Mortality Prediction Model Research Articles

‘ALERT’, a machine learning predictive model to identify six-month mortality in patients with advanced cancer: Perception of the utility and acceptability among patients.

189 Background: Machine learning (ML) models can predict mortality and guide end-of-life (EOL) care for patients with advanced solid cancers. Early advanced care planning (ACP) leads to more goal-concordant care, yet many patients, especially from minoritized groups, do not engage in serious illness conversations (SIC). Reasons for delay include inability of oncologists to identify patients at high short-term mortality risk, and uncertainties about prognosis. While ML may improve prognosis accuracy, research on its use and impact on racial biases affecting SIC is limited. This study aims to evaluate the acceptability of deploying a ML model to predict 6-month mortality among patients with advanced solid cancers. Methods: In order to inform the implementation of ‘ALERT,’ a predictive mortality model into a clinical alert system, we conducted one-on-one semi-structured interviews with patients with advanced solid cancer receiving care at a single institution (N=26). Interviews were recorded, transcribed verbatim, coded, and themes were reviewed by a team of interdisciplinary investigators. Results: Benefits and challenges to guide the future implementation of ALERT were identified. Patient-perceived benefits were: 1) EOL Planning: preparing family members (e.g., financial, emotional); 2) Emotional Freedom: knowledge would give freedom to live out life more carefree due to limited time left; 3) Medical Autonomy: the treatment could be worse than the actual disease, if at any point it was not helping, knowing allows individuals to make the decision to stop. Challenges identified were: 1) Loss of Control: “when you don’t know [the outcome], you just keep moving, you just keep pushing”- knowing can result in a loss of control patients may desire for their course of treatment 2) Emotional Impact: knowledge of mortality could lead to grief among patients and families leading to distress around acceptability; 3) Faith: mortality prediction can be seen “playing God,” many patients believed when their time comes, it will happen and a ML tool cannot accurately predict when that time will be. Conclusions: Patients expressed equal balance of acceptability and resistance of ALERT for themselves and their families. This resulted in the difference of preferences regarding whether or not the knowledge this tool could provide would be helpful. This could lead to clinicians taking a patient-centered approach where the delivery of results using ALERT can be based on patient preference. The findings underscore the complexity of implementing ALERT, balancing significant benefits with notable challenges. These insights will guide the refinement of ALERT, ensuring it supports patient care while ensuring ALERT’s benefits are maximized and challenges are overcome.

Development and validation of predictive models for mortality of cases with COVID-19 (Omicron BA.5.2.48 and B.7.14): a retrospective study

ObjectivesWith the emergence of new COVID-19 variants (Omicron BA.5.2.48 and B.7.14), predicting the mortality of infected patients has become increasingly challenging due to the continuous mutation of the virus. Existing...

Mitral-specific cardiac damage score (m-CDS) predicts risk of death in functional mitral regurgitation: a study from the National Echo Database of Australia

AimsWe set out to explore associations between a ‘mitral-specific’ cardiac damage score (m-CDS) and survival outcomes in mitral regurgitation (MR) and compare the performance of the m-CDS and an ‘aortic-specific’...

Multicentre Validation of the RESECT-90 Prediction Model for 90-Day Mortality After Lung Resection

BackgroundThe RESECT-90 model was developed to predict 90-day mortality for patients undergoing lung resection but hasn't been externally validated. The aim of this study was to validate the RESECT-90 clinical prediction model using multicentre patient data from across the United Kingdom (UK). Materials and MethodsData from 12 UK thoracic surgery centers for patients undergoing lung resection between 2016 and 2020 with available 90-day mortality status were used to externally validate the RESECT-90 model. Measures of discrimination (area under the receiving operator characteristic curve [AUC]) and calibration (calibration slope, calibration intercept and flexible calibration plot) were assessed as measures of model performance. Model recalibration was also performed by updating the original model intercept and coefficients. ResultsA total of 12,241 patients were included. Overall 90-day mortality was 2.9% (n = 360). Acceptable model discrimination was demonstrated (AUC 0.74 [0.73, 0.75]). Calibration varied between centers with some evidence of overall model miscalibration (calibration slope 0.80 [0.66, 0.95] and calibration intercept 0.40 [0.29, 0.52]) despite acceptable appearances of the flexible calibration plot. The model was subsequently recalibrated, after which all measures of calibration indicated excellent performance. ConclusionsAfter external validation and recalibration using a large contemporary cohort of patients undergoing surgery in multiple geographical locations across the UK, the RESECT-90 model demonstrated satisfactory statistical performance for the prediction of 90-day mortality after lung resection. Whilst the recalibrated model will require ongoing validation, the results of this study suggest that routine use of the RESECT-90 model in UK thoracic surgery practice should be considered.

