A risk prediction model for 28-day mortality of elderly sepsis patients after transfer to ICU and a construction method thereof

Abstract

The application provides a risk prediction model for 28-day mortality of an elderly sepsis patient after being transferred into an ICU and a construction method thereof, determines independent risk factors related to death within 28 days through single factor and multi-factor logistic regression analysis, makes a nomogram, and constructs the prediction model of the patent, and the comparison result of the AUC of the ROC curve of the standard APACHE II score and the frailty index (FI-lab) or other multi-index model in recent years indicates that the prediction model proposed in the patent has better performance.

Description

Technical Field

[0001] This invention belongs to the field of medical testing technology and relates to a risk prediction model for mortality rate of elderly sepsis patients 28 days after admission to the ICU and its construction method. Background Technology

[0002] Sepsis, one of the most common and fatal diseases in intensive care units (ICUs), has a complex pathogenesis involving dysregulation of the host immune response and excessive inflammatory response, with alterations in the number and function of lymphocytes playing a crucial role. Elderly patients (≥65 years old) are more susceptible to life-threatening sepsis due to increased age, declining immune function, and a higher risk of adverse health outcomes.

[0003] The patent application No. 202410761508.7, entitled "A method and its application for time trajectory typing based on peripheral blood lymphocyte count 72 hours after the transfer of sepsis patients to the ICU", discloses that: a nomogram is made using six indicators, namely heart rate, patient age, lymphocyte time trajectory typing, APACHE II score, peripheral blood NK cell count, and pulmonary infection, to predict the mortality risk of sepsis patients 28 days after the transfer to the ICU. However, this patent does not specifically study the mortality risk of a special population, namely elderly sepsis patients, within 28 days after the transfer to the ICU.

[0004] The article "Construction and Validation of Nonodal Chart Model for Mortality Risk in Elderly Sepsis Patients" [1] reported that the following risk factors were identified on the first day of admission: decreased minimum blood oxygen saturation, decreased minimum mean arterial pressure, increased maximum heart rate, decreased minimum body temperature, increased maximum respiratory rate, decreased minimum plasma albumin, increased maximum blood bilirubin, increased maximum blood lactate, increased maximum blood urea nitrogen, mechanical ventilation, increased age, liver dysfunction, and history of metastatic malignant tumors. The area under the ROC curve (AUC) of the nonodal chart model constructed using the above indicators to predict the mortality risk of elderly sepsis patients within 30 days of admission was 0.777 (95%). The percentage CI was 0.743–0.810, and the AUC for predicting mortality risk with external validation was 0.747 (95% CI = 0.733–0.762). This patent did not study the risk of death within 28 days in elderly sepsis patients under special circumstances (after admission to the ICU). The ROC-AUC value was low. The patent collected data from 1066 cases at the Second Affiliated Hospital of Wenzhou Medical University from 2010 to 2020, and compared it with 335 clinical data points of elderly sepsis patients from a large multicenter intensive care database in the United States. The data collected by this patent has a large time span, many confounding factors, and the model includes many indicators, which affects the accuracy and ease of use of the research results.

[0005] Other studies have used immune status indicators of sepsis patients, including the neutrophil-to-lymphocyte ratio (NLR) and the neutrophil-to-platelet ratio (NPR), to assess prognosis [2]. Some studies have also analyzed the relationship between the frailty index (FI-lab) based on laboratory tests and the prognosis of patients with septic shock [3]. These indicators reflect the level of inflammatory response and immunosuppression in the body and the degree of frailty in patients to a certain extent, but they have certain limitations: First, these indicators are usually based on measurements at a single time point, ignoring the dynamic changes of indicators during the disease process; second, simply performing ratios on different cell types makes it difficult to comprehensively reflect the interaction of immune cells and the overall immune status; finally, the selection of cutoff values ​​is easily affected by the sample and measurement time point, and may lack robustness and universality.

[0006] To develop a more accurate predictive model for the mortality risk of elderly Chinese sepsis patients 28 days after admission to the ICU, a large sample size of elderly Chinese sepsis patients admitted to the ICU who meet the inclusion and exclusion criteria is required. To improve the predictive performance of the model, it is necessary to collect and accurately and in detail collect factors that may affect the mortality of these patients within a short period of time and conduct in-depth research, which is objectively quite difficult.

[0007] In summary, there is an urgent need to establish a mortality risk prediction model based on a large sample size of elderly Chinese sepsis patients 28 days after admission to the ICU, in order to more comprehensively, accurately and efficiently predict the mortality risk of elderly Chinese sepsis patients.

[0008] References:

[0009] [1]Xia Yizi 1 Zhang Jing 2 Liu Chen 2 Weng Jie 1 Song Jiaze 3 Tian Weihua 3 Xu Hui 1 Construction and validation of a nomogram model for mortality risk in elderly patients with sepsis [J]. Journal of Wenzhou Medical University, 2023, (No. 6).

[0010] [2] Li Lijuan, Liu Guoliang, Li Jun, Tao Xincao, Ruan Dongshi, Zhang Yunjian. Clinical value of peripheral blood lymphocyte count in assessing the prognosis of sepsis patients [J]. Chinese Journal of Emergency Medicine, 2015, (No. 11).

