Construction method of risk prediction model for children living liver transplantation recipient to return to ICU (Intensive Care Unit)

Abstract

The invention relates to the field of medical prediction, and discloses a method for constructing a risk prediction model for children live liver transplantation recipients to return to ICU (Intensive Care Unit), which comprises the following steps of: incorporating factors possibly related to children liver transplantation recipients to return to ICU through literature research and expert consultation; general information of a child liver transplantation recipient, ICU hospitalization information and operation related information are obtained, and ICU intraday blood examination information and immunosuppressor drug concentration are transferred out; and screening variables related to the return ICU by adopting single-factor Cox regression. Carrying out regression on the single factor Cox to obtain plt; and introducing a variable of 0.05 into multi-factor Cox regression analysis, screening a model based on an AIC criterion, determining independent prediction factors influencing the outcome, and drawing a column graph. According to the method, a time-dependent ROC curve, a calibration curve, a consistency index and a clinical decision curve are adopted to evaluate the line graph, a bootstrap method is adopted for internal verification, and the process is repeated for 1000 times.

Description

Technical Field

[0001] This invention relates to the field of medical prediction, and in particular to a method for constructing a risk prediction model for pediatric living donor liver transplant recipients returning to the ICU. Background Technology

[0002] Pediatric liver transplantation is the only effective treatment for end-stage liver disease in children. Pediatric liver transplantation is a complex procedure, with postoperative conditions often being variable and rapidly progressing, typically requiring close monitoring in the intensive care unit (ICU). However, some pediatric liver transplant recipients require readmission to the ICU after initial discharge due to changes in their condition, a phenomenon known as ICU re-admission. ICU re-admission increases mortality, hospital stay, and hospitalization costs, making risk identification for pediatric liver transplant recipients crucial. Studies have shown that age, length of initial hospitalization, and respiratory rate at ICU discharge are risk factors for ICU re-admission in adult liver transplant recipients during the same hospital stay. However, the risk factors for pediatric liver transplant recipients differ significantly from those in adults, and the global sample size of pediatric liver transplant patients is much smaller than that of adults, posing a significant challenge to preventing pediatric liver transplant recipients from re-admitting to the ICU.

[0003] Therefore, there is an urgent need to establish a risk prediction model for pediatric living donor liver transplant recipients returning to the ICU, so as to identify risks in a timely manner, take early intervention measures, reduce the ICU return rate of pediatric living donor liver transplant recipients, and thus improve the clinical outcomes of children. Summary of the Invention

[0004] The main objective of this invention is to address the technical problem that the risk factors for pediatric liver transplant recipients to return to the ICU differ significantly from those for adults. A method for constructing a risk prediction model for pediatric living donor liver transplant recipients returning to the ICU includes the following steps: S1. Obtain clinical data of pediatric living donor liver transplant recipients who meet the preset inclusion and exclusion criteria, wherein the clinical data includes at least outcome information such as whether a return to the intensive care unit occurred; S2. Based on the outcome information, a propensity score matching method is used to match recipients who were readmitted to the intensive care unit with those who were not readmitted to the intensive care unit. S3. Select candidate predictor variables from the matched dataset; S4. Using whether or not a return to the intensive care unit occurs as the outcome variable, and the aforementioned candidate predictor variables as independent variables, a univariate Cox proportional hazards regression analysis is used to screen predictor variables related to the outcome variable. S5. Incorporate the predictive variables obtained from the single-factor Cox proportional hazards regression analysis into the multi-factor Cox proportional hazards regression model, and select the predictive factors for constructing the risk prediction model based on the Akaike information criterion. S6. Based on the aforementioned multifactor Cox proportional hazards regression model, establish a risk function; S7. Based on the risk function, construct a nomogram model to predict the risk of pediatric living donor liver transplant recipients returning to the intensive care unit.