Development and Validation of a Prediction Model of Early Mortality for Conservative Management vs. Dialysis among US Veterans with Advanced CKD

Development and Validation of a Prediction Model of Early Mortality for Conservative Management vs. Dialysis among US Veterans with Advanced CKD

Development of a CVD mortality risk score using nutritional predictors: A risk prediction model in the Golestan Cohort Study

Background and AimWe aimed to develop a dietary score using prediction model method for evaluating the risk of cardiovascular disease (CVD) mortality and suggesting a simple and practical scoring system within the healthcare context. Method and ResultsA total of 43878 adult participants (aged 37-80 years) from the Golestan Cohort Study (GCS) were included in analysis. A random split of the subjects into the derivation (n=28930) and the validation sets (n=14948) was done. The Cox proportional hazard model was used to develop prediction model for the 8-year risk of CVD mortality. The model’s discrimination and calibration were assessed by C-statistic and calibration plot, respectively. To enhance clinical utility, we devised a point-based scoring system derived from our model. This prediction model was developed by nine predictors including age, physical activity level (MET minutes/week), waist-to-hip ratio, tea intake (cup/day), vegetable intake (gr/1000 kcal/day), white meat intake (gr/1000 kcal/day), salt intake (gr/1000 kcal/day), dairy intake (Cup/1000 kcal/day), and percentage of protein intake. The model had an acceptable discrimination in both derivation (C-statistic: 0.76, p<0.001) and validation (C- statistic: 0.77, p<0.001) samples. Also, the calibration of model in both derivation and validation datasets was 0.81. ConclusionThis is the first attempt to develop a risk prediction model of CVD mortality and the risk scoring system by the majority of nutritional predictors in a large cohort study. This nutritional risk assessment tool is suitable for motivating at-risk individuals to make lifestyle and dietary pattern changes to reduce future risk to prevent health problems.

Development and validation of the PHM-CPA model to predict in-hospital mortality for cirrhotic patients with acute kidney injury

BackgroundThe presence of acute kidney injury (AKI) significantly increases in-hospital mortality risk for cirrhotic patients. Early prognosis prediction for these patients is crucial. We aimed to develop and validate a machine learning model for in-hospital mortality prediction for cirrhotic patients with AKI. MethodsData from cirrhotic patients with AKI hospitalized at the First Affiliated Hospital of Zhejiang University between January 1, 2013, and December 31, 2020 were used to train and validate an extreme Gradient Boosting model to predict in-hospital mortality risk. The Boruta algorithm was used for variable selection. The optimal model was selected and named as PHM-CPA (Prediction of in-Hospital Mortality for Cirrhotic Patients with AKI). The PHM-CPA model was then externally validated in patients from eICU Collaborative Research Database (eICU-CRD) and Medical Information Mart for Intensive Care III dataset (MIMIC). The predictive performance of PHM-CPA model was compared with that of logistic regression (LR) model and 25 previously reported models. ResultsA total of 519 cirrhotic patients with AKI were enrolled in model training cohort, of whom 118 (23%) died during hospitalization. Fifteen variables from common laboratory tests were selected to develop the PHM-CPA model. The PHM-CPA model achieved an AUROC of 0.816 (95% CI, 0.763–0.861) in the internal validation cohort and 0.787 (95% CI, 0.745–0.830) in the external validation cohort. The PHM-CPA model consistently outperformed the LR model and 25 previously reported models. ConclusionWe developed and validated the PHM-CPA model, comprising readily available clinical variables, which demonstrated superior performance and calibration in predicting in-hospital mortality for cirrhotic patients with AKI.