[0011] [3]Ding H, Li X, Zhang Summary of the Invention

[0012] The problem that the invention aims to solve:

[0013] In view of the fact that there is no method in the existing technology to accurately predict the risk of death of elderly sepsis patients 28 days after being transferred to the ICU, the purpose of this invention is to provide a predictive model and its construction method that can be used to accurately predict the risk of death of elderly sepsis patients 28 days after admission.

[0014] To achieve the above objectives, the specific technical solution of the present invention comprises the following steps:

[0015] S1. Elderly sepsis patients who needed to be admitted to the ICU due to their condition were included. The patient cohort was determined by screening according to the exclusion criteria. Baseline data, clinical information and laboratory test results of each patient in the cohort were collected. The study cohort consisted of 1205 elderly sepsis patients from the ICUs of four general tertiary hospitals in Beijing. Among them, 160 patients died within 28 days after being transferred to the ICU.

[0016] S2. Based on the frailty index (FI-lab) obtained after admission to the ICU, specific laboratory test data include complete blood cell count (hemoglobin, white blood cell count, platelet count), electrolytes (sodium, potassium, chloride, calcium, phosphate, magnesium), renal function (blood urea nitrogen, creatinine), liver function (alanine aminotransferase, aspartate aminotransferase, albumin, total protein, bilirubin), coagulation function (prothrombin time, activated partial thromboplastin time), glucose, and inflammatory markers (C-reactive protein, erythrocyte sedimentation rate). If the laboratory test results are within the normal reference range, the score for this item is 0; if they exceed the normal reference range, the score for this item is 1. FI-lab is the ratio of the sum of all abnormal values ​​to the total number of items. This invention divides FI-lab into three categories according to different risks: low risk (FI-lab < 0.45), medium risk (FI-lab 0.45-0.65), and high risk (FI-lab > 0.65).

[0017] S3. Peripheral blood lymphocytes and subsets were collected and analyzed within 72 hours after admission to the ICU. The collected lymphocyte count data were analyzed by time series analysis using a latent class mixed effects model. Based on the clustering results, patients were divided into four different phenotypes: high-declining (α type), stable-intermediate (β type), high-rising (γ type), and stable-low (δ type).

[0018] S4. The included patient cohort was randomly divided into two groups at a ratio of 7:3. The group with a ratio of 7 was set as the training set, and the group with a ratio of 3 was set as the validation set. The balance between the training set and the validation set was compared by statistical analysis.

[0019] S5. Convert the collected patient data, information, test results, and FI-lab and lymphocyte time trajectory typing into statistical variables, and use univariate analysis in the training set to screen out variables related to 28-day mortality.

[0020] S6. In the training set, the variables related to 28-day mortality obtained in S5 were further identified through binary logistic regression analysis to determine the independent risk factors associated with 28-day mortality, including: APACHE II score, patient origin, heart rate at admission, respiratory rate at admission, peripheral blood lymphocyte time trajectory phenotype within 72 hours after admission to the ICU, peripheral blood natural killer cell (NK cell) count, whether the admission was due to pulmonary infection and FI-lab risk category, as well as the complex regression equations between each risk factor and 28-day mortality.

[0021] S7. This complex equation was visualized simply using a nomogram to obtain a 28-day mortality prediction model for elderly sepsis patients, and the model discrimination and calibration were analyzed and evaluated respectively.

[0022] S8. Validate the prediction model obtained in S7 on the validation set, analyze the model discrimination and model calibration, and compare and evaluate it with the training set results.

[0023] S9. The ROC curve of the validated prediction model was compared with the ROC curves of the other seven prediction models. The results showed that the prediction model of S7 had the largest AUC value. The AUC values ​​of the ROC curves of the four models were below 0.7 and could not be used for prediction. Although the AUC values ​​of the ROC curves of the other three models were above 0.7, their prediction performance was low.

[0024] Compared with existing technologies, the beneficial effects of this invention are: it establishes a mortality risk prediction model based on a large sample size of elderly Chinese sepsis patients admitted to the ICU 28 days after admission.

[0025] 1. The model has good predictive performance;

[0026] 2. This model is more suitable for the Chinese population;

[0027] 3. The independent risk factors included in this model (such as FI-lab) can be easily obtained and calculated using electronic medical record systems in clinical practice. Attached Figure Description

[0028] Figure 1 Case screening flowchart and grouping of model training and validation sets;

[0029] Figure 2 Survival curves for 28-day mortality rate established using a frailty index based on laboratory testing;

[0030] Figure 3 Lymphocyte temporal trajectory typing in the model training set (A) and model validation set (B);

[0031] Figure 4 Risk prediction model for mortality rate of elderly sepsis patients 28 days after admission to ICU (nomogram);

[0032] Figure 5 (B) The ROC curve, AUC, and calibration curve of the model training set; (C) The ROC curve, AUC, and calibration curve of the model validation set; (D)

[0033] Figure 6 A comparison of the ROC curves and AUC values ​​of eight models, among which...

[0034] AUC1: The predicted ROC curve and its AUC value are derived from a model built using six indicators: frailty index (FI-lab), peripheral blood lymphocyte time trajectory classification, patient heart rate, APACHE II score, patient origin, and whether there is a pulmonary infection.