[0005] In S2, the matching ratio of recipients who returned to the ICU to those who did not was 1:3, the matching method was nearest neighbor matching, and the clamp value was 0.2.

[0006] The nomogram model includes: The first line is a score scale, with a score range of 0-100; The second line shows the blood chloride concentration, ranging from 124 to 96. The third line is the heart rate, with a count range of 60-170. The fourth line represents diastolic blood pressure, ranging from 90 to 45. The fifth row shows the total protein concentration, ranging from 75 to 35. The sixth column shows uric acid, ranging from 50 to 450. The seventh line is the total score scale, with a score range of 0-260; The eighth line represents the probability of pediatric liver transplant recipients returning to the ICU 6978.5 minutes post-surgery, ranging from 0.1 to 0.8. The ninth line represents the probability of a pediatric liver transplant recipient returning to the ICU 87-135 minutes post-surgery, ranging from 0.1 to 0.9.

[0007] This invention also relates to a prediction system for a risk prediction model of pediatric living donor liver transplant recipients returning to the ICU, wherein the prediction system is a nomogram model, and the nomogram model includes: The first line is a score scale, with a score range of 0-100; The second line shows the blood chloride concentration, ranging from 124 to 96. The third line is the heart rate, with a count range of 60-170. The fourth line represents diastolic blood pressure, ranging from 90 to 45. The fifth row shows the total protein concentration, ranging from 75 to 35. The sixth column shows uric acid, ranging from 50 to 450. The seventh line is the total score scale, with a score range of 0-260; The eighth line represents the probability of pediatric liver transplant recipients returning to the ICU 6978.5 minutes post-surgery, ranging from 0.1 to 0.8. The ninth line represents the probability of a pediatric liver transplant recipient returning to the ICU 87-135 minutes post-surgery, ranging from 0.1 to 0.9.

[0008] The present invention has the following beneficial effects: By introducing propensity score matching to mitigate sample imbalance, and combining univariate and multivariate Cox regression analysis with the AIC criterion to screen statistically significant and clinically interpretable core predictive factors, a nomogram model based on real clinical data was constructed to achieve individualized and dynamic prediction of the risk of pediatric living donor liver transplant recipients returning to the ICU. This method avoids the bias caused by directly applying adult models, solves the problem of variable overfitting under small sample conditions, and provides an intuitive and operable clinical decision support tool in the form of nomograms, suitable for risk stratification management in multicenter, heterogeneous clinical settings. Attached Figure Description

[0009] Figure 1 A nomogram model for pediatric liver transplant recipients returning to the ICU.

[0010] Figure 2 This is the ROC curve for the prediction model.

[0011] Figure 3 This is the calibration curve for the prediction model.

[0012] Figure 4 This is the DCA curve for the prediction model.

[0013] Figure 5 Application of a nomogram model for predicting the return to the ICU for pediatric living donor liver transplant recipients. Detailed Implementation

[0014] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] This invention provides a method for constructing a risk prediction model for pediatric living donor liver transplant recipients returning to the ICU. The specific implementation process is as follows: First, a retrospective cohort study was conducted, including children who underwent living donor liver transplantation as the initial study subjects. Inclusion criteria included: age ≤18 years, first-time living donor liver transplantation, and survival upon transfer out of the ICU after surgery. Exclusion criteria included: death during postoperative ICU stay, readmission to the ICU for non-medical reasons (such as lack of family caregivers, bed turnover, etc.), missing data rate exceeding 30%, or serious postoperative complications preventing the completion of routine perioperative management procedures. Clinical data meeting the above criteria were extracted through electronic medical record systems, laboratory information systems, and intensive care unit databases, resulting in an original sample of 847 cases. Among these, 32 cases experienced ICU return events, while 815 cases did not, indicating a significant class imbalance.