Comparative analysis of Glasgow Coma Scale, quick Sepsis-related Organ Failure Assessment, base excess, and lactate for mortality prediction in critically ill emergency department patients

Abstract OBJECTIVES: It is crucial to promptly identify high-mortality patients in emergency departments and initiate their treatment as soon as possible. Although many parameters have been studied to select patients with high mortality, no comprehensive evaluation exists in previous literature on these parameters in critically ill patients, regardless of patient groups. The aim of this study is to evaluate the Glasgow Coma Scale (GCS), quick Sepsis-related Organ Failure Assessment (qSOFA), blood gas base excess (BE), and blood gas lactate in predicting mortality in critically ill patients admitted to the emergency department. METHODS: This prospective observational cohort study included adult patients with Emergency Severity Index 1–2 (critically ill) admitted to the emergency department. All patients were evaluated by the physician within 10 min, and blood gas samples were taken. The data collection forms recorded the patients’ GCS and qSOFA scores at the time of first evaluation by the physician. The qSOFA score assessment was performed in all patients with ESI levels 1 and 2, regardless of whether infective pathology was suspected. Blood gas BE and lactate values were also from laboratory test results. Patients or their relatives were contacted by phone at the end of the 1st month to obtain information about the clinical condition (survival or mortality). RESULTS: A total of 868 patients were included, with 163 deaths observed within 30 days. GCS score, qSOFA score, and lactate value were significant in predicting mortality within 30 days. While the BE value was significant for predicting 30-day mortality for values equal to or below the lower limit of −1.5 (P &lt; 0.001), it was not significant for values equal to or above the upper limit of +3 (P &gt; 0.05). The most successful prediction model for predicting 30-day mortality was found to be qSOFA with a cutoff value of ≥1. CONCLUSION: In emergency departments, each of the GCS, qSOFA scores, BE, and lactate values can be used independently as a practical mortality prediction model in critically ill patients. Among these four models, qSOFA is the most successful practical mortality prediction model in critically ill patients.

A machine learning algorithm-based risk prediction score for in-hospital/30-day mortality after adult cardiac surgery.

A study of the performance of in-hospital/30-day mortality risk prediction models using an alternative machine learning algorithm (XGBoost) in adults undergoing cardiac surgery. Retrospective analyses of prospectively routinely collected data on adult patients undergoing cardiac surgery in the UK from January 2012 to March 2019. Data were temporally split 70:30 into training and validation subsets. Independent mortality prediction models were created using sequential backward floating selection starting with 61 variables. Assessments of discrimination, calibration, and clinical utility of the resultant XGBoost model with 23 variables were then conducted. A total of 224,318 adults underwent cardiac surgery during the study period with a 2.76% (N = 6,100) mortality. In the testing cohort, there was good discrimination (area under the receiver operator curve 0.846, F1 0.277) and calibration (especially in high-risk patients). Decision curve analysis showed XGBoost-23 had a net benefit till a threshold probability of 60%. The most important variables were the type of operation, age, creatinine clearance, urgency of the procedure and the New York Heart Association score. Feature-selected XGBoost showed good discrimination, calibration and clinical benefit when predicting mortality post-cardiac surgery. Prospective external validation of a XGBoost-derived model performance is warranted.

Feature Selection Methods for ICU Mortality Prediction Model

Feature Selection Methods for ICU Mortality Prediction Model

Development and validation of a population-based risk algorithm for premature mortality in Canada: the Premature Mortality Population Risk Tool (PreMPoRT)

IntroductionTo develop and validate the Premature Mortality Population Risk Tool (PreMPoRT), a population-based risk algorithm that predicts the 5-year incidence of premature mortality among the Canadian adult population.MethodsRetrospective cohort analysis...

A comparison of some existing and novel methods for integrating historical models to improve estimation of coefficients in logistic regression

Abstract Model integration refers to the process of incorporating a fitted historical model into the estimation of a current study to increase statistical efficiency. Integration can be challenging when the current model includes new covariates, leading to potential model misspecification. We present and evaluate seven existing and novel model integration techniques, which employ both likelihood constraints and Bayesian informative priors. Using a simulation study of logistic regression, we quantify how efficiency—assessed by bias and variance—changes with the sample sizes of both historical and current studies and in response to violations to transportability assumptions. We also apply these methods to a case study in which the goal is to use novel predictors to update a risk prediction model for in-hospital mortality among pediatric extracorporeal membrane oxygenation patients. Our simulation study and case study suggest that (i) when historical sample size is small, accounting for this statistical uncertainty is more efficient; (ii) all methods lose efficiency when there exist differences between the historical and current data-generating mechanisms; (iii) additional shrinkage to zero can improve efficiency in higher-dimensional settings but at the cost of bias in estimation.

