the main source of uncertainty is the uncertainty of missing data, then retesting after completing the key context should be prioritized; if the monitoring status is boundary pending confirmation or continuous deviation pending confirmation, then paired sampling retesting, continuous sampling retesting, or external reference detection confirmation should be prioritized.

[0022] In one embodiment, the preset monitoring state includes at least three of the following: insufficient baseline, stable baseline, stable baseline but large fluctuations, short-term disturbances, boundary to be confirmed, continuous deviation to be confirmed, continuous deviation confirmed, and recent decline or recovery; wherein, the baseline may be an individual baseline of a single processable indicator, or a comprehensive baseline state formed by the individual baseline states of multiple processable indicators through a preset aggregation rule.

[0023] In one embodiment, the baseline update admission flag for the current processable index is determined based on at least one of the following: the uncertainty of transient events u_event,t, the uncertainty of measurement errors u_measure,t, the uncertainty of missing data u_missing,t, and the probability of the state to be confirmed P_pending,t; wherein, P_pending,t can be obtained by summing the probabilities of each state in the set of states to be confirmed S_pending; the active confirmation sampling strategy includes: the baseline update admission flag.

[0024] In one embodiment, the method further includes: generating a standardized individual trigger boundary; the active confirmation sampling strategy includes: the standardized individual trigger boundary; the calculation formula for the standardized individual trigger boundary is: trigger^Y_{i,j,l,t}=RB^Y_{j,l}+δ^Y_{i,j,t}+p^Y_{i,j,l,t}; δ^Y_{i,j,t}=w_{i,j,t}(mean(Y_{i,j,B})-μ^Y_{group,j}); w_{i,j,t}=n_eff / (n_eff+k_δ); p^Y_{i,j,l,t}=d_{j,l}(n_1u_boundary,t+n_2u_measure,t+n_3u_missing,t+n_4u_baseline,t); Where, trigger^Y_{i,j,l,t} represents the standardized confirmation sampling trigger boundary used at time t for the j-th physiological index of target object i at the l-th reference boundary, RB^Y_{j,l} represents the recognized reference boundary after transformation function and standardization, δ^Y_{i,j,t} represents the individual offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, p^Y_{i,j,l,t} represents the protective offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, mean represents the mean value, and Y_{i,j,B} represents the j-th physiological index of target object i. The standardized values ​​of the physiological indicators entering the baseline candidate pool are: μ^Y_{group,j} represents the standardized center value of the j-th physiological indicator of target object i in the population; d_{j,l} represents the boundary direction of the j-th physiological indicator; d_{j,l} takes +1 at high boundaries and -1 at low boundaries; the d_{j,l} value of the j-th physiological indicator without a directional boundary is specified by the indicator configuration table; η_1, η_2, η_3, and η_4 represent preset coefficients; the standardized individual trigger boundary can be used together with the target candidate sampling action to generate or adjust the sampling time, sample type or measurement conditions, number of samplings, and verification type in the confirmation sampling strategy.

[0025] In one embodiment, after the number of historical log samples reaches a preset number, at least one of the probability of the future observation result or the action reward can be estimated by an offline training model; the offline training model includes at least one of a generalized linear model, a gradient boosting tree, a temporal neural network, a contextual gambling machine, or an offline strategy evaluation model; the offline training model only sorts candidate actions within the candidate confirmation sampling action set and preset constraints. Attached Figure Description

[0026] Figure 1 This is a flowchart of the active confirmation sampling strategy generation method for physiological indicators according to the first embodiment of the present invention. Detailed Implementation

[0027] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the purpose, features, and advantages of the present invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative of the essential spirit of the technical solution of the present invention.

[0028] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0029] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0030] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0031] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to include the meaning of “or / and” unless otherwise expressly stated herein.

[0032] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0033] Therefore, a data processing method is needed that, given existing detection results, selects the next confirmation sampling action based on the probability of the monitoring state and the source of uncertainty, and simultaneously outputs a baseline update admission flag. This method does not limit the specific detection hardware structure and does not protect reagent formulations, sensor structures, chip structures, mechanical sampling components, or reader structures.

[0034] The first embodiment of this invention relates to a method for generating an active confirmation sampling strategy for physiological indicators, applied to a processing device. This processing device can be a mobile phone, desktop computer, laptop computer, server, home testing device, POCT (point-of-care testing) device, portable testing device, wearable device, urine testing device, blood testing device, saliva testing device, sweat testing device, or continuous monitoring device. The test values ​​can be obtained directly from the aforementioned devices, or they can be transmitted from the testing device to a mobile terminal or server for execution of this method. The following description focuses on a urine testing device as a key embodiment, but the invention is not limited to urine testing indicators.

[0035] In other words, the physiological indicators include at least one of biological sample detection indicators, body fluid detection indicators, sensor detection indicators, or wearable device detection indicators.

[0036] The physiological indicators include at least one of the following: albumin-to-creatinine ratio (ACR), urine albumin-to-creatinine ratio (UACR), protein-to-creatinine ratio (PCR), urine protein-to-creatinine ratio (UPCR), urine pH, urine specific gravity, urine glucose, urine ketones, urine occult blood, urine red blood cell count, blood pressure, fasting blood glucose, random blood glucose, continuous blood glucose statistics, glycated hemoglobin, body mass index (BMI), and resting heart rate.

[0037] The specific process of the active confirmation sampling strategy generation method for physiological indicators in this embodiment is as follows: Figure 1 As shown.

[0038] Step 101: Obtain indicator-related data for at least one physiological indicator of the target object. The indicator-related data includes current detection data and historical detection data.

[0039] Specifically, the target object can be a user with authorized testing records, which can come from home testing devices, wearable devices, laboratory reports, user-authorized imported data, or other testing data sources. This invention only processes the data records and strategy generation results generated after testing; the specific testing hardware, sample collection structure, and reagent reaction process are not limited to the scope of this invention.

[0040] Obtain data related to one or more physiological indicators of the target object. The data related to the indicators include: current test data and historical test data. The current detection data and the historical detection data include at least three of the following: detection value, detection time, sample type, data source, quality marker, and missing marker.

[0041] For example, the detection data R_{i,j,t} of the j-th physiological indicator of target object i at time t can be represented as: + R_{i,j,t}=(x_{i,j,t},time_t,source_t,type_t,q_{j,t},c_t,miss_t); Where x_{i,j,t} represents the original detection value of the j-th physiological indicator at time t or the detection value after conversion to a standard scale, time_t represents the detection time of the j-th physiological indicator, source_t represents the data source of the j-th physiological indicator at time t, type_t represents the sample type or measurement conditions of the j-th physiological indicator at time t, q_{j,t} represents the detection quality vector of the j-th physiological indicator at time t (where the quality label can be one item in the detection quality vector or generated by the detection quality vector), c_t represents the context record data of the j-th physiological indicator at time t, and miss_t represents the missing label of the j-th physiological indicator at time t.

[0042] Furthermore, indicator-related data may also include: indicator configuration data and context record data.

[0043] The indicator-related data also includes: indicator configuration data; the indicator configuration data includes: the indicator type, unit, reference boundary, transformation function and quality requirements of the physiological indicator.

[0044] For example, the indicator configuration data T_j for the j-th physiological indicator of target object i can be represented as: T_j=(name_j,type_j,unit_j,RB_j,f_j,quality_j); Where T_j represents the configuration of the j-th physiological indicator; name_j represents the name of the j-th physiological indicator; type_j represents the type of the j-th physiological indicator; unit_j represents the unit of the j-th physiological indicator; RB_j represents the recognized reference boundary, reference range, grading boundary, or positive / negative boundary of the j-th physiological indicator; f_j represents the transformation function of the j-th physiological indicator; and quality_j represents the quality requirements of the j-th physiological indicator.

[0045] Recognized reference boundaries can be derived from publicly available guidelines, national standards, industry standards, registered test methods, equipment manuals, or audited specification tables; these boundaries are used for algorithm configuration.

[0046] If a certain indicator does not have a traceable reference boundary, that indicator will not trigger the confirmation sampling strategy on its own. The indicator can still be used as contextual input or auxiliary observation.