[0035] AUC2: Based on the patent application number 202410761508.7, "A method and its application for time trajectory typing of peripheral blood lymphocyte count 72 hours after the transfer of sepsis patients to ICU", the predicted ROC curve and its AUC value of the model were established using six indicators: heart rate, patient age, lymphocyte time trajectory typing, APACHE II score, peripheral blood NK cell count, and whether there is a lung infection.

[0036] AUC3: The predicted ROC curve and its AUC value of a model based on peripheral blood lymphocyte time trajectory typing.

[0037] AUC4: The predicted ROC curve and its AUC value of the model built using the decay index (FI-lab); AUC5: The predicted ROC curve and its AUC value of the model built using the APACHE II score.

[0038] AUC6: The predicted ROC curve and its AUC value of a model built using three indicators: frailty index (FI-lab), peripheral blood lymphocyte time trajectory classification, and patient heart rate.

[0039] AUC7: The predicted ROC curve and its AUC value of a model built using four indicators: frailty index (FI-lab), peripheral blood lymphocyte time trajectory classification, patient heart rate, and APACHE II score.

[0040] AUC8: The predicted ROC curve and its AUC value are based on a model established using five indicators: frailty index (FI-lab), peripheral blood lymphocyte time trajectory classification, patient heart rate, APACHE II score, and patient origin. Detailed Implementation

[0041] The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the invention. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0042] Unless otherwise specified, the experimental methods used in the following examples are conventional methods.

[0043] Unless otherwise specified, all materials and reagents used in the following examples are commercially available.

[0044] Example 1: Method for constructing a model to predict 28-day mortality rate in elderly sepsis patients based on laboratory-detected frailty index and lymphocyte time trajectory typing.

[0045] 1. Patient screening, data collection, and model dataset grouping

[0046] Inclusion criteria: Sepsis patients aged ≥65 years with an expected treatment time of more than 48 hours in the ICU and more than 2 lymphocyte counts were selected.

[0047] Exclusion criteria:

[0048] (1) Patients who survive no more than 48 hours;

[0049] (2) Patients with primary or acquired immunodeficiency or hematologic disorders, to exclude factors that may interfere with lymphocyte count;

[0050] (3) Patients with a missing item rate greater than 30% required to construct the Frailty Index (FI-lab).

[0051] This prospective study included 1,425 elderly sepsis patients admitted to the ICU of four tertiary hospitals in Beijing between June 2023 and November 2024. Based on the inclusion and exclusion criteria, 165 patients who survived less than 48 hours were excluded, 34 patients with primary or acquired immunodeficiency or hematologic diseases were excluded, and 21 patients with a missing item rate greater than 30% for constructing the Frailty Index (FI-lab) were excluded. A total of 1,205 elderly sepsis patients were included in this study, of whom 160 died within 28 days of admission to the ICU.

[0052] Using a random allocation method, 1205 subjects were randomly divided into two groups at a ratio of 7:3. The model training set had 844 subjects, and the model validation set had 361 subjects.

[0053] Figure 1 The flowchart for case screening in the cohort and the grouping of the model training and validation sets are presented.

[0054] 2. Data collection and variable selection

[0055] Elderly patients with sepsis were consecutively enrolled, and demographic and clinical data were recorded, including patient age, sex, medical history, whether admitted as an emergency patient, whether transferred due to pulmonary infection, infection focus, comprehensive organ function (expressed using the Sequential Organ Failure Scale (SOFA)), disease severity (expressed using the Acute Physiology and Chronic Health Evaluation II (APACHE II) score), baseline vital signs (including body temperature, heart rate, mean arterial blood pressure, respiratory rate, serum lactate, superior vena cava oxygen saturation, arterial-venous carbon dioxide partial pressure difference, and amount of vasoactive drugs used), and various organ functions (including left ventricular ejection fraction, oxygenation index, serum creatinine, blood urea nitrogen, alanine aminotransferase, total bilirubin, direct bilirubin, and serum amylase). All subjects' outcome-related indicators were collected, including ICU stay, hospital stay, ventilator use time, and 28-day mortality. Peripheral blood samples were collected from all subjects to analyze their immune and inflammatory responses. Status, including interleukin-6, interleukin-8, interleukin-10, tumor necrosis factor-α, neutrophil count, lymphocyte count, neutrophil-to-lymphocyte ratio, high-sensitivity C-reactive protein, and procalcitonin; peripheral blood was collected to analyze immunoglobulin A, immunoglobulin G, immunoglobulin M, complement 3, complement 4, and lymphocyte subsets [including total CD19+ B lymphocytes, total CD3+ T lymphocytes, CD4+ T lymphocytes, CD8+ T lymphocytes, and CD3-CD16+CD56+ natural killer (NK) cells].

[0056] In addition, peripheral blood samples were collected at 0, 24, 48, and 72 hours after admission to the ICU to detect absolute lymphocyte counts and generate longitudinal sequence measurements. Peripheral blood mononuclear cells were stained with fluorescent monoclonal antibodies and then analyzed by flow cytometry using a three-color EPICS-XL flow cytometer (Beckman Coulter, Brea, CA) to detect T cells (CD3+), CD4+ and CD8+ T cell subsets, B cells (CD19+), and NK cells (CD3-CD16+CD56+).