Interpretable mortality prediction model for ICU patients with pneumonia: using shapley additive explanation method

BackgroundPneumonia, a leading cause of morbidity and mortality worldwide, often necessitates Intensive Care Unit (ICU) admission. Accurate prediction of pneumonia mortality is crucial for tailored prevention and treatment plans. However, existing mortality prediction models face limited adoption in clinical practice due to their lack of interpretability.ObjectiveThis study aimed to develop an interpretable model for predicting pneumonia mortality in ICUs. Leveraging the Shapley Additive Explanation (SHAP) method, we sought to elucidate the Extreme Gradient Boosting (XGBoost) model and identify prognostic factors for pneumonia.MethodsConducted as a retrospective cohort study, we utilized electronic health records from the eICU-CRD (2014–2015) for all adult pneumonia patients. The first 24 h of each ICU admission records were considered, with 70% of the dataset allocated for model training and 30% for validation. The XGBoost model was employed, and performance was assessed using the area under the receiver operating characteristic curve (AUC). The SHAP method provided insights into the XGBoost model.ResultsAmong 10,962 pneumonia patients, in-hospital mortality was 16.33%. The XGBoost model demonstrated superior predictive performance (AUC: 0.778 ± 0.016)) compared to traditional scoring systems and other machine learning method, which achieved an improvement of 10% points. SHAP analysis identified Aspartate Aminotransferase (AST) as the most crucial predictor.ConclusionsInterpretable predictive models enhance mortality risk assessment for pneumonia patients in the ICU, fostering transparency. AST emerged as the foremost predictor, followed by patient age, albumin, BMI et al. These insights, rooted in strong correlations with mortality, facilitate improved clinical decision-making and resource allocation.

Prediction Model of In-Hospital Death for Acute Exacerbation of Chronic Obstructive Pulmonary Disease Patients Admitted to Intensive Care Unit: The PD-ICU Score

Introduction: Patients with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) admitted to intensive care unit (ICU) are exposed to poor clinical outcomes, and no specific prognostic models are available among this population. We aimed to develop and validate a risk score for prognosis prediction for these patients. Methods: This was a multicenter observation study. AECOPD patients admitted to ICU were included for model derivation from a prospective, multicenter cohort study. Logistic regression analysis was applied to identify independent predictors for in-hospital death and establish the prognostic risk score. The risk score was further validated and compared with DECAF, BAP-65, CURB-65, and APACHE II score in another multicenter cohort. Results: Five variables were identified as independent predictors for in-hospital death in APCOPD patients admitted to ICU, and a corresponding risk score (PD-ICU score) was established, which was composed of procalcitonin >0.5 μg/L, diastolic blood pressure <60 mm Hg, need for invasive mechanical ventilation, disturbance of consciousness, and blood urea nitrogen >7.2 mmol/L. Patients were classified into three risk categories according to the PD-ICU score. The in-hospital mortality of low-risk, intermediate-risk, and high-risk patients was 0.3%, 7.3%, and 27.9%, respectively. PD-ICU score displayed excellent discrimination ability with an area under the receiver-operating characteristic curve (AUC) of 0.815 in the derivation cohort and 0.754 in the validation cohort which outperformed other prognostic models. Conclusion: We derived and validated a simple and clinician-friendly prediction model (PD-ICU score) for in-hospital mortality among AECOPD patients admitted to ICU. With good performance and clinical practicability, this model may facilitate early risk stratification and optimal decision-making among these patients.

Machine learning models for mortality prediction in critically ill patients with acute pancreatitis-associated acute kidney injury.