[0047] Contextual logging data includes at least two of the following: activity logs, hydration logs, sleep logs, stress logs, fever logs, medication or supplement change logs, device logs, and calibration status. Sleep logs, hydration logs, activity logs, and stress logs are typically used to interpret uncertainties and are not used alone as confirmatory sampling triggers unless they are configured with traceable, recognized reference boundaries.

[0048] Step 102: Determine the manageable indicator from the at least one physiological indicator.

[0049] Specifically, after obtaining the relevant data of at least one physiological indicator of the target user, the first step is to determine whether each physiological indicator can be used to execute the subsequent active confirmation sampling strategy. For each physiological indicator: If the indicator configuration data of the physiological indicator indicates that the physiological indicator has a recognized threshold, reference range, grading boundary, or positive / negative boundary, then the physiological indicator is determined to be a processable indicator and participates in the subsequent process of determining the active confirmation sampling strategy.

[0050] If the indicator configuration data of the physiological indicator does not indicate that the physiological indicator has a recognized threshold, reference range, grading boundary, or positive / negative boundary, then the physiological indicator is determined to be an unprocessable indicator and will not participate in the subsequent steps of determining the active confirmation sampling strategy.

[0051] In addition, for physiological indicators that are not processable, the indicator-related data of these physiological indicators are used as contextual record data or as auxiliary inputs in the uncertainty source vector.

[0052] Based on the above process, all physiological indicators that can participate in the subsequent determination of the active confirmation sampling strategy can be selected and processed.

[0053] Step 103: Based on the current detection data and historical detection data of the processable indicators, generate a monitoring status probability distribution of the target object. The monitoring status probability distribution indicates the conditional probability of multiple preset monitoring states.

[0054] Specifically, the processing device is pre-configured with a set of preset monitoring states, which includes multiple preset monitoring states. For example, the set of preset monitoring states includes at least three of the following eight preset monitoring states: insufficient baseline, stable baseline, stable baseline but large fluctuations, short-term disturbances, boundary to be confirmed, continuous deviation to be confirmed, continuous deviation confirmed, and recent decline or recovery.

[0055] For example, the preset monitoring state set S = {S0, S1, S2, S3, S4, S5, S6, S7}; where S0 indicates insufficient baseline; S1 indicates stable baseline; S2 indicates stable baseline but large fluctuations; S3 indicates short-term disturbances dominating; S4 indicates boundaries to be confirmed; S5 indicates persistent deviation to be confirmed; S6 indicates persistent deviation confirmed; and S7 indicates recent decline or recovery. The persistent deviations mentioned in S5 and S6 include persistent increases in high-boundary indicators, persistent decreases in low-boundary indicators, and persistent abnormal deviations in undirected boundary indicators.

[0056] Among them, the states such as insufficient baseline, stable baseline, and stable baseline but large fluctuation can be determined for the individual baseline of a single manageable indicator; when there are multiple manageable indicators, the baseline state of each manageable indicator can be determined first, and then the comprehensive monitoring state of the target object can be formed according to the rules specified in the indicator configuration table, such as trigger indicator priority, maximum value, weighted average, or indicator configuration table.

[0057] For each preset monitoring state in the preset monitoring state set S, the conditional probability of each preset monitoring state is calculated; in the case of a single processable indicator, the feature calculation is based on the processable indicator; in the case of multiple processable indicators, the feature calculation is based on the features of multiple processable indicators and their aggregated features. For a preset monitoring state s in the set of monitoring states S, the expression for the conditional probability b_t(s) that the target object is in the preset monitoring state s at time t is: b_t(s)=P(S_t=s|Y_{1:J,1:t},C_{1:t},Q_{1:t}); Where P represents the conditional probability, S_t=s represents the target object being in the preset monitoring state s at time t; Y_{1:J,1:t} represents the standardized values ​​from the first to the Jth processable indicators from the initial time to time t; C_{1:t} represents the context record data from the initial time to time t; Q_{1:t} represents the data quality record from the initial time to time t or the data quality score sequence obtained by aggregating the detection quality vectors q_{j,t} at each time; and J is the total number of processable indicators.

[0058] The conditional probability of the preset monitoring state can be estimated by a rule table, a Bayesian state model, a hidden Markov model, a gradient boosting tree, or a temporal neural network. In the small sample stage, rule tables and Bayesian state models are preferred to estimate the conditional probability of the preset monitoring state.

[0059] In one example, the conditional probability of a preset monitoring state can be obtained through state scoring normalization. The expression for the conditional probability b_t(s) of the target object being in the preset monitoring state s at time t is: b_t(s)=exp(r_t(s)) / Σ_{v∈S}exp(r_t(v)); r_t(s) = ω_s^TF_t + ξ_s; r_t(v) = ω_v^TF_t + ξ_v; Where r_t(s) represents the state score of the preset monitoring state s at time t, T represents the transpose, F_t represents the feature vector of the target object at time t, ω_s represents the weight vector of the preset monitoring state s, ξ_s represents the bias term of the preset monitoring state s, v represents the v-th preset monitoring state in the preset monitoring state set S, exp represents the natural exponential function, and Σ represents the summation function.

[0060] Specifically, if the number of processable indicators is one, then F_t includes the target indicator feature of the processable indicator at time t; if the number of processable indicators is multiple, then F_t includes the aggregated feature formed by the target indicator features of all the processable indicators at time t; the target indicator feature of the processable indicator includes at least two of the following: the standardized value, recent change, boundary distance, uncertainty component, and data quality score of the processable indicator; that is, in a single processable indicator scenario, F_t includes at least two of the standardized value, recent change, boundary distance, uncertainty component, and data quality score of the processable indicator, and in a multiple processable indicator scenario, it includes the above features of each processable indicator and the aggregated feature formed according to trigger priority, maximum value, weighted average, or indicator configuration table rules.

[0061] ω_s and ξ_s are preset values ​​that can be estimated from the indicator configuration table, rule scoring table, or authorized historical logs.

[0062] The feature vector F_t of the target object at time t includes: the standardized value of the detection value of each processable indicator at time t, the standardized value of the change in the detection value of each processable indicator in the recent specified time period, the boundary distance between the detection value of each processable indicator at time t and the boundary value, the uncertainty source vector of the target object, and the data quality score.

[0063] When there is only a single tangible metric, F_t can be composed of the standardized value of the tangible metric, recent change, boundary distance, uncertainty source component, and data quality score; when there are multiple tangible metrics, F_t can include the above characteristics of each tangible metric, and further include aggregated characteristics obtained by trigger priority, maximum value, weighted average, or metric configuration table rules.

[0064] The calculation process from the detected values ​​of each processable indicator to the standardized values ​​is as follows: Taking the j-th manageable index as an example, the formula for calculating the equivalent value y_{j,t} of the j-th manageable index at time t is: y_{j,t}=f_j(x_{j,t}) Where y_{j,t} represents the equivalent value of the j-th processable indicator at time t, x_{j,t} represents the detection value of the j-th processable indicator at time t, and f_j represents the transformation function of the j-th processable indicator.

[0065] The transformation function of physiological indicators is determined by the type of physiological indicator, specifically: For continuous ratio-based physiological indicators, such as UACR or UPCR, the conversion function is a log function; for example, the equivalent value of the j-th manageable indicator is y_{j,t}=log(x_{j,t}+ε_j); where log represents the natural logarithm function; ε_j represents a small positive number of the j-th manageable indicator, used to handle zero values, values ​​below the detection limit, or low-resolution readings.

[0066] For continuous or semi-continuous numerical physiological indicators, such as the j-th manageable indicator, y_{j,t}=(x_{j,t}-μ_{ref,j}) / s_{ref,j}; where μ_{ref,j} represents the reference range center value or training sample center value of the j-th manageable indicator, and s_{ref,j} represents the reference range half-width, validation sample standard deviation, or scale given in the configuration table for the j-th manageable indicator.

[0067] For semi-quantitative graded physiological indicators, the conversion function can map negative, trace, 1+, 2+, 3+, 4+ to increasing grade scores.