[0057] 3. Weakness Index

[0058] The Frailty Index-Laboratory (FI-lab) is a scoring system that measures an individual's overall health by comprehensively analyzing common laboratory data and clinical interventions. The FI-lab score typically uses a standard normal range to binarize relevant variables, converting variable values ​​to 0 (normal) or 1 (abnormal), and then calculates the proportion of items considered abnormal to obtain the FI-lab score.

[0059] This invention calculates the FI-lab score using complete blood cell count (hemoglobin, white blood cell count, platelet count), electrolytes (sodium, potassium, chloride, calcium, phosphate, magnesium), renal function (blood urea nitrogen, creatinine), liver function (alanine aminotransferase, aspartate aminotransferase, albumin, total protein, bilirubin), coagulation function (prothrombin time, activated partial thromboplastin time), glucose, and inflammatory markers (C-reactive protein, erythrocyte sedimentation rate). If the laboratory test results are within the normal reference range, the score for that item is 0; if they exceed the normal reference range, the score for that item is 1. FI-lab is the ratio of the sum of all abnormal values ​​to the total number of items; a higher score indicates a weaker patient.

[0060] Based on previous research, this invention categorizes FI-lab into three risk groups: low risk (FI-lab < 0.45), intermediate risk (FI-lab 0.45-0.65), and high risk (FI-lab > 0.65). Survival analysis based on these different risk levels of the frailty index suggests that elderly sepsis patients with a low-risk frailty index have a better prognosis. Details of the results can be found in [link to relevant documentation]. Figure 2 .

[0061] Reference to previous research results:

[0062] [1]Sapp D., Cormier B., Rockwood K., et al. (2023). The frailty indexbased on laboratory test data as a tool to investigate the impact of frailtyon health outcomes: A systematic review and meta-analysis. Age and Ageing, 52, 1-13.

[0063] [2]Nagae M., Umegaki H., Nakashima H., Nishiuchi T. (2025). FI-lab in the emergency department and adverse outcomes among acutely hospitalized olderadults. Archives of Gerontology and Geriatrics, 129, 105649.

[0064] 4. Lymphocyte time trajectory typing

[0065] 4.1 Latent Class Mixture Model

[0066] Latent Class Mixed Models (LCMMs) in machine learning are statistical models that allow researchers to group individuals based on their longitudinal data trajectories. This model assumes the existence of unobserved (latent) heterogeneous subgroups with different developmental trajectories. LCMMs combine features of traditional mixture-effects models and latent class models, capturing heterogeneity among individuals by introducing latent classes into the mixture-effects model. Its mathematical model is as follows:

[0067]

[0068] Among them, Y it The observation value representing individual i at time t,

[0069] X it It is a time variable.

[0070] β0, β1, and β2 are fixed-effect parameters.

[0071] b 0i b 1i It is a random effects parameter.

[0072] ∈it is the residual term.

[0073] This method has significant advantages over commonly used single-time-point immune indicators and traditional statistical analysis methods:

[0074] (1) LCMM can identify hidden dynamic change patterns from longitudinal data. Even if the absolute count values ​​of different patients are very different, they can be classified into the same category according to the relative change trend, which has stronger robustness and generalization ability.

[0075] (2) The categories obtained by time trajectory clustering have clear clinical interpretations, and each category corresponds to a specific pattern of immune status change, which is convenient for guiding treatment decisions;

[0076] (3) Trajectory-based typing can capture the nonlinear changes in immune status over time, which is more comprehensive and accurate than indicators based on a single time point, and can provide richer prognostic prediction information. In addition, as a semi-parametric model, LCMM has statistical advantages in parameter estimation and category inference, and can better balance model complexity and goodness of fit.

[0077] 4.2 Time series analysis of collected lymphocyte count data using LCMM

[0078] The steps for classifying the trend of changes in the patient's lymphocyte count are as follows:

[0079] (1) Data Acquisition

[0080] We collected patients' lymphocyte counts and combined them with time points (0, 24, 48, and 72 hours after admission to the ICU) to form a longitudinal dataset.

[0081] (2) LCMM modeling

[0082] Using the R language lcmm package (version: 2.1.0, https: / / cran.r-project.org / web / packages / lcmm / index.html), with patients as random effects, and time (wave) and its quadratic term (wave) 2 To construct a latent class mixture model for peripheral blood lymphocyte counts, considering both fixed and mixed effects, the following R language code is used (e.g., specifying 4 latent classes):

[0083] lcmm::lcmm(LymLevel~wave+I(wave^2),mixture=~wave,random=~wave,subject='icuaId',ng=4,data=datlong,B=m1,link="3-equi-splines")

[0084] Where LymLevel is the dependent variable lymphocyte count, wave is the time variable (0-3 correspond to 0, 24, 48 and 72h after admission to the ICU, respectively), mixture and random specify the relationship between latent class and individual random effect and time, ng is the specified number of latent classes, and link is the link function.

[0085] (3) Determining the optimal number of categories

[0086] By setting the parameter 'ng' in the code to different values ​​from 1 to 9, genotyping analysis was performed sequentially. The category affiliation of each individual was obtained using the BIC clustering curve. The results are shown below. Figure 3 For the PUMCH and MIMIC queues, the lowest score for the Bayesian Information Criterion (BIC) for different numbers of clusters was 4, and the optimal number of clusters was finally determined to be 4.