[0016] To mitigate the bias caused by sample imbalance during model training, a propensity score matching method was used to balance the two groups. Specifically, whether or not a return to the ICU occurred was used as a binary outcome variable Y (Y=1 indicates occurrence, Y=0 indicates non-occurrence), and candidate predictor variables were used as covariates X to construct a logistic regression model: The propensity score P(Y=1|X) for each recipient was calculated. Then, a nearest neighbor matching method was used, with the caliper value set to 0.2 times the standard deviation of the propensity score. Recipients in the ICU return group were matched with 96 recipients in the non-ICU return group at a 1:3 ratio, resulting in a balanced dataset of 128 recipients. After matching, the distributional differences of each covariate between the two groups were tested using the standardized mean difference (SMD) test, ensuring that SMD < 0.1, indicating good matching and balanced baseline characteristics between groups.

[0017] In the matched dataset, candidate predictive variables are systematically extracted from the original clinical records, including but not limited to: age (months), sex (male / female), and body mass index (BMI, kg / m²). 2), ABO blood type, Rh blood type, donor-recipient ABO compatibility (compatible / incompatible), graft recipient body weight ratio (GRWR, %), primary disease type (biliary atresia, metabolic disease, acute liver failure, etc.), ICU stay duration (hours), mechanical ventilation duration (hours), specific date of ICU transfer (working day / non-working day) and time period (early shift / afternoon shift / night shift), return-to-ICU interval (minutes), vital signs upon ICU transfer (temperature ℃, heart rate bpm, respiratory rate bpm, systolic blood pressure mmHg, diastolic blood pressure mmHg, pulse oxygen saturation %), surgical procedure (left lobe of liver / right lobe of liver), intraoperative red blood cell transfusion volume (U), drainage volume in the 24 hours prior to ICU transfer (mL), blood test indicators on the day of ICU transfer (sodium, potassium, chloride, calcium, total protein, albumin, creatinine, uric acid, ALT, AST, INR, PT, etc.), and tacrolimus glutamate concentration (ng / mL). All variables were reviewed by clinical experts to confirm their availability and clinical relevance.

[0018] Subsequently, using the occurrence of re-ICU admission as the outcome variable, univariate Cox proportional hazards regression analysis was performed on each of the aforementioned candidate variables. Continuous variables were directly included in the model; categorical variables were coded using dummy variables. An independent Cox model was fitted for each variable, and its hazard ratio (HR), 95% confidence interval, and p-value were calculated. A significance threshold of α=0.10 was set, and variables with p<0.10 were retained for the next step of multivariate analysis. Preliminary screening yielded 15 potential predictive variables, including serum chloride concentration, total protein concentration, heart rate, diastolic blood pressure, uric acid, GRWR, duration of mechanical ventilation, drainage volume, tacrolimus concentration, and primary disease type.

[0019] Using the aforementioned 15 variables as input, a multivariate Cox proportional hazards regression model was constructed. To avoid overfitting and improve the model's generalization ability, a stepwise regression method combined with the Akaike information criterion (AIC) was used for variable selection. Specifically, starting from the full model, variables that caused the least increase in AIC were successively removed until removing any further variable would lead to an increase in AIC. After iterative optimization, five variables were ultimately retained to form the optimal model: serum chloride concentration (Cl, mmol / L), total protein concentration (TP, g / L), heart rate (HR, beats / min), diastolic blood pressure (DBP, mmHg), and uric acid (UA, μmol / L). This five-variable model had the smallest AIC value and satisfied the proportional hazards assumption of the Cox model (through the Schoenfeld residual test, p>0.05).