[0068] For physiological indicators with positive and negative types, the conversion function can map negative to 0 and positive to 1.

[0069] For count-type physiological indicators, the transformation function can be: using the original count as the equivalent value of the physiological indicator, or using square root transformation, for example, the equivalent value of the j-th processable indicator is y_{j,t}=sqrt(x_{j,t}+0.5); sqrt represents the square root function.

[0070] Then, based on the equivalent values ​​of the manageable indicators, the standardized values ​​of the manageable indicators are determined. The specific process is as follows: When perturbation condition parameters exist, the standardized value Y_{j,t} of the j-th manageable index at time t is calculated as follows: Y_{j,t}=y_{j,t}-G_{j,t}-H_{j,t}-M_{j,t} Where Y_{j,t} represents the standardized value of the j-th processable index at time t, G_{j,t} represents the sampling condition perturbation value of the j-th processable index at time t, H_{j,t} represents the transient event perturbation value of the j-th processable index at time t, and M_{j,t} represents the detection quality perturbation value of the j-th processable index at time t.

[0071] In the above formula for calculating the standardized value Y_{j,t}: G_{j,t}=B_{z,j}^TZ_{j,t}+b_{u,j}+b_{m,j} Where B_{z,j} represents the coefficient of the j-th processable index on the sampling condition vector, T represents the transpose, Z_{j,t} represents the sampling condition vector of the j-th processable index at time t, b_{u,j} represents the individual offset of the target object on the j-th processable index, and b_{m,j} represents the batch offset of the current detection batch on the j-th processable index.

[0072] H_{j,t}=Σ_kβ_{e,j,k}S_{k,t}exp(-d_{k,t} / τ_k)r_{k,t} Where Σ represents summation, k represents the k-th transient event, β_{e,j,k} represents the coefficient of the j-th processable index for the k-th transient event, S_{k,t} represents the severity of the k-th transient event at time t, exp represents the exponential function with the natural constant e as the base, d_{k,t} represents the time interval between the k-th transient event and the current detection time t, τ_k represents the decay time constant of the k-th transient event, and r_{k,t} represents the credibility of the event record of the k-th transient event at time t.

[0073] M_{j,t}=B_{q,j}^Tq_{j,t} Where B_{q,j} represents the coefficient of the j-th processable index on the detection quality vector, and q_{j,t} represents the detection quality vector of the j-th processable index at time t. The detection quality vector may include device error flags, sample size flags, calibration status, repeated measurement coefficient of variation, linear range flags, and storage condition flags.

[0074] For example, the sampling condition vector may include: sample type or data source type, standard sampling time window, standard testing posture, standard testing site, no preset transient events, qualified sample volume or signal quantity, qualified equipment calibration, valid consumables or reagents, qualified reading quality or signal quality, environmental conditions within permissible range, and indicator specificity correction conditions. Sampling condition information may include sample type, sampling time, testing time, interval from wake-up time, time interval between the current sampling or measurement and the previous sampling or measurement, sample dilution or concentration index, endogenous correction index, equipment batch, reagent batch, consumable batch, testing temperature, ambient humidity, testing posture, testing site, pre-test dietary status, and pre-test exercise status.

[0075] Transient event vectors include events that may affect test results within a preset time window before testing (24 to 72 hours prior to testing), including but not limited to strenuous exercise, abnormal water intake, abnormal food intake, fever or acute discomfort, sleep deprivation, severe stress, infection symptoms, changes in medication or supplements, caffeine or alcohol intake, menstruation or sample contamination risk, changes in posture before testing, dehydration or rehydration, and abnormal sensor wearing. The occurrence of each type of event can be recorded to obtain its corresponding transient event vector; transient event information includes all transient event vectors.

[0076] The detection quality vector includes at least one of the following: device error identifier, sample size or signal quantity identifier, calibration status identifier, coefficient of variation of multiple physiological index tests, detection range identifier, reading quality identifier, image quality identifier, signal stability identifier, reagent quality identifier, consumable quality identifier, sensor contact quality identifier, sample preservation condition identifier, consumable preservation condition identifier, abnormal ambient temperature identifier, abnormal ambient humidity identifier, communication error identifier, or device power error identifier.

[0077] The standardized values ​​of the processable indicators obtained from the above calculations are used to calculate the state probability and boundary distance of the preset monitoring state.

[0078] Step 104: Perform uncertainty assessment on the target object to obtain the uncertainty vector of the target object.

[0079] Specifically, if the number of processable indicators is one, then the uncertainty source vector includes at least two of the following: uncertainty of sampling conditions, uncertainty of transient events, uncertainty of reference boundary distance, uncertainty of individual baseline reliability, uncertainty of measurement error, and uncertainty of missing data.

[0080] If there are multiple processable indicators, the uncertainty source vector is obtained as follows: First, obtain the uncertainty source components of each of the processable indicators; the uncertainty source components of each processable indicator still include at least two of the following: uncertainty of sampling conditions, uncertainty of transient events, uncertainty of reference boundary distance, uncertainty of individual baseline reliability, uncertainty of measurement error, and uncertainty of missing data.

[0081] According to the trigger priority, maximum value, weighted average, or rules corresponding to each of the processable indicators, the uncertainty source components of all the processable indicators are aggregated to obtain the uncertainty source vector. That is, for all processable indicators, each uncertainty item in all these uncertainty source components is processed by weighting, selecting the maximum value, or selecting according to priority to obtain a new uncertainty item. Combining all the determined new uncertainty items yields the uncertainty source vector; this uncertainty source vector can be used for calculating the action reward of candidate sampling actions and determining baseline update admission.

[0082] Among them, the uncertainty of sampling conditions is determined by sample type, sampling time or measurement condition field; the uncertainty of transient events is determined by event occurrence identifier, event severity, event time interval and event record reliability; the uncertainty of reference boundary distance is determined by the distance between the standardized value and the reference boundary; the uncertainty of individual baseline reliability is determined by the number of effective baseline samples and the baseline interval width; the uncertainty of measurement error is determined by equipment error, repeated measurement coefficient of variation or calibration status; and the uncertainty of missing data is determined by the proportion of missing necessary fields.

[0083] For example, the expression for the uncertainty source vector u_t of the target object is: u_t=[u_sample,t,u_event,t,u_boundary,t,u_baseline,t,u_measure,t,u_missing,t]; Where u_sample,t represents the uncertainty of sampling conditions; u_event,t represents the uncertainty of transient events; u_boundary,t represents the uncertainty of the distance to the reference boundary; u_baseline,t represents the uncertainty of the reliability of the individual baseline; u_measure,t represents the uncertainty of measurement error; and u_missing,t represents the uncertainty of missing data.

[0084] In this embodiment, the uncertainties of sampling conditions, transient events, reference boundary distance, individual baseline reliability, measurement error, and missing data can be normalized to 0 to 1; the higher the value, the greater the impact of the uncertainty on the current judgment of the target object.

[0085] In the uncertainty source vector u_t: The expression for the uncertainty of the reference boundary distance is: u_boundary,t=max_{j,l}exp(-|Y_{j,t}-RB^Y_{j,l}| / (σ_{j,t}+ε)); Where max represents the maximum value; j represents the index of the j-th processable index; l represents the l-th reference boundary value; RB^Y_{j,l} represents the standardized value obtained by transforming the l-th reference boundary value of the j-th processable index (i.e., the transformation from the detected value to the standardized value mentioned above); σ_{j,t} represents the comprehensive uncertainty of the j-th processable index at time t; || represents the absolute value; the closer the detection result is to the reference boundary, the higher the boundary uncertainty.

[0086] RB^Y_{j,l}=standardize_j(f_j(θ^{ref}_{j,l})); Where θ^{ref}_{j,l} represents the l-th recognized reference boundary of the j-th manageable index, f_j represents the transformation function of the j-th manageable index, and standardize_j represents the same standardization method as Y_{j,t}; thus, RB^Y_{j,l} and Y_{j,t} are on the same scale.