[0087] Figure 3 The x-axis represents the number of clusters, and the y-axis represents the Bayesian information criterion score.

[0088] (4) Clustering results analysis

[0089] Based on the changing trends, the lymphocyte change patterns of various categories were analyzed, and patients were divided into four different phenotypes: high-declining (α type), stable-intermediate (β type), high-rising (γ type), and stable-low (δ type). The results are shown in […]. Figure 3 In the model training set ( Figure 3 In A), the α type had the highest lymphocyte count at 0 hours, and the lymphocyte count decreased linearly from 0 hours to 72 hours; the β type had a moderate lymphocyte count at 0 hours, and the lymphocyte count remained basically stable from 0 hours to 72 hours; the γ type had a relatively high lymphocyte count at 0 hours, and the lymphocyte count increased linearly from 0 hours to 72 hours; and the δ type had the lowest lymphocyte count at 0 hours, and the lymphocyte count remained basically stable from 0 hours to 72 hours.

[0090] Analysis revealed that in the model validation set ( Figure 3 In B), the LCMM model also divides the lymphocyte counting trajectory optimization into four similar time trajectory phenotypic results, consistent with the model training set.

[0091] Figure 3 In the diagram, the x-axis represents the time points for lymphocyte measurement (0, 24, 48, and 72 hours after ICU admission), and the y-axis represents the lymphocyte count (x10⁻¹⁰). 9 / L); The shaded area around the smooth curve represents the confidence interval with a confidence level of 0.95.

[0092] Example 2: Establishing a predictive model for the mortality risk of elderly sepsis patients 28 days after admission to the ICU using nomograms.

[0093] 1. Independent risk factors for 28-day mortality in elderly patients with sepsis

[0094] In the model training set, variables associated with 28-day mortality in elderly sepsis patients were obtained through univariate analysis (see Table 1), and independent risk factors for 28-day mortality in elderly sepsis patients were further obtained through multivariate logistic regression analysis (see Table 2).

[0095] Table 1. Univariate analysis of 28-day survivors and deaths among elderly sepsis patients.

[0096]

[0097]

[0098] Variables with p-values ​​<0.15 in the univariate analysis of Table 1, including APACHE II score, heart rate, respiratory rate, oxygenation index, lactate, patient origin, pulmonary infection, T lymphocyte count, CD4+ T lymphocyte count, natural killer cell count, peripheral blood lymphocyte count time trajectory classification, and frailty index risk stratification, were included in the multivariate logistic regression analysis. The multivariate logistic regression analysis of 28-day mortality in elderly sepsis patients is shown in Table 2.

[0099] Table 2. Multivariate logistic regression analysis of 28-day mortality rate in elderly patients with sepsis.

[0100]

[0101] As shown in Table 2, the independent risk factors are: APACHE II score, heart rate, patient origin (transferred from internal medicine ward), pulmonary infection, peripheral blood lymphocyte count time trajectory classification, and frailty index risk stratification.

[0102] 2. Establishment of a 28-day mortality prediction model for elderly sepsis patients

[0103] 2.1 Establishment of a Clinical Risk Prediction Model

[0104] A nomogram is based on a multi-factor model. It integrates multiple independent variables and uses line segments with scales to draw the fitted functional relationships in the multi-factor model on the same plane. It is used to express the interrelationships and relative importance of the independent variables in the prediction model.

[0105] Independent risk factors associated with mortality within 28 days were incorporated into the predictive model. Following Occam's razor, the optimal model should have as few variables as possible. Further screening using nomograms resulted in the predictive model retaining six independent risk factors: heart rate, APACHE II score, patient origin, pulmonary infection, lymphocyte time trajectory typing, and frailty index risk stratification. A nomogram simplifies the complex relationship between 28-day mortality and these six independent risk factors. (Normally shown). Figure 4 As shown.

[0106] The first line is a score scale, with a score range of 0 to 100;

[0107] The second line shows the heart rate upon admission to the ICU. 20 beats / min corresponds to a score of 0 on the scale, and 220 beats / min corresponds to a score of 76 on the scale. The intervals between these values ​​are averaged.

[0108] The third row is the APACHE II score. An APACHE II score of 5 corresponds to a score of 0 on the score scale, and an APACHE II score of 55 corresponds to a score of 100 on the score scale. The intervals between these scores are averaged.

[0109] The fourth line is the source of the patient: 1 for elective surgery, 2 for internal medicine ward, and 4 for emergency surgery.

[0110] The fifth line indicates whether the patient was transferred due to a lung infection. 0 indicates no and corresponds to 0 points, while 1 indicates yes and corresponds to 19 points.

[0111] The sixth row shows the time trajectory typing of peripheral blood lymphocytes, where β type is 0 points, γ type is 2 points, α type is 15 points, and δ type is 25 points.

[0112] The seventh row is the risk classification of the frailty index, with low risk corresponding to 0 points, medium risk corresponding to 10 points, and high risk corresponding to 25 points.

[0113] The eighth and ninth rows represent the total score and its corresponding predicted 28-day mortality rate, with a probability range of 0.1 to 0.8.