[0020] Based on extensive experimental data, the present invention uses the following expression for the risk function P(t): To further enable clinical visualization applications, a nomogram model is constructed based on the aforementioned risk function, such as... Figure 1 As shown in the figure, this nomogram contains nine parallel bars arranged from top to bottom, with vertical lines connecting the bars to map values. The first row is the base score bar 1, with a score range of 0-100, used to convert the values ​​of each variable into corresponding scores; the second row is the blood chloride concentration bar 2, with a concentration range of 96-124 mmol / L, covering the clinically common range of low to high chloride; the third row is the heart rate bar 3, with a count range of 60-170 beats / min, covering the range from normal to tachycardia after surgery in children; the fourth row is the diastolic blood pressure bar 4, with a range of 45-90 mmHg, reflecting postoperative hypotension to normal diastolic blood pressure levels; the fifth row is the total protein concentration bar 5, with a range of 35-75 g / L, corresponding to hypoproteinemia to normal protein levels; and the sixth row is the uric acid bar 6, with a range of 50-450. μmol / L, covering the common postoperative uric acid fluctuation range; the seventh row is the total score scale 7, with a score range of 0-260, obtained by summing the scores of each variable; the eighth row is the risk probability scale 8 at the first preset time point after surgery, set as the probability of returning to the ICU 5 days after surgery; the ninth row is the risk probability scale 9 at the second preset time point after surgery, set as the probability of returning to the ICU 60 days after surgery.

[0021] The nomogram is used as follows: Based on the actual test values ​​of the child on the day of ICU discharge, the clinician locates the corresponding values ​​on scales 2 through 6. For example, if a child's blood chloride is 102 mmol / L, find the position 102 on scale 2; if the heart rate is 120 beats / min, locate 120 on scale 3; if the diastolic blood pressure is 60 mmHg, locate 60 on scale 4; if the total protein is 50 g / L, locate 50 on scale 5; if the uric acid is 300 μmol / L, locate 300 on scale 6. Then, draw a perpendicular line upwards from each variable position, intersecting the baseline score scale 1, and read the corresponding score (e.g., Cl=102 corresponds to approximately 45, HR=120 corresponds to approximately 60, DBP=60 corresponds to approximately 30, TP=50 corresponds to approximately 50, UA=300 corresponds to approximately 40). Add all the scores together to get the total score (e.g., 45+60+30+50+40=225). Locate 225 on the total score scale 7, and then draw vertical lines downwards to intersect scale 8 and scale 9 respectively. This will give you the probability of the child returning to the ICU 5 days after surgery (median time to return to the ICU 6978.5 minutes ≈ 5 days) and 60 days after surgery (maximum observed time to return to the ICU 87135 minutes ≈ 60 days) (e.g., approximately 28% and 35% respectively).

[0022] The nomogram model was constructed strictly based on 128 matched real clinical cases. The scale ranges of all bars were determined according to the 5th to 95th percentiles of the actual observed values, ensuring coverage of the vast majority of clinical scenarios. The allocation of scores for each variable was achieved through linear transformation: score = (variable coefficient / absolute value of the largest coefficient) × 100, ensuring that the score is proportional to the risk contribution. The scales of risk probability bars 8 and 9 were generated by substituting the total score into the Cox model to calculate the cumulative risk function, and then estimating the calibration baseline risk using the Kaplan-Meier method. Internal model validation used the Bootstrap resampling method (repeated 1000 times). The results showed a C-index of 0.82 and a calibration curve close to the 45-degree diagonal, indicating that the model has good discrimination and calibration.

[0023] At the system implementation level, this invention also provides an electronic device, which includes one or more processors and a storage device. The storage device stores a computer program, which, when executed by the processor, sequentially calls a data acquisition module, a matching module, a variable screening module, a model building module, and an output module. The data acquisition module acquires clinical data meeting inclusion and exclusion criteria from a hospital information system interface in real time or in batches; the matching module calls a statistical software library (such as Python's sklearn or R's MatchIt package) to perform PSM; the variable screening module integrates the statsmodels or survival package to complete univariate and multivariate Cox regression and AIC screening; the model building module generates a risk function based on the regression coefficients and renders a nomogram; the output module outputs... Figure 1 The nine-line scale shown is presented on the clinical workstation via an interactive graphical interface, supporting mouse clicks for variable value input and automatic calculation of risk probabilities. The system is deployed on the hospital's intranet server and interfaces with the electronic medical record system via the HL7 protocol to ensure data security and real-time performance.