[0087] σ_{j,t}=sqrt(σ^2_{meas,j,t}+σ^2_{sample,j,t}+σ^2_{event,j,t}) Where sqrt represents the square root function, σ^2_{meas,j,t} represents the measurement error variance, σ^2_{sample,j,t} represents the variance introduced by the sampling conditions, and σ^2_{event,j,t} represents the variance introduced by the insufficient explanation of transient events. The measurement error variance, the variance introduced by the sampling conditions, and the variance introduced by the insufficient explanation of transient events can be preset values, such as those estimated from the indicator configuration table, equipment error parameters, historical repeated measurement results, or authorized logs.

[0088] The expression for the uncertainty of individual baseline reliability is: u_baseline,t=min(1,σ_{base,j,t} / (s_j+ε)+1 / (n_{eff,j,t}+1)); Where min represents taking a smaller value, σ_{base,j,t} represents the standard error or confidence interval width of the current individual baseline for the j-th manageable indicator; s_j represents the scaling parameter of the j-th manageable indicator; and n_{eff,j,t} represents the number of effective samples used for baseline estimation for the j-th manageable indicator. Baseline uncertainty increases when the number of effective samples is small or the baseline interval width is too large.

[0089] In addition, when there is only a single quantifiable metric, u_baseline,t can be directly taken as the value of that metric; when there are multiple quantifiable metrics, the overall u_baseline,t can be obtained according to the aforementioned multi-quantifiable metric aggregation rules.

[0090] The expression for the uncertainty of missing data is: u_missing,t=count(missingrequiredfields) / count(requiredfields); Here, `count` represents the counting function, `missingrequiredfields` represents the necessary fields that are missing from the current processable metric, and `requiredfields` represents the fields required for the current processable metric and action. Missing context must not be interpreted as the event not occurring. Missing fields increase the uncertainty of `u_missing,t` or its related sources.

[0091] The expression for the uncertainty of the sampling conditions is: u_sample,t=min(1,α_1m_type,t+α_2m_time,t+α_3m_condition,t+α_4v_creatinine,t); Wherein, m_type,t represents the missing sample type or measurement condition marker at time t; m_time,t represents the abnormal sampling or measurement time marker at time t; m_condition,t represents the marker that the measurement condition at time t is inconsistent with the indicator configuration; v_creatinine,t represents the degree of abnormality of urine creatinine or equivalent quality control item at time t; α1 to α_4 represent the preset weights.

[0092] The expression for the uncertainty of transient events is: u_event,t=min(1,Σ_kγ_ke_{k,t}S_{k,t}exp(-d_{k,t} / τ_k)+Σ_kγ^m_kmissing_{k,t}); Wherein, γ_k represents the weight of the k-th transient event, γ^m_k represents the weight of the missing marker of the k-th transient event, e_{k,t} represents whether the k-th transient event occurred at time t, S_{k,t} represents the severity of the k-th transient event at time t; d_{k,t} represents the time interval between the k-th transient event and the current detection time t, τ_k represents the event decay time constant of the k-th transient event, and missing_{k,t} represents the missing event record marker of the k-th transient event at time t.

[0093] The expression for the uncertainty of measurement error is: u_measure,t=min(1,β_1err_t+β_2CV_{j,t}+β_3range_{j,t}+β_4cal_t); Where err_t represents the device error or reading error flag at time t; CV_{j,t} represents the repeated measurement variation coefficient of the j-th processable index at time t; range_{j,t} represents the flag that the detection value of the j-th processable index at time t exceeds the corresponding linear range; cal_t represents the calibration abnormality or calibration unknown flag at time t; β_1 to β_4 represent the preset weights.

[0094] Step 105: Based on the monitoring state probability distribution and the uncertainty vector, generate a candidate sampling action set for the target object, wherein the candidate sampling action set includes multiple candidate sampling actions.

[0095] Specifically, candidate sampling actions can be selected from the candidate sampling action library based on the triggered processable indicators, main sources of uncertainty, current preset monitoring status, data quality requirements, user authorization status, and availability of external reference detection methods.

[0096] Specifically, the processing device is pre-configured with a candidate sampling action library, which can select multiple candidate sampling actions from the candidate sampling action library to form a candidate sampling action set based on the probability distribution of the current monitoring state of the target object and the uncertainty vector. The candidate sampling actions in the candidate sampling action library include at least three of the following: maintaining regular detection, retesting within a preset time window, retesting according to preferred sample conditions or measurement conditions, retesting after a brief event decay, paired sampling retesting, continuous sampling retesting, retesting after completing the key context, and external reference detection confirmation.

[0097] For example, the candidate sampling action set A_t={a0,a1,a2,a3,a4,a5,a6,a7}. Here, a0 represents maintaining regular detection; a1 represents retesting within a preset time window; a2 represents retesting according to preferred sample conditions or measurement conditions; a3 represents retesting after a brief event decay; a4 represents paired sampling retesting; a5 represents continuous sampling retesting; a6 represents retesting after completing the key context; and a7 represents external reference detection confirmation, i.e., external reference detection confirmation method confirmation suggestion.

[0098] Specifically, if the main source of uncertainty is the uncertainty of sampling conditions, then retesting under preferred sample conditions or measurement conditions should be prioritized; if the main source of uncertainty is the uncertainty of transient events, then retesting after the transient events have decayed should be prioritized; if the main source of uncertainty is the uncertainty of missing data, then retesting after completing the key context should be prioritized; if the monitoring status is boundary pending confirmation or continuous deviation pending confirmation, then paired sampling retesting, continuous sampling retesting, or external reference detection confirmation should be prioritized.

[0099] Candidate sampling actions can be selected based on the current sample type and the source of uncertainty of the target object.

[0100] For example, if the processable indicator is a urine indicator and the sample type is urine, the selected candidate sampling actions include: a0 (urine sample testing at a fixed time), a1 (e.g., morning urine retesting within 24 to 72 hours), a2 (e.g., morning urine retesting on two consecutive days), and a4 (e.g., daytime urine and the next day's morning urine paired for retesting). For example, for a manageable indicator such as blood pressure, the selected candidate sampling actions include: a0 (fixed time period detection), a1 (retest within 24 to 72 hours), a2 (retest in a resting state), and a5 (continuous sampling and retest).

[0101] For example, for manageable indicators such as blood glucose or glycated hemoglobin, the selected candidate sampling actions include: a2 (retesting in the morning on an empty stomach or after a meal) and a7 (confirmation by laboratory testing).

[0102] In cases where transient events dominate, the selected candidate sampling actions are added to a3 (retesting after transient event decay).

[0103] In cases where context record data is missing, the selected candidate sampling action is added to a6 (the key context is filled in first and then retested).

[0104] The examples above are for illustrative purposes only and do not constitute disease diagnosis or treatment advice.

[0105] Step 106: Based on the monitoring state probability distribution and the uncertainty vector, predict the action benefit of performing each of the candidate sampling actions.

[0106] Specifically, for each candidate sampling action, the action benefit of performing the candidate sampling action is calculated based on at least three of the following: uncertainty change of performing the candidate sampling action, false alarm cost, false negative cost, acknowledgment delay cost, sampling burden cost, and resource consumption cost.

[0107] The uncertainty change in performing the candidate sampling action is represented by the change between the current state entropy and the expected state entropy after performing the candidate sampling action. The state entropy is used to measure the dispersion of the probability distribution of the current monitored state. The formula for calculating the state entropy at the current moment is: E_t=-Σ_s b_t(s)log(b_t(s)+ε); Where E_t represents the state entropy at time t, s represents the preset monitoring state s in the preset monitoring state set S, b_t(s) represents the conditional probability of being in the preset monitoring state s at time t, log represents the natural logarithm function, and ε represents a small positive number used to avoid taking the logarithm of zero. The more dispersed the conditional probabilities of the multiple preset monitoring states indicated by the monitoring state probability distribution, the higher the current state entropy.

[0108] Σ_s b_t(s)log(b_t(s)+ε) means: traverse each preset monitoring state in the preset monitoring state set S, calculate the product of b_t(s) and log(b_t(s)+ε) for each preset monitoring state, and then sum the calculated products for all preset monitoring states.