[0114] In the nomogram, rows 2 through 7 represent independent risk factors for death within 28 days. Different risk factors correspond to different scores on the scale. The sum of the scores from rows 2 through 7 is projected onto the corresponding positions in rows 8 and 9, which represents the predicted probability of death within 28 days.

[0115] 2.2 Multiple Validation Analysis of Clinical Risk Prediction Model

[0116] After the prediction model is successfully established, the receiver operating characteristic (ROC) curves of the model training set are first plotted using receiver operating characteristic (ROC) curve analysis (see...). Figure 5A) and the area under the curve was calculated to be 0.866, indicating that the model has good discriminative power in the training set; correspondingly, the receiver operating characteristic (ROC) curve of the model validation set was plotted again using the same method (see A). Figure 5 C), and the area under the curve was calculated to be 0.873, indicating that the prediction model also has good discriminative power in the model validation set and is similar to the results in the model training set, indicating that the prediction model has good consistency in terms of discriminative power between the model training set and the model validation set.

[0117] Secondly, the predictive performance of the prediction model is graphically analyzed on the model training set using calibration curves (see...). Figure 5 As shown in B), the calibration curve on the model training set does not deviate significantly from the ideal curve, indicating that the prediction model has good calibration in the model training set. Correspondingly, the prediction performance of the prediction model is again graphically analyzed using the calibration curve on the model validation set (see [link to analysis]). Figure 5 As can be seen from D), the calibration curve of the model validation set does not deviate significantly from the ideal curve, indicating that the prediction model has good calibration in the model validation set and is similar to the results of the model training set. This shows that the prediction model has good consistency in terms of calibration between the model training set and the model validation set.

[0118] 2.3 Practical application of clinical risk prediction models:

[0119] When admitting elderly sepsis patients to the ICU, it is necessary to predict the risk of death within 28 days. This can be achieved by collecting clinical information such as heart rate, APACHE II score, patient origin, whether the transfer was due to pulmonary infection, peripheral blood lymphocyte count time series typing, and frailty index-related laboratory indicators. Then, scores are obtained based on the intervals of each risk factor, and the total score is calculated by summing these scores. The probability of death within 28 days is then displayed in the mortality probability column. A total score of 128 corresponds to a predicted mortality probability of 0.1 within 28 days, a total score of 186 corresponds to a predicted mortality probability of 0.5 within 28 days, and a total score of 226 corresponds to a predicted mortality probability of 0.8 within 28 days.

[0120] For example, suppose an 80-year-old sepsis patient from the internal medicine ward is transferred to the ICU due to a lung infection. Upon admission, the patient's heart rate is 120 beats / min, the APACHE II score upon admission is 25, the peripheral blood lymphocyte count time series genotype is α, and the FI-lab is calculated to be high-risk. According to the predictive model, the patient's total score is calculated to be 157 (with a heart rate of 120 beats / min corresponding to 38 points, an APACHE II score of 25 corresponding to 40 points, being from the internal medicine ward corresponding to 20 points, lung infection corresponding to 19 points, peripheral blood lymphocyte time series α corresponding to 15 points, and high-risk FI-lab corresponding to 25 points). The probability of death within 28 days is approximately 0.25.

[0121] Examples 3-10 below illustrate the comparison of predictive performance of different models. The sample set consisted of the entire cohort of 1205 patients. Different predictive models were used, and each patient was assigned a predictive score representing the probability of death within 28 days. Next, ROC curves were plotted using these predictive scores and the patients' actual survival status to evaluate the model's predictive performance. On the ROC curve, each point represents a specific predictive score threshold and corresponds to the true positive rate and false positive rate at that threshold. Finally, the area under the ROC curve (AUC) was calculated to quantify the model's predictive accuracy; the closer the AUC value is to 1, the stronger the model's predictive ability. In this way, the performance of different models in predicting 28-day mortality can be compared and evaluated.

[0122] Example 3: ROC curve of predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients from a cohort of 1205 patients using the model established in Example 2.

[0123] The prediction model constructed in Example 2 was used to predict the 28-day mortality risk of all 1205 elderly sepsis patients in the cohort. The results are as follows: Figure 6 As shown, the ROC-AUC1 is 0.880 (95% confidence interval 0.836–0.924).

[0124] Example 4: ROC curves (Comparative Example 1) of the model in patent application number 202410761508.7, "A method for time trajectory typing based on peripheral blood lymphocyte count 72 hours after septic patients are transferred to the ICU and its application," used to predict the risk of death of elderly septic patients 28 days after being transferred to the ICU in a cohort of 1205 patients.

[0125] Following the technical solution in patent application number 202410761508.7, "A Method and Application of Temporal Trajectory Classification Based on Peripheral Blood Lymphocyte Count 72 Hours After Transfer to ICU for Sepsis Patients," six indicators (including heart rate, patient age, lymphocyte temporal trajectory classification, APACHE II score, peripheral blood NK cell count, and pulmonary infection) were used to model and predict the 28-day mortality risk of all 1205 elderly sepsis patients in this cohort. The results are as follows: Figure 6 As shown, the ROC-AUC2 was 0.656 (95% confidence interval 0.576-0.736), which does not meet the requirement of AUC ≥ 0.7 for accurate prediction. Therefore, it cannot be used to predict the risk of death in elderly sepsis patients 28 days after admission to the ICU.