The occurrence of acute kidney injury (AKI) was associated with an increased mortality rate among acute pancreatitis (AP) patients, indicating the importance of accurately predicting the mortality rate of critically ill patients with acute pancreatitis-associated acute kidney injury (AP-AKI) at an early stage. This study aimed to develop and validate machine learning-based predictive models for in-hospital mortality rate in critically ill patients with AP-AKI by comparing their performance with the traditional logistic regression (LR) model. This study used data from three clinical databases. The predictors were identified by the Recursive Feature Elimination algorithm. The LR and two machine learning models-random forest (RF) and eXtreme Gradient Boosting (XGBoost)-were developed using 10-fold cross-validation to predict in-hospital mortality rate in AP-AKI patients. A total of 1089 patients from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) and eICU Collaborative Research Database (eICU-CRD) were included in the training set and 176 patients from Xiangya Hospital were included in the external validation set. The in-hospital mortality rates of the training and external validation sets were 13.77% and 54.55%, respectively. Compared with the area under the curve (AUC) values of the LR model and the RF model, the AUC value of the XGBoost model {0.941 [95% confidence interval (CI) 0.931-0.952]} was significantly higher (both P<.001) and the XGBoost model had the smallest Brier score of 0.039 in the training set. In the external validation set, the performance of the XGBoost model was acceptable, with an AUC value of 0.724 (95% CI 0.648-0.800). However, it did not differ significantly from the LR and RF models. The XGBoost model was superior to the LR and RF models in terms of both the discrimination and calibration in the training set. Whether the findings can be generalized needs to be further validated.

Machine learning-based prognostic model for 30-day mortality prediction in Sepsis-3

BackgroundSepsis poses a critical threat to hospitalized patients, particularly those in the Intensive Care Unit (ICU). Rapid identification of Sepsis is crucial for improving survival rates. Machine learning techniques offer advantages over traditional methods for predicting outcomes. This study aimed to develop a prognostic model using a Stacking-based Meta-Classifier to predict 30-day mortality risks in Sepsis-3 patients from the MIMIC-III database.MethodsA cohort of 4,240 Sepsis-3 patients was analyzed, with 783 experiencing 30-day mortality and 3,457 surviving. Fifteen biomarkers were selected using feature ranking methods, including Extreme Gradient Boosting (XGBoost), Random Forest, and Extra Tree, and the Logistic Regression (LR) model was used to assess their individual predictability with a fivefold cross-validation approach for the validation of the prediction. The dataset was balanced using the SMOTE-TOMEK LINK technique, and a stacking-based meta-classifier was used for 30-day mortality prediction. The SHapley Additive explanations analysis was performed to explain the model’s prediction.ResultsUsing the LR classifier, the model achieved an area under the curve or AUC score of 0.99. A nomogram provided clinical insights into the biomarkers' significance. The stacked meta-learner, LR classifier exhibited the best performance with 95.52% accuracy, 95.79% precision, 95.52% recall, 93.65% specificity, and a 95.60% F1-score.ConclusionsIn conjunction with the nomogram, the proposed stacking classifier model effectively predicted 30-day mortality in Sepsis patients. This approach holds promise for early intervention and improved outcomes in treating Sepsis cases.

A Common Longitudinal Intensive Care Unit data Format (CLIF) to enable multi-institutional federated critical illness research.

Critical illness, or acute organ failure requiring life support, threatens over five million American lives annually. Electronic health record (EHR) data are a source of granular information that could generate crucial insights into the nature and optimal treatment of critical illness. However, data management, security, and standardization are barriers to large-scale critical illness EHR studies. A consortium of critical care physicians and data scientists from eight US healthcare systems developed the Common Longitudinal Intensive Care Unit (ICU) data Format (CLIF), an open-source database format that harmonizes a minimum set of ICU Data Elements for use in critical illness research. We created a pipeline to process adult ICU EHR data at each site. After development and iteration, we conducted two proof-of-concept studies with a federated research architecture: 1) an external validation of an in-hospital mortality prediction model for critically ill patients and 2) an assessment of 72-hour temperature trajectories and their association with mechanical ventilation and in-hospital mortality using group-based trajectory models. We converted longitudinal data from 94,356 critically ill patients treated in 2020-2021 (mean age 60.6 years [standard deviation 17.2], 30% Black, 7% Hispanic, 45% female) across 8 health systems and 33 hospitals into the CLIF format, The in-hospital mortality prediction model performed well in the health system where it was derived (0.81 AUC, 0.06 Brier score). Performance across CLIF consortium sites varied (AUCs: 0.74-0.83, Brier scores: 0.06-0.01), and demonstrated some degradation in predictive capability. Temperature trajectories were similar across health systems. Hypothermic and hyperthermic-slow-resolver patients consistently had the highest mortality. CLIF facilitates efficient, rigorous, and reproducible critical care research. Our federated case studies showcase CLIF's potential for disease sub-phenotyping and clinical decision-support evaluation. Future applications include pragmatic EHR-based trials, target trial emulations, foundational multi-modal AI models of critical illness, and real-time critical care quality dashboards.