[0024] In summary, this invention constructs a nomogram prediction tool with a clear structure, concise variables, and strong visualization through a rigorous statistical modeling process and a clinical data-driven approach. Figure 1The spatial layout of scales 1 to 9 is fixed, arranged vertically in parallel, with consistent spacing between adjacent scales. All scales are of equal length, and the scale lines are perpendicular to the scale direction. Numerical labels are located on the outer side of the scales. The numerical direction of variables on scales 2-6 is determined according to the sign of their regression coefficients: for negative predictors (such as Cl, DBP, TP), the values ​​decrease from left to right; for positive predictors (such as HR, UA), the values ​​increase from left to right, ensuring that the score monotonically increases with risk. The total score scale 7 is located in the middle, serving as a hub connecting input variables and output probabilities. Risk probability scales 8 and 9 are located at the bottom, corresponding to two clinically relevant time points. Their probability values ​​are correlated with the total score through nonlinear interpolation, forming a complete risk mapping link. This design allows clinicians to quickly assess the risk of individual children returning to the ICU without complex calculations, providing an objective basis for early intervention and resource allocation.

[0025] The nomogram model achieves high-precision risk prediction because it eliminates inter-group confounding bias caused by the scarcity of events in the original data through propensity score matching during the variable screening stage. This allows for accurate estimation of the true effects of five variables: serum chloride, heart rate, diastolic blood pressure, total protein, and uric acid. Simultaneously, the numerical arrangement of scales 2-6 in the nomogram strictly follows the sign of the regression coefficients, ensuring that changes in score caused by variations in any variable value align with the actual risk change, thus guaranteeing a monotonic correlation between the total score and individual risk. Furthermore, the scale range is set based on the 5th to 95th percentiles of the matched data to avoid extreme values ​​interfering with clinical judgment. The total score scale 7 acts as a hub, compressing multidimensional variable information into a one-dimensional risk measure through a vertical connection mechanism. Finally, the bottom two probability scales 8 and 9 output quantitative risk with a clear time window, forming a complete mapping link from raw physiological indicators to clinical decision support.

[0026] Example: This invention focuses on 847 children who underwent living donor liver transplantation at Renji Hospital, affiliated with Shanghai Jiao Tong University School of Medicine, the world's largest liver transplant center, between January 1, 2019, and December 31, 2021. Return to the ICU refers to the need for further medical intervention during the same hospitalization due to complications or deterioration of the condition. It is closely related to increased medical costs, higher mortality rates, longer hospital stays, and adverse events.

[0027] Propensity score matching (PSM) involves selecting individuals from the control group whose propensity score is the same or similar to that of an individual in the treatment group and pairing them together. Commonly used matching methods include nearest neighbor matching, caliper matching, and global optimal matching.

[0028] The inclusion criteria were: (1) age < 18 years; (2) first admission to the ICU due to postoperative monitoring of liver transplantation; (3) survival when first transferred out of the ICU; (4) immunosuppressive regimen: tacrolimus + methylprednisolone; (5) the liver transplant donor was a living donor.

[0029] Exclusion criteria were: (1) death during ICU hospitalization after liver transplantation; (2) discharge from the ICU on the first admission without treatment; (3) not the first liver transplantation; (4) combined with other organ transplantation; (5) incomplete data that could not be analyzed.

[0030] To establish a multivariate predictive model, PSM was matched using the nearest neighbor matching method based on inclusion and exclusion criteria, with 1 case returning to the ICU and 3 cases not returning to the ICU. A total of 128 pediatric living donor liver transplant recipients were included in the retrospective analysis (32 cases returned to the ICU and 96 cases did not return to the ICU).