[0109] Then, the expected change in state entropy, i.e., the reduction in state entropy, is calculated for each preset candidate sampling action performed by the target object. For example, the formula for calculating the change in state entropy ΔE(a) for performing preset candidate sampling action a is: ΔE(a)=E_t-Σ_oP(o|a,b_t,u_t)E_{t+1}(o,a); Where 'o' represents the possible future observation result after performing candidate sampling action 'a', and the future observation result may include at least one of the following: future detection value, future standardized value, sample type or measurement conditions, sampling time, quality label, missing label, context completion result, and external reference detection result; P(o|a,b_t,u_t) represents the probability of obtaining future observation result 'o' under the conditions of the monitoring state probability distribution b_t and uncertainty source vector u_t, and E_{t+1}(o,a) represents the expected state entropy after obtaining future observation result 'o' after performing candidate sampling action 'a'. If there are not enough log samples, P(o|a,b_t,u_t) can be estimated using expert rule tables, simulation samples, or historical experience transition tables.

[0110] Subsequently, the action benefit of the target object performing each candidate sampling action can be predicted. Taking the execution of the preset candidate sampling action a as an example, the calculation formula for the action benefit of executing the preset candidate sampling action a is as follows: U(a)=λ_I·ΔE(a)-λ_F·C_false(a)-λ_M·C_miss(a)-λ_D·C_delay(a)-λ_B·C_burden(a)-λ_R·C_resource(a); Wherein, U(a) represents the benefit of performing candidate sampling action a, ΔE(a) represents the expected reduction in state entropy after performing candidate sampling action a, C_false(a) represents the false alarm cost, C_miss(a) represents the missed alarm cost, C_delay(a) represents the acknowledgment delay cost, C_burden(a) represents the sampling burden cost, C_resource(a) represents the resource consumption cost, λ_I represents the weight corresponding to the state entropy reduction benefit, λ_F represents the weight corresponding to the false alarm cost, λ_M represents the weight corresponding to the missed alarm cost, λ_D represents the weight corresponding to the acknowledgment delay cost, λ_B represents the weight corresponding to the sampling burden cost, and λ_R represents the weight corresponding to the resource consumption cost. The weights represented by λ_I, λ_F, λ_M, λ_D, λ_B, and λ_R are preset values, which can be specifically set by the indicator configuration table, risk level, user-acceptable sampling frequency, and business rules.

[0111] In the above formula for calculating the benefits of actions, the costs (false alarm cost, false negative cost, confirmation delay cost, sampling burden cost, and resource consumption cost) can be determined using values ​​from the preset configuration table, or they can be determined using the following normalized scoring formula: C_false(a)=κ_F×P_transient,t×repeat_intensity(a) C_miss(a)=κ_M×P_pending,t×delay_factor(a) C_delay(a)=κ_D×max(0,Δt(a)-t^{max}_j) C_burden(a)=κ_{B1}×n_sample(a)+κ_{B2}×d_cont(a)+κ_{B3}×h_condition(a)+κ_{B4}×(1-adherence_{i,t}) C_resource(a)=κ_R×resource(a) Wherein, κ_F, κ_M, κ_D, κ_{B1}, κ_{B2}, κ_{B3}, κ_{B4} and κ_R represent preset cost coefficients; the above cost coefficients can be given by the indicator configuration table, the strategy configuration table, the user authorization history log, or the business rules fixed by the configuration table; P_transient,t represents the probability of the dominant transient perturbation state at time t, which can be b_t(S3) or the sum of transient perturbation state probabilities specified in the indicator configuration table; repeat_intensity(a) represents the retest intensity of candidate sampling action a; P_pending,t represents the sum of probabilities of the set of states to be confirmed at time t. P_pending,t=Σ_{s∈S_pending}b_t(s); `delay_factor(a)` represents the delay factor of candidate sampling action a; `Δt(a)` represents the waiting time corresponding to candidate sampling action a; `t^{max}_j` represents the maximum allowed confirmation waiting time for the j-th processable indicator; `n_sample(a)` represents the number of samples corresponding to candidate sampling action a; `d_cont(a)` represents the number of consecutive sampling days corresponding to candidate sampling action a; `h_condition(a)` represents the difficulty of the sample type or measurement condition requirements corresponding to candidate sampling action a; `adherence_{i,t}` represents the historical compliance estimate of target object i at time t, which can be estimated by authorization history logs or configuration tables and is a preset value; `resource(a)` represents the resource consumption score corresponding to candidate sampling action a.

[0112] Step 107: Based on the action benefits of each of the candidate sampling actions, select a target candidate sampling action from the plurality of candidate sampling actions, and output a confirmation sampling strategy corresponding to the target candidate sampling action. The confirmation sampling strategy includes at least two of the following: sampling time, sample type or measurement conditions, number of samplings, and verification type.

[0113] For example, the candidate sampling action with the highest action benefit among all candidate sampling actions can be directly selected as the target candidate sampling action; In some embodiments, executable candidate sampling actions can be first selected from all candidate sampling actions according to preset selection rules, and then candidate sampling actions whose action benefits conform to the preset selection rules can be selected as target candidate sampling actions. The preset selection rules may include at least one of the following: highest action benefit, priority of parallel actions, quality and safety rules, user authorization status, availability of external reference detection confirmation method, or coverage rules specified in the configuration table.

[0114] For example, for urine ratio indicators such as UACR or UPCR, if the sample type is unknown or the degree of dilution and concentration is abnormal, morning urine retesting, paired sampling retesting, or continuous sampling retesting can be set as the priority candidate actions; for blood pressure indicators, if the resting conditions are unclear, repeated measurements at fixed time periods under resting conditions can be selected as the priority; for fasting blood glucose, random blood glucose, continuous blood glucose statistics, or glycated hemoglobin, if the measurement conditions or data sources are inconsistent, fasting, postprandial, continuous monitoring statistical verification, or external reference test confirmation that meets the indicator configuration can be selected as the priority.

[0115] a_t=argmax_{a∈A_t,L(a)=1}U(a) Where argmax represents the independent variable that maximizes the function value in the default implementation; a_t represents the target candidate sampling action selected at time t; A_t represents the set of available candidate sampling actions at time t; L(a) represents the constraints on candidate sampling action a, and L(a)=1 indicates that candidate sampling action a meets the corresponding constraints and can be executed; the constraints may include sampling time window, upper limit of action frequency, data quality requirements, availability of external reference detection confirmation method confirmation suggestions, and user authorization status.

[0116] The preset selection rules may include selecting the candidate sampling action with the highest action benefit, or may include parallel action priority, quality and safety rules, user authorization status, availability of external reference detection confirmation methods, or coverage rules specified in the configuration table.

[0117] After selecting the target candidate sampling action, at least two of the following can be determined based on the target candidate sampling action: sampling time, sample type or measurement conditions, number of samplings, and verification type. The confirmation sampling strategy corresponding to the target candidate sampling action is then obtained and output.

[0118] When the sample size is small, this method can employ rule tables, Bayesian state models, and explicit reward functions. Rule tables are used to define executable actions. Bayesian state models are used to estimate the probability of monitored states; explicit reward functions are used to compare candidate actions.

[0119] After the number of historical log samples reaches a preset quantity, at least one of the probability of the future observation result or the action reward can be estimated by an offline training model; the offline training model includes at least one of a generalized linear model, gradient boosting tree, temporal neural network, contextual gambling machine, or offline strategy evaluation model; the offline training model only ranks candidate actions within the candidate sampled action set and preset constraints. The learning model only ranks candidate actions within the candidate action set and constraints, without removing quality constraints, action frequency constraints, and external reference detection confirmation method confirmation suggestion constraints.

[0120] Training data uses only authorized logs. After training, the model needs to be compared with replay data in terms of action frequency, acknowledgment latency, invalid retests, and state entropy decrease. If the preset evaluation criteria are not met, the method retains the rule table as the primary action selection method.