Machine Learning-Based Predictive Model for Mortality in Female Breast Cancer Patients Considering Lifestyle Factors.

To construct a free and accurate breast cancer mortality prediction tool by incorporating lifestyle factors, aiming to assist healthcare professionals in making informed decisions. In this retrospective study, we utilized a ten-year follow-up dataset of female breast cancer patients from a major Chinese hospital and included 1,390 female breast cancer patients with a 7% (96) mortality rate. We employed six machine learning algorithms (ridge regression, k-nearest neighbors, neural network, random forest, support vector machine, and extreme gradient boosting) to construct a mortality prediction model for breast cancer. This model incorporated significant lifestyle factors, such as postsurgery sexual activity, use of totally implantable venous access ports, and prosthetic breast wear, which were identified as independent protective factors. Meanwhile, ten-fold cross-validation demonstrated the superiority of the random forest model (average AUC = 0.918; 1-year AUC = 0.914, 2-year AUC = 0.867, 3-year AUC = 0.883). External validation further supported the model's robustness (average AUC = 0.782; 1-year AUC = 0.809, 2-year AUC = 0.785, 3-year AUC = 0.893). Additionally, a free and user-friendly web tool was developed using the Shiny framework to facilitate easy access to the model. Our breast cancer mortality prediction model is free and accurate, providing healthcare professionals with valuable information to support their clinical decisions and potentially promoting healthier lifestyles for breast cancer patients.

Abstract P415: The Current Cardiac Disease in Pregnancy -II Risk Score as a Predictor of Pregnancy Morbidity is Not Affected by Maternal Obesity

Cardiovascular disease (CVD) is the most common cause of death in adults over 40 years of age. CVD remains a common cause of death in pregnancy and within the first-year post-partum, and Black women have higher cardiovascular risk during pregnancy than White women. There are several established mortality and morbidity prediction models for maternal cardiac complications including the Cardiac Disease in Pregnancy (CARPREG) risk score. CARPREG-II is the newest risk score which uses ten criteria to predict maternal cardiac complications. However, the only study to validate the CARPREG-II risk score had a large sample size with only 15% of patients with obesity (body mass index (BMI) &gt;30 kg/m2). Obesity and increased obesity class are known risks factors for pregnancy related morbidity and mortality. Thus, the validity of CARPREG-II risk score as a predictor of cardiovascular risks and pregnancy outcomes in women with cardiac disease and obesity remains unknown. Therefore, the goals of our study were to assess the validity of the CARPREG-II score in Black and White women as a predictor of risk for cardiovascular morbidity and mortality in obese pregnant women and determine if obesity influences CARPREG-II score and pregnancy morbidity and mortality. For this study, we completed an IRB-approved retrospective chart review of 232 pregnant adult Black and White women who received care at the University of Mississippi Medical Center (UMMC) from January 2013 - 2020. We completed a CARPREG-II risk score and compared morbidity in patients with normal BMI (18-24.9 kg/m2), overweight (25-29.9 kg/m2), obesity class I (30-34.9 kg/m2), obesity class II (35-39.9 kg/m2), and obesity class III (&gt;40 kg/m2). By race, the majority of White women were categorized as overweight; whereas the majority of Black women were categorized as obesity class II or III (p=0.0016). Although frequency of BMI category differed by race, CARPREG-II score did not differ as BMI increased in Black or White women, and obesity class did not influence CARPREG-II risk score. Our data indicate that Black women were more likely to have increased obesity, but obesity class did not influence CARPREG-II score. The lack of association between obesity class and CARPREG-II risk score indicate that the current CARPREG-II assessment is not an appropriate risk assessment tool for pregnant obese women and additional studies/tools are warranted to determine cardiovascular risks in pregnant obese women.

321.7: Construct and validate a mortality prediction model for severe pulmonary infection post-kidney transplantation.

321.7: Construct and validate a mortality prediction model for severe pulmonary infection post-kidney transplantation.