[0031] Statistical methods: Qualitative data were described using frequencies and percentages (%). Normally distributed quantitative data were described using mean (x) ± standard deviation; non-normally distributed quantitative data were described using median and interquartile range, denoted as M (P25, P75). Baseline covariate balance between matched groups was tested using t-tests or chi-square tests. Univariate Cox regression was used to screen variables associated with ICU readmission. Variables with p < 0.05 in univariate Cox regression were included in multivariate Cox regression analysis, and the model was screened based on the AIC criterion to identify independent predictors of outcome, and nomograms were plotted. Figure 1 Time-dependent ROC curves were used (the median follow-up time was selected as the evaluation time point). Figure 2 ), calibration curve ( Figure 3 ), C-index and DCA curve ( Figure 4 The performance of the predictive model was internally validated. The results showed an area under the ROC curve of 0.789 (95% CI: 0.685–0.909), indicating that the model has good discriminative ability regarding whether pediatric liver transplant recipients should return to the intensive care unit. The C-index of the predictive model was 0.778, and the calibration curve showed that the overall trend of the predicted probability and the actual observed probability was close to the ideal diagonal, indicating good model calibration. The DCA curve showed that, over a wide threshold probability range, the model's net benefit (red curve) was significantly higher than the two extreme strategies of "full intervention" (grey curve) and "no intervention" (black curve), indicating that the model has good clinical practical value within this threshold range and can bring more benefits to clinical decision-making. All statistical analyses were performed using R 4.2.2 software, and p < 0.05 was considered statistically significant.

[0032] Figure 1 This is a nomogram model for the return to the ICU of pediatric liver transplant recipients. In the nomogram, the first row is a score scale, ranging from 0 to 100; the second row is blood chloride concentration, ranging from 96 to 124; the third row is heart rate, ranging from 60 to 170; the fourth row is diastolic blood pressure, ranging from 45 to 90; the fifth row is total protein concentration, ranging from 35 to 75; the sixth row is uric acid, ranging from 50 to 450; the seventh row is the total score scale, ranging from 0 to 260; the eighth row is the risk probability of pediatric liver transplant recipients returning to the ICU 6978.5 minutes post-surgery, ranging from 0.1 to 0.8; and the ninth row is the risk probability of pediatric liver transplant recipients returning to the ICU 87135 minutes post-surgery, ranging from 0.1 to 0.9.

[0033] like Figure 5 As shown by the red arrows, one pediatric living donor liver transplant recipient had a Cl concentration of 110 mmol / L, a heart rate of 109 beats / min, a diastolic blood pressure of 55 mmHg, TP of 40 g / L, and uric acid of 210 mmol / L when transferred out of the ICU. The corresponding risk values ​​were 50, 40, 30, 45, and 20, respectively, with a total risk score of 185. The corresponding risks of returning to the ICU were 0.15 at 6978.5 minutes after transfer and 0.65 at 87135 minutes after transfer. That is, under these circumstances, the pediatric living donor liver transplant recipient had a 15% probability of returning to the ICU 6978.5 minutes (approximately 4.85 days) after transfer, and a 65% probability of returning to the ICU 87135 minutes (approximately 60.5 days) after transfer.

[0034] This invention also provides an electronic device, which can vary considerably due to differences in configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media may be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0035] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the electronic device structure in this embodiment does not constitute a limitation on the electronic device itself, and may include more or fewer components, or combinations of certain components, or different component arrangements.

[0036] The electronic device provided in this embodiment of the invention can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0037] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the aforementioned method.