[0126] Example 5: ROC curve for predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients from a cohort of 1205 patients using peripheral blood lymphocyte time trajectory typing (Comparative Example 2).

[0127] Using peripheral blood lymphocyte time trajectory typing as a single indicator, logistic regression analysis was conducted to predict the 28-day mortality risk of all 1205 elderly sepsis patients in the cohort. The constructed logistic regression equation is as follows:

[0128] logit(p) = -5.543 - 0.781 × peripheral blood lymphocyte time trajectory γ type

[0129] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6 As shown, the ROC-AUC3 was 0.657 (95% confidence interval 0.580-0.733), which does not meet the requirement of AUC ≥ 0.7 for accurate prediction. Therefore, it cannot be used to predict the risk of death 28 days after elderly sepsis patients are transferred to the ICU.

[0130] Example 6: ROC curve of a method for predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients using a laboratory-detected frailty index in a cohort of 1205 patients (Comparative Example 3).

[0131] Using a single indicator, the frailty index (based on laboratory tests at ICU admission), logistic regression analysis was employed to predict the 28-day mortality risk of all 1205 elderly sepsis patients in the cohort. The constructed logistic regression equation is as follows:

[0132] logit(p) = -5.216 + 0.851 × Weakness Index

[0133] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6As shown, the ROC-AUC4 was 0.647 (95% confidence interval 0.569-0.724), which does not meet the requirement of AUC≥0.7 for accurate prediction. Therefore, it cannot be used to predict the risk of death 28 days after elderly sepsis patients are transferred to the ICU.

[0134] Example 7: ROC curve of a method for predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients using the APACHE II score in a cohort of 1205 patients (Comparative Example 4).

[0135] Using the APACHE II score at ICU admission as a single indicator, logistic regression analysis was performed to predict the 28-day mortality risk of all 1205 elderly sepsis patients in the cohort. The constructed logistic regression equation is as follows:

[0136] logit(p) = -4.903 + 0.129 × APACHE II score

[0137] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6 As shown, the ROC-AUC5 was 0.685 (95% confidence interval 0.611-0.759), which does not meet the requirement of AUC ≥ 0.7 for accurate prediction. Therefore, it cannot be used to predict the risk of death 28 days after elderly sepsis patients are transferred to the ICU.

[0138] Example 8: ROC curve of a method for predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients from a cohort of 1205 patients using three indicators: frailty index (laboratory-detected), peripheral blood lymphocyte time trajectory typing, and patient heart rate (Comparative Example 5).

[0139] The 28-day mortality risk of all 1205 elderly sepsis patients in the cohort was predicted using logistic regression analysis based on three indicators: frailty index (tested in the laboratory), peripheral blood lymphocyte time trajectory typing, and patient heart rate. The constructed logistic regression equation is as follows:

[0140] logit(p) = -6.381 + 0.687 × frailty index - 0.673 × peripheral blood lymphocyte time trajectory γ type + 0.022 × patient heart rate

[0141] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6 As shown, ROC-AUC6 is 0.748 (95% confidence interval 0.681-0.814), which meets the requirement of AUC ≥ 0.7 for accurate prediction. However, the AUC value is still low, and the prediction performance of this model is average, lower than the prediction model constructed in Example 2 (AUC1 = 0.880).

[0142] Example 9: ROC curve of a method for predicting the risk of death 28 days after admission to the ICU in elderly sepsis patients from a cohort of 1205 patients using four indicators: frailty index (laboratory test), peripheral blood lymphocyte time trajectory typing, patient heart rate, and APACHE II score (Comparative Example 6).

[0143] The 28-day mortality risk of all 1205 elderly sepsis patients in the cohort was predicted using logistic regression analysis based on four indicators: frailty index (tested in the laboratory), peripheral blood lymphocyte time trajectory typing, patient heart rate, and APACHE II score. The constructed logistic regression equation is as follows:

[0144] logit(p) = -7.195 + 0.495 × frailty index - 0.502 × peripheral blood lymphocyte time trajectory delta type + 0.017 × patient heart rate + 0.101 × APACHE II score

[0145] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6 As shown, the ROC-AUC7 is 0.814 (95% confidence interval 0.759-0.869), which meets the requirement of AUC≥0.7 for accurate prediction, but is lower than the prediction model constructed in Example 2 (AUC1=0.880).

[0146] Example 10: ROC curve of a method for predicting the risk of death 28 days after admission to the ICU in an entire cohort of 1205 patients with sepsis using five indicators: frailty index (laboratory test), peripheral blood lymphocyte time trajectory typing, patient heart rate, APACHE II score, and patient origin (Comparative Example 7).

[0147] Five indicators—frailty index (tested in the laboratory), peripheral blood lymphocyte time trajectory typing, patient heart rate, APACHE II score, and patient origin—were used to predict the 28-day mortality risk of all 1205 elderly sepsis patients in the cohort using logistic regression analysis. The constructed logistic regression equation is as follows:

[0148] logit(p) = -7.547 + 0.443 × frailty index - 0.439 × peripheral blood lymphocyte time trajectory delta type + 0.017 × patient's heart rate + 0.084 × APACHE II score + 0.572 The patient was from the internal medicine ward.