[0121] Step 108: Based on at least one of the following: the uncertainty of transient events u_event,t, the uncertainty of measurement errors u_measure,t, the uncertainty of missing data u_missing,t, and the probability of the state to be confirmed P_pending,t, determine the baseline update admission flag for the current processable index; wherein, P_pending,t can be obtained by summing the probabilities of each state in the set of states to be confirmed S_pending; the active confirmation sampling strategy includes: the baseline update admission flag.

[0122] Specifically, for the manageable indicators of this detection, it can be determined whether the manageable indicators of this detection can be used for baseline update based on at least one of the following: the uncertainty of the corresponding transient event, the uncertainty of measurement error, the uncertainty of missing data, and the probability of the state to be confirmed. The corresponding baseline update admission flag is then generated and output.

[0123] For example, if the baseline update eligibility flag is denoted as baseline_update_eligible_t, then: baseline_update_eligible_t=I(Q_t≥q_min and u_event, t≤θ_E and u_measure, t≤θ_M and u_missing, t≤θ_N and Σ_{s∈S_pending}b_t(s)≤θ_S); Where I represents the indicator function, taking the value 1 when the condition is met and 0 when the condition is not met; Q_t represents the data quality score at time t; q_min represents the preset minimum quality threshold; θ_E represents the upper limit of uncertainty for transient events; θ_M represents the preset upper limit of uncertainty for measurement errors; θ_N represents the preset upper limit of uncertainty for missing data; S_pending represents the set of states to be confirmed; and θ_S represents the preset upper limit of probability for states to be confirmed. S_pending can include boundary states to be confirmed, continuous deviation states to be confirmed, and other states to be confirmed specified in the indicator configuration table; q_min, θ_E, θ_M, ​​θ_N, and θ_S are all preset values, which can be estimated from the indicator configuration table, data quality rules, action strategy configuration table, or authorized historical logs.

[0124] If `baseline_update_eligible_t` is 0, the detected values ​​of the processable metrics obtained this time will not enter the long-term baseline candidate pool. If the flag is 1, the sample can enter the long-term baseline candidate pool, but it still needs to meet the requirements of sample quantity, time interval, and state consistency.

[0125] Q_t= 1-min(1,ρ_1u_sample,t+ρ_2u_measure,t+ρ_3u_missing,t+ρ_4e_quality,t); Where ρ_1 to ρ_4 represent the preset quality deduction weights; e_quality,t represents the merging flag for insufficient sample size, abnormal storage conditions, abnormal reading images, or other quality abnormalities.

[0126] Q_t ranges from 0 to 1, with higher values ​​indicating higher data quality; Q_t can be obtained by aggregating the q_{j,t} of one or more indicators according to preset rules. Where q_{j,t} represents the detection quality vector of the j-th processable indicator at time t, and Q_t represents the data quality score used for baseline update admission judgment.

[0127] Step 109: Generate standardized individual trigger boundaries; the active confirmation sampling strategy includes: the standardized individual trigger boundaries.

[0128] Specifically, the formula for generating the standardized individual trigger boundary is: trigger^Y_{i,j,l,t}=RB^Y_{j,l}+δ^Y_{i,j,t}+p^Y_{i,j,l,t}; δ^Y_{i,j,t}=w_{i,j,t}(mean(Y_{i,j,B})-μ^Y_{group,j}); w_{i,j,t}=n_eff / (n_eff+k_δ); p^Y_{i,j,l,t}=d_{j,l}(n_1u_boundary,t+n_2u_measure,t+n_3u_missing,t+n_4u_baseline,t); Where, trigger^Y_{i,j,l,t} represents the standardized confirmation sampling trigger boundary used at time t for the j-th physiological index of target object i at the l-th reference boundary, RB^Y_{j,l} represents the recognized reference boundary after transformation function and standardization, δ^Y_{i,j,t} represents the individual offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, p^Y_{i,j,l,t} represents the protective offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, mean represents the mean value, and Y_{i,j,B} represents the entry of the j-th physiological index of target object i into the baseline candidate. The standardized value of the pool, μ^Y_{group,j}, represents the standardized center value of the j-th physiological indicator of target object i in the population. d_{j,l} represents the boundary direction of the j-th physiological indicator. d_{j,l} is +1 at high boundaries and -1 at low boundaries. The d_{j,l} value for the j-th physiological indicator without a directional boundary is specified by the indicator configuration table. η_1, η_2, η_3, and η_4 represent preset coefficients. The standardized individual trigger boundary can be used in conjunction with the target candidate sampling action to generate or adjust the sampling time, sample type or measurement conditions, number of samples, and verification type in the confirmation sampling strategy. It is only used to confirm the generation of the sampling strategy and does not replace the recognized reference boundary. If it is necessary to display the original detection value, it can be converted according to the inverse transformation and inverse standardization rules in the indicator configuration.

[0129] In some embodiments, the output confirmation sampling strategy fields may include: strategy number, sampling time window, sample type or measurement conditions, number of samplings, number of consecutive days, context fields to be completed, confirmation suggestion flag for external reference detection confirmation method, baseline update admission flag, triggering reason, and main sources of uncertainty. The output results are used for subsequent detection process scheduling and data processing.

[0130] For example, in the retesting of the urine albumin-creatinine ratio (UACR), the random urine UACR of the target subject is higher than the boundary in the configuration table, the sample type is daytime urine, and the urine creatinine indicates abnormal sample dilution or concentration; high sampling condition uncertainty and moderate boundary uncertainty are calculated. By comparison, the expected state entropy decrease of candidate sampling action a2 is determined to be higher than that of candidate sampling action a0, and candidate sampling action a2 is selected as the target candidate sampling action. The method output is to retest according to the preferred sample conditions within a preset time window, and then pause the processable indicators of this test to enter the long-term baseline.

[0131] For example, for a retest after a transient event decays, if the target object recorded strenuous exercise, fever, sleep deprivation, or severe stress before the detection, and the manageable index exceeds the confirmation sampling trigger boundary, and a high transient event uncertainty is calculated, if the retest after the transient event decays can significantly reduce the state entropy, then candidate sampling action a3 is selected as the target candidate sampling action, and the manageable index of this detection is suspended and enters the long-term baseline.

[0132] For example, for boundary verification retests, the standardized values ​​of the manageable indicators are close to the reference boundary, resulting in high boundary uncertainty. A comparison is made between paired sampling retests, continuous sampling retests, and conventional retests. If the continuous sampling retest meets the preset selection rules, it is selected as the target candidate sampling action, and the monitoring state probability distribution is recalculated after subsequent samples enter the system.

[0133] For example, when confirming blood pressure or blood glucose-related indicators as manageable indicators, if the test record shows incomplete measurement conditions, unclear rest status, or missing key context, repeat testing after completing the key context is selected as the target candidate sampling action. If multiple records are close to the same reference boundary under the same measurement conditions, repeated measurements at fixed time intervals or external reference detection confirmation are selected as the target candidate sampling action.

[0134] The second embodiment of the present invention relates to a processing device, which may be a mobile phone, desktop computer, laptop computer, server, home testing device, POCT (point-of-care testing) device, portable testing device, wearable device, urine testing device, blood testing device, saliva testing device, sweat testing device, or continuous monitoring device. The test values ​​can be obtained directly from the above-mentioned devices, or they can be transmitted from the testing device to a mobile terminal or server for execution of the method. The following description focuses on a urine testing device as an example, but the present invention is not limited to urine testing indicators.

[0135] In other words, the physiological indicators include at least one of biological sample detection indicators, body fluid detection indicators, sensor detection indicators, or wearable device detection indicators.

[0136] The physiological indicators include at least one of the following: albumin-to-creatinine ratio (ACR), urine albumin-to-creatinine ratio (UACR), protein-to-creatinine ratio (PCR), urine protein-to-creatinine ratio (UPCR), urine pH, urine specific gravity, urine glucose, urine ketones, urine occult blood, urine red blood cell count, blood pressure, fasting blood glucose, random blood glucose, continuous blood glucose statistics, glycated hemoglobin, body mass index (BMI), and resting heart rate.