[0038] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0039] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0040] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a risk prediction model for pediatric living donor liver transplant recipients returning to the ICU, characterized in that, Includes the following steps: S1. Obtain clinical data of pediatric living donor liver transplant recipients who meet the preset inclusion and exclusion criteria, wherein the clinical data includes at least outcome information such as whether a return to the intensive care unit occurred; S2. Based on the aforementioned outcome information, a propensity score matching method is used to match recipients who were readmitted to the intensive care unit with those who were not readmitted to the intensive care unit. S3. Select candidate predictor variables from the matched dataset; S4. Using whether or not a return to the intensive care unit occurs as the outcome variable, and the aforementioned candidate predictor variables as independent variables, a univariate Cox proportional hazards regression analysis is used to screen predictor variables related to the outcome variable. S5. Incorporate the predictive variables obtained from the single-factor Cox proportional hazards regression analysis into the multi-factor Cox proportional hazards regression model, and select the predictive factors for constructing the risk prediction model based on the Akaike information criterion. S6. Based on the aforementioned multifactor Cox proportional hazards regression model, establish a risk function; S7. Based on the risk function, construct a nomogram model to predict the risk of pediatric living donor liver transplant recipients returning to the intensive care unit.

2. The method as described in claim 1, characterized in that, In step S2, the matching ratio of recipients who returned to the intensive care unit to those who did not is 1:3, the matching method is nearest neighbor matching, and the clamp value is 0.

2.

3. The method as described in claim 1, characterized in that, Candidate predictive variables include age, sex, BMI, blood type, donor-recipient ABO blood type compatibility, GRWR, primary disease, ICU stay duration, mechanical ventilation duration, ICU discharge date and time, ICU return interval, vital signs at discharge, surgical procedure, intraoperative and postoperative RBC transfusions, drainage volume the day before ICU discharge, blood test information and immunosuppressant drug concentration on the day of ICU discharge.

4. The method as described in claim 1, characterized in that, The risk function includes regression coefficients corresponding to blood chloride concentration, total protein concentration, heart rate, diastolic blood pressure, and uric acid levels.

5. The method as described in claim 1, characterized in that, The expression for the risk function P(t) is: P(t) = 1 - [S0(t)] exp(-0.14852×血氯浓度+0.03366×心率-0.03590×舒张压-0.05370×总蛋白浓度+0.00521×尿酸) Where, S0(t): the baseline survival function; Blood chloride concentration, heart rate, diastolic blood pressure, total protein concentration, and uric acid each represent their respective linear predictive values; P(t): The probability of an event occurring at time t for an individual.

6. The method as described in claim 1, characterized in that, The nomogram model includes: The first line is a score scale, with a score range of 0-100; The second line shows the blood chloride concentration, ranging from 124 to 96. The third line is the heart rate, with a count range of 60-170. The fourth line represents diastolic blood pressure, ranging from 90 to 45. The fifth row shows the total protein concentration, ranging from 75 to 35. The sixth column shows uric acid, ranging from 50 to 450. The seventh line is the total score scale, with a score range of 0-260; The eighth line represents the probability of pediatric liver transplant recipients returning to the ICU 6978.5 minutes post-surgery, ranging from 0.1 to 0.

8. The ninth line represents the probability of a pediatric liver transplant recipient returning to the ICU 87-135 minutes post-surgery, ranging from 0.1 to 0.

9.

7. A pediatric living donor liver transplant recipient return-to-ICU risk prediction system comprising the method described in any one of claims 1-6, characterized in that, The prediction system is a nomogram model, which includes: The first line is a score scale, with a score range of 0-100; The second line shows the blood chloride concentration, ranging from 124 to 96. The third line is the heart rate, with a count range of 60-170. The fourth line represents diastolic blood pressure, ranging from 90 to 45. The fifth row shows the total protein concentration, ranging from 75 to 35. The sixth column shows uric acid, ranging from 50 to 450. The seventh line is the total score scale, with a score range of 0-260; The eighth line represents the probability of pediatric liver transplant recipients returning to the ICU 6978.5 minutes post-surgery, ranging from 0.1 to 0.

8. The ninth line represents the probability of a pediatric liver transplant recipient returning to the ICU 87-135 minutes post-surgery, ranging from 0.1 to 0.

9.

8. An electronic device, the electronic device comprising a memory and at least one processor, the memory storing instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1-6.

Images