[0149] The prediction model was evaluated using ROC curves, and the results are as follows: Figure 6As shown, ROC-AUC8 is 0.824 (95% confidence interval 0.769-0.879), which meets the minimum accurate prediction requirement of AUC ≥ 0.7, but is lower than the prediction model constructed in Example 2 (AUC1 = 0.880).

[0150] In summary, compared with single indicators (such as the frailty index and peripheral blood lymphocyte time trajectory typing in laboratory tests), mature standard methods (APACHE II score), and different combinations of multiple indicators, the clinical risk prediction model proposed in this patent has the best ROC curve for predicting the probability of death risk in elderly sepsis patients 28 days after admission to the ICU, and the prediction method proposed in this invention has better efficacy.

[0151] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting the mortality rate of elderly sepsis patients 28 days after admission to the ICU, the method comprising: (1) Create a nomogram of independent risk factors associated with death within 28 days. The independent risk factors mentioned are: peripheral blood lymphocyte count time trajectory classification within 72 hours after ICU admission, patient origin, presence of pulmonary infection, APACHE II score, heart rate at ICU admission, frailty index FI-lab risk stratification; The independent risk factors were determined by multivariate logistic regression analysis of 28-day mortality in sepsis patients; The method for constructing the time trajectory typing of peripheral blood lymphocyte counts in elderly sepsis patients 72 hours after admission to the ICU is as follows: Data acquisition: Lymphocyte counts of sepsis patients at 0, 24, 48 and 72 hours after admission to the ICU are collected and combined with time to form a longitudinal dataset; LCMM modeling; Determining the optimal number of model categories through BIC clustering curves; Clustering result analysis; The patient sources mentioned refer to elective surgery, internal medicine wards, or emergency surgery. The frailty index FI-lab is determined by detecting complete blood cell count, electrolytes, renal function, liver function, coagulation function, glucose, and inflammatory markers. If the laboratory test results are within the normal reference range, the score for this item is 0; if they exceed the normal reference range, the score for this item is 1. FI-lab is the ratio of the sum of all abnormal values ​​to the total number of items. Complete blood cell count includes hemoglobin, white blood cell count, and platelet count; electrolytes include sodium, potassium, chloride, calcium, phosphate, and magnesium; renal function indicators include blood urea nitrogen and creatinine; liver function indicators include alanine aminotransferase, aspartate aminotransferase, albumin, total protein, and bilirubin; coagulation function indicators include prothrombin time and activated partial thromboplastin time; glucose and inflammatory markers include C-reactive protein and erythrocyte sedimentation rate. The aforementioned decay index FI-lab risk stratification is divided into low risk, medium risk and high risk, with low risk FI-lab < 0.45, medium risk FI-lab between 0.45 and 0.65, and high risk FI-lab > 0.65; (2) Using the nomogram to predict the risk of death within 28 days after elderly sepsis patients are transferred to the ICU. Clinical information was collected within 72 hours after the patient was transferred to the ICU, including peripheral blood lymphocyte count time trajectory classification, patient origin, presence of pulmonary infection, APACHE II score, heart rate at ICU admission, and frailty index risk stratification. Then, corresponding scores were obtained based on the intervals of each independent risk factor of the patient. The total score of the patient at this stage was obtained by summing the scores of each factor. The specific probability value of the patient dying within 28 days was obtained in the mortality probability column.

2. The method as described in claim 1, characterized in that, The nomogram is constructed by creating a nomogram based on independent risk factors associated with mortality within 28 days. This nomogram simplifies the complex relationship between the 28-day mortality probability and six independent risk factors. The nomogram includes: The first line is a score scale, with a score range of 0 to 100; The second line shows the heart rate upon admission to the ICU. 20 beats / min corresponds to a score of 0 on the scale, and 220 beats / min corresponds to a score of 76 on the scale. The intervals between these values ​​are averaged. The third row is the APACHE II score. An APACHE II score of 5 corresponds to a score of 0 on the score scale, and an APACHE II score of 55 corresponds to a score of 100 on the score scale. The intervals between these scores are averaged. The fourth line is the source of the patient: 1 for elective surgery, 2 for internal medicine ward, and 4 for emergency surgery. The fifth line indicates whether the patient was transferred due to a lung infection. 0 indicates no and corresponds to 0 points, while 1 indicates yes and corresponds to 19 points. The sixth row shows the time trajectory typing of peripheral blood lymphocytes, where β type is 0 points, γ type is 2 points, α type is 15 points, and δ type is 25 points. The seventh row is the risk classification of the frailty index, with low risk corresponding to 0 points, medium risk corresponding to 10 points, and high risk corresponding to 25 points. The eighth and ninth rows represent the total score and its corresponding predicted 28-day mortality rate, with a probability range of 0.1 to 0.

8. In the nomogram, rows two through seven represent independent risk factors for death within 28 days. Different risk factors correspond to different scores on the scale. The sum of the scores from rows two through seven is projected onto the corresponding positions in rows eight and nine, which represents the predicted probability of death within 28 days.

3. The method as described in claim 1, characterized in that, The AUC value of the ROC curve used by the method to predict the risk of death within 28 days in elderly patients with sepsis was 0.880.

Images