[0137] For example, if the processing device is a toilet-mounted or freestanding urine testing device, then the physiological indicators include: urine testing indicators.

[0138] The processing device is used to execute the active confirmation sampling strategy generation method for physiological indicators as described in the first embodiment.

[0139] In one example, the processing device may include an active confirmation sampling strategy generation apparatus to implement an active confirmation sampling strategy generation method for physiological indicators. The active confirmation sampling strategy generation apparatus specifically includes: The data acquisition unit is used to acquire indicator-related data of the physiological indicators of the target object, including: current detection data and historical detection data; A manageable indicator determination unit is used to determine a manageable indicator from at least one physiological indicator; A state probability calculation unit is used to generate a monitoring state probability distribution of the target object corresponding to the physiological indicators based on the current detection data and the historical detection data. The monitoring state probability distribution indicates the conditional probability of multiple preset monitoring states. An uncertainty calculation unit is used to perform uncertainty assessment on the physiological indicators and obtain the uncertainty vector of the physiological indicators; A candidate action generation unit is used to generate a set of candidate sampling actions for the physiological indicator based on the monitoring state probability distribution and the uncertainty vector. The set of candidate sampling actions includes multiple candidate sampling actions. An action benefit calculation unit is used to predict the action benefit of performing each of the candidate sampling actions based on the monitoring state probability distribution and the uncertainty vector. The strategy output unit is used to select a target candidate sampling action from the plurality of candidate sampling actions based on the action benefits of each candidate sampling action, and output a confirmation sampling strategy corresponding to the target candidate sampling action. The confirmation sampling strategy includes at least two of the following: sampling time, sample type or measurement conditions, number of samplings, and verification type.

[0140] Since the first embodiment corresponds to this embodiment, this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment remain valid in this embodiment, and the technical effects achievable in the first embodiment can also be achieved in this embodiment. To reduce repetition, they will not be repeated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the first embodiment.

[0141] The third embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor of the active confirmation sampling strategy generation method in the first embodiment.

[0142] The preferred embodiments of the present invention have been described in detail above, but it should be understood that, if necessary, aspects of the embodiments can be modified to utilize aspects, features, and concepts from various patents, applications, and publications to provide other embodiments.

[0143] In light of the detailed description above, these and other changes can be made to the embodiments. Generally, the terminology used in the claims should not be considered limited to the specific embodiments disclosed in the specification and claims, but should be understood to include all possible embodiments together with the full scope of equivalents enjoyed by these claims.

Claims

1. A method for generating an active confirmation sampling strategy for physiological indicators, characterized in that, include: Obtain indicator-related data for at least one physiological indicator of the target object, wherein the indicator-related data includes: current detection data and historical detection data; A manageable indicator is determined from the at least one of the physiological indicators; Based on the current and historical detection data of the processable indicators, a monitoring status probability distribution of the target object is generated, wherein the monitoring status probability distribution indicates the conditional probability of multiple preset monitoring states. An uncertainty assessment is performed on the target object to obtain its uncertainty vector. Based on the monitoring state probability distribution and the uncertainty vector, a candidate sampling action set for the target object is generated, the candidate sampling action set including: multiple candidate sampling actions; Based on the monitoring state probability distribution and the uncertainty vector, predict the action benefit of performing each of the candidate sampling actions; Based on the action benefits of each of the candidate sampling actions, a target candidate sampling action is selected from the plurality of candidate sampling actions, and a confirmation sampling strategy corresponding to the target candidate sampling action is output. The confirmation sampling strategy includes at least two of the following: sampling time, sample type or measurement conditions, number of samplings, and verification type.

2. The active confirmation sampling strategy generation method according to claim 1, characterized in that, For each preset monitoring state in the preset monitoring state set S, the conditional probability of each preset monitoring state is calculated; the formula for calculating the conditional probability of each preset monitoring state is: b_t(s)=exp(r_t(s)) / Σ_{v∈S}exp(r_t(v)); r_t(s) = ω_s^TF_t + ξ_s; Where b_t(s) represents the conditional probability that the target object is in the preset monitoring state s at time t. If the number of processable indicators is 1, then F_t includes the target indicator features of the processable indicator at time t; if the number of processable indicators is multiple, then F_t includes the aggregated features formed by the target indicator features of all the processable indicators at time t. The target indicator features of the processable indicators include at least two of the following: the standardized value, recent change, boundary distance, uncertainty component, and data quality score of the processable indicator; ω_s represents the weight vector of the preset monitoring state s, ξ_s represents the bias term of the preset monitoring state s, and v represents any preset monitoring state in the preset monitoring state set S.

3. The active confirmation sampling strategy generation method according to claim 1, characterized in that, Based on the monitored state probability distribution and the uncertainty vector, predict the action benefit of performing each of the candidate sampling actions, including: For each candidate sampling action, the action benefit of performing the candidate sampling action is calculated based on at least three of the following: uncertainty change, false alarm cost, false negative cost, acknowledgment delay cost, sampling burden cost, and resource consumption cost.

4. The active confirmation sampling strategy generation method according to claim 3, characterized in that, The uncertainty change in performing the candidate sampling action is represented by the change between the current state entropy and the expected state entropy after performing the candidate sampling action; the state entropy is used to measure the dispersion of the current monitoring state probability distribution, and the formula for calculating the current state entropy is: E_t=-Σ_s b_t(s)log(b_t(s)+ε); Where E_t represents the state entropy at time t, s represents the preset monitoring state, b_t(s) represents the conditional probability of being in the preset monitoring state s at time t, log represents the natural logarithm function, and ε represents a small positive number used to avoid taking the logarithm of zero.

5. The active confirmation sampling strategy generation method according to claim 3, characterized in that, The formula for calculating the action benefit of performing the candidate sampling action is: U(a)=λ_I·ΔE(a)-λ_F·C_false(a)-λ_M·C_miss(a)-λ_D·C_delay(a)-λ_B·C_burden(a)-λ_R·C_resource(a); Where U(a) represents the action benefit of performing candidate sampling action a, ΔE(a) represents the expected reduction in state entropy after performing candidate sampling action a, C_false(a) represents the false alarm cost, C_miss(a) represents the missed alarm cost, C_delay(a) represents the acknowledgment delay cost, C_burden(a) represents the sampling burden cost, C_resource(a) represents the resource consumption cost, λ_I represents the weight corresponding to the state entropy reduction benefit, λ_F represents the weight corresponding to the false alarm cost, λ_M represents the weight corresponding to the missed alarm cost, λ_D represents the weight corresponding to the acknowledgment delay cost, λ_B represents the weight corresponding to the sampling burden cost, and λ_R represents the weight corresponding to the resource consumption cost.

6. The active confirmation sampling strategy generation method according to claim 5, characterized in that, The formula for calculating the expected change in state entropy ΔE(a) after performing candidate sampling action a is: ΔE(a)=E_t-Σ_oP(o|a,b_t,u_t)E_{t+1}(o,a); Where o represents the possible future observation result after performing candidate sampling action a, and the future observation result includes at least one of the following: future detection value, future standardized value, sample type or measurement condition, quality label, missing label, context completion result and external reference detection result; P(o|a,b_t,u_t) represents the probability of obtaining future observation result o under the conditions of the monitoring state probability distribution b_t and the uncertainty source vector u_t, and E_{t+1}(o,a) represents the state entropy at the next moment after obtaining future observation result o by performing candidate sampling action a.

7. The active confirmation sampling strategy generation method according to claim 1, characterized in that, Based on the action benefits of each of the candidate sampling actions, a target candidate sampling action is selected from the plurality of candidate sampling actions, including: From all the candidate sampling actions, select the candidate sampling action that satisfies the preset selection rules as the target candidate sampling action; the preset selection rules include at least one of the following: highest action benefit rule, parallel action priority rule, quality and safety rule, user authorization status rule, external reference detection availability rule, or coverage rule specified in the indicator configuration table.

8. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The indicator-related data also includes: indicator configuration data; the indicator configuration data includes: the indicator type, unit, reference boundary, transformation function, and quality requirements of the physiological indicator; The manageable indicator is determined from the at least one physiological indicator, including: For each of the physiological indicators, if the indicator configuration data of the physiological indicator indicates that the physiological indicator has a recognized threshold, reference range, grading boundary or positive / negative boundary, then the physiological indicator is determined to be a processable indicator. If the indicator configuration data of the physiological indicator does not indicate that the physiological indicator has a recognized threshold, reference range, grading boundary, or positive / negative boundary, then the physiological indicator is determined to be a non-processable indicator, and the physiological indicator is used as an auxiliary input in the context record data or the uncertainty source vector.

9. The active confirmation sampling strategy generation method according to claim 1, characterized in that, If the number of processable indicators is one, then the uncertainty source vector includes at least two of the following: uncertainty of sampling conditions, uncertainty of transient events, uncertainty of reference boundary distance, uncertainty of individual baseline reliability, uncertainty of measurement error, and uncertainty of missing data; wherein, the uncertainty of sampling conditions is determined by sample type, sampling time, or measurement condition field; the uncertainty of transient events is determined by event occurrence identifier, event severity, event time interval, and event record reliability; the uncertainty of reference boundary distance is determined by the standardized value and the reference boundary distance; the uncertainty of individual baseline reliability is determined by the number of valid baseline samples and the baseline interval width; the uncertainty of measurement error is determined by equipment error, repeated measurement coefficient of variation, or calibration status; and the uncertainty of missing data is determined by the proportion of missing necessary fields.

10. The active confirmation sampling strategy generation method according to claim 1, characterized in that, If there are multiple processable indicators, the uncertainty source vector is obtained as follows: First, obtain the uncertainty source components of each of the processable indicators; According to the trigger priority, maximum value, weighted average, or rules corresponding to each of the processable indicators, the uncertainty source components of all the processable indicators are aggregated to obtain the uncertainty source vector.

11. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The physiological indicators include at least one of the following: urine albumin-to-creatinine ratio, urine total protein-to-creatinine ratio, urine pH, urine specific gravity, urine glucose, urine ketones, urine occult blood, urine red blood cell count, blood pressure, fasting blood glucose, random blood glucose, continuous blood glucose statistics, glycated hemoglobin, body mass index, and resting heart rate.

12. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The current detection data and the historical detection data include at least three of the following: detection value, detection time, sample type, data source, quality marker, and missing marker.

13. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The data related to the indicator also includes: context recording data, which includes at least two of the following: exercise records, water intake records, sleep records, stress records, fever records, medication or supplement change records, equipment logs, and calibration status.

14. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The candidate sampling actions in the candidate sampling action set are selected from the candidate sampling action library. The candidate sampling actions in the candidate sampling action library include at least three of the following: maintaining regular detection, retesting within a preset time window, retesting according to preferred sample conditions or measurement conditions, retesting after a brief event decay, paired sampling retesting, continuous sampling retesting, retesting after completing the key context, and external reference detection confirmation. Specifically, if the main source of uncertainty is the uncertainty of sampling conditions, then retesting under preferred sample conditions or measurement conditions should be prioritized; if the main source of uncertainty is the uncertainty of transient events, then retesting after the transient events have decayed should be prioritized; if the main source of uncertainty is the uncertainty of missing data, then retesting after completing the key context should be prioritized; if the monitoring status is boundary pending confirmation or continuous deviation pending confirmation, then paired sampling retesting, continuous sampling retesting, or external reference detection confirmation should be prioritized.

15. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The preset monitoring states include at least three of the following: insufficient baseline, stable baseline, stable baseline but large fluctuations, short-term disturbances, boundary to be confirmed, continuous deviation to be confirmed, continuous deviation confirmed, and recent decline or recovery. The baseline can be an individual baseline of a single processable indicator or a comprehensive baseline state formed by the individual baseline states of multiple processable indicators through preset aggregation rules.

16. The active confirmation sampling strategy generation method according to claim 1, characterized in that, Based on at least one of the following: the uncertainty of transient events u_event,t, the uncertainty of measurement errors u_measure,t, the uncertainty of missing data u_missing,t, and the probability of the state to be confirmed P_pending,t, the baseline update admission flag for the current processable index is determined; wherein, P_pending,t can be obtained by summing the probabilities of each state in the set of states to be confirmed S_pending; the active confirmation sampling strategy includes: the baseline update admission flag.

17. The active confirmation sampling strategy generation method according to claim 1, characterized in that, The method further includes: generating standardized individual trigger boundaries; the active confirmation sampling strategy includes: the standardized individual trigger boundaries; the calculation formula for the standardized individual trigger boundaries is: trigger^Y_{i,j,l,t}=RB^Y_{j,l}+δ^Y_{i,j,t}+p^Y_{i,j,l,t}; δ^Y_{i,j,t}=w_{i,j,t}(mean(Y_{i,j,B})-μ^Y_{group,j}); w_{i,j,t}=n_eff / (n_eff+k_δ); p^Y_{i,j,l,t}=d_{j,l}(n_1u_boundary,t+n_2u_measure,t+n_3u_missing,t+n_4u_baseline,t); Where, trigger^Y_{i,j,l,t} represents the standardized confirmation sampling trigger boundary used at time t for the j-th physiological index of target object i at the l-th reference boundary, RB^Y_{j,l} represents the recognized reference boundary after transformation function and standardization, δ^Y_{i,j,t} represents the individual offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, p^Y_{i,j,l,t} represents the protective offset of the j-th physiological index of target object i at time t when transformed to the standardized scale, mean represents the mean value, and Y_{i,j,B} represents the j-th physiological index of target object i. The standardized values ​​of the physiological indicators entering the baseline candidate pool are: μ^Y_{group,j} represents the standardized center value of the j-th physiological indicator of target object i in the population; d_{j,l} represents the boundary direction of the j-th physiological indicator; d_{j,l} takes +1 at high boundaries and -1 at low boundaries; the d_{j,l} value of the j-th physiological indicator without a directional boundary is specified by the indicator configuration table; η_1, η_2, η_3, and η_4 represent preset coefficients; the standardized individual trigger boundary can be used together with the target candidate sampling action to generate or adjust the sampling time, sample type or measurement conditions, number of samplings, and verification type in the confirmation sampling strategy.

18. The active confirmation sampling strategy generation method according to claim 6, characterized in that, After the number of historical log samples reaches a preset number, at least one of the probability of the future observation result or the action benefit can be estimated by an offline training model; the offline training model includes at least one of a generalized linear model, gradient boosting tree, temporal neural network, contextual gambling machine or offline strategy evaluation model; the offline training model only sorts candidate actions within the candidate confirmation sampling action set and preset constraints.

19. An active confirmation sampling strategy generation device, characterized in that, include: The data acquisition unit is used to acquire indicator-related data of the physiological indicators of the target object, including: current detection data and historical detection data; A manageable indicator determination unit is used to determine a manageable indicator from at least one physiological indicator; A state probability calculation unit is used to generate a monitoring state probability distribution of the target object corresponding to the physiological indicators based on the current detection data and the historical detection data. The monitoring state probability distribution indicates the conditional probability of multiple preset monitoring states. An uncertainty calculation unit is used to perform uncertainty assessment on the physiological indicators and obtain the uncertainty vector of the physiological indicators; A candidate action generation unit is used to generate a set of candidate sampling actions for the physiological indicator based on the monitoring state probability distribution and the uncertainty vector. The set of candidate sampling actions includes multiple candidate sampling actions. An action benefit calculation unit is used to predict the action benefit of performing each of the candidate sampling actions based on the monitoring state probability distribution and the uncertainty vector. The strategy output unit is used to select a target candidate sampling action from the plurality of candidate sampling actions based on the action benefits of each candidate sampling action, and output a confirmation sampling strategy corresponding to the target candidate sampling action. The confirmation sampling strategy includes at least two of the following: sampling time, sample type, number of samplings, and verification type.

20. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a program that, when executed by the processor, performs the active confirmation sampling strategy generation method according to any one of claims 1 to 18.

21. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it completes the active confirmation sampling strategy generation method according to any one of claims 1 to 18.

Images