In light of global population aging and the increasing prevalence of Rheumatoid Arthritis (RA) with age, strategies are needed to address this public health challenge. Machine learning (ML) may play a vital role in early identification of RA, allowing an early start of treatment, thereby reducing costs. This study aims first to identify potential variables related to RA, and second to explore and evaluate the potential of ML to identify RA patients in people over 50 years.

We developed ML predictive models (lightGBM, logistic regression, k nearest neighbor, naive Bayes, random forrest, and XGBoost) using patient-reported outcomes collected from the SHARE database (7th and 9th wave).

Difficulties in daily life such as stooping and pulling are risk factors for RA. Lifestyle activities participation is negatively associated with RA. ML models performed differently with the lightGBM model achieving the highest AUC (0.748, 95 % CI: 0.739-0.758), and logistic regression and lightGBM showing the highest accuracy at 0.902. The sensitivity of naive Bayes was highest at 0.442. Significant differences were observed in the Hosmer-Lemeshow test (P < 0.05).

The predictive models based on patient-reported outcome measures achieved fair performance with limited potential to early identify RA patients. Lifestyle activities and difficulties in daily life were associated with risk of RA and should be considered in anamnesis.

To develop and internally validate a new severity score to more accurately assess the clinical severity forms of acute gastroenteritis (AGE) in children from birth to age 5 years.

We included children consulting for AGE in the emergency department of the University Hospital of Nantes (March 2017-June 2019). We developed and evaluated a new predictive score (GASTROVIM score) using the classification and regression trees. We compared its diagnostic performance with the two existing scores: the Vesikari score and clinical dehydration scale (CDS). A clinical expert a posteriori evaluated children's medical records to determine the severity form of AGE as the gold standard.

Of the 200 children included, 129 (64.5%) had severe forms of AGE according to the GASTROVIM score (maximal number of liquid stools and vomiting per day, weight loss and CDS), with sensitivity 90.0% (95% CI: 83.5-94.6) and specificity 82.9% (72.0-90.8). The Vesikari score had similar sensitivity (97.3%) but lower specificity (17.0%) and the CDS had lower sensitivity (28.3%) and higher specificity (100%) than the GASTROVIM score.

The GASTROVIM score could discriminate severe forms of AGE with good diagnostic performance. Nevertheless, external validation in other populations and/or other countries is needed.

Despite advances in research and patient management, atherosclerosis and its dreaded acute and chronic sequelae continue to account for one out of three deaths globally. The vast majority of acute coronary syndromes (ACS) arise from either plaque rupture or erosion, but other mechanisms, including calcific nodules, embolism, spontaneous coronary artery dissection, coronary spasm, and microvascular dysfunction, can also cause ACS. This ACS heterogeneity necessitates a paradigm shift in its management that extends beyond the binary interpretation of electrocardiographic and biomarker data. Indeed, given the evolution in the global risk factor profile, the increasing importance of previously underappreciated mechanisms, the evolving appreciation of sex-specific disease characteristics, and the advent of rapidly evolving technologies, a precision medicine approach is warranted. This review provides an update of the mechanisms of ACS, delineates the role of previously underappreciated contributors, discusses sex-specific differences, and explores novel tools for contemporary and personalized management of patients with ACS. Beyond mechanistic insights, it examines evolving imaging techniques, biomarkers, and regression- and machine learning-based approaches for the diagnosis (e.g. CoDE-ACS, MI3) and prognosis (e.g. PRAISE, GRACE, SEX-SHOCK scores) of ACS, along with their implications for future ACS management. A more individualized approach to patients with ACS is advocated, emphasizing the need for innovative studies on emerging technologies, including artificial intelligence, which may collectively facilitate clinical decision-making within a more mechanistic framework, thereby personalizing patient care and potentially improving long-term outcomes.

BackgroundDementia is a major public health challenge, yet existing prediction models often overlook sleep-related symptoms, despite their known links to cognitive decline.ObjectiveTo develop and validate a four-year Dementia Risk Score (DRS) incorporating self-reported sleep-related symptoms with demographic and clinical factors to predict all-cause dementia, including Alzheimer's disease.MethodsData from 3082 Korean adults aged 60-79 years were analyzed. Predictors were selected using LASSO regression and included in a multivariate logistic regression model. A point-based scoring system, the DRS, was constructed from the model coefficients. Internal validation was conducted using bootstrapping and a separate dataset.ResultsThe DRS achieved robust predictive performance, with AUC values of 0.824 in the training set and 0.826 in the validation set. Key predictors included sleep disturbance, use of sleep medications, daytime dysfunction, leg discomfort, and urge to move legs.ConclusionsThe DRS provides a practical, scalable tool for predicting dementia risk, supporting community-based screening and early intervention. External validation is needed to confirm its broader applicability.

Research-funding agencies are sometimes criticized for their ineffective review of grant proposals and for prioritizing competition for grants over project outcomes. Deficiencies in the review process for grants may limit the diversity of a program's applicant pool and restrict opportunities for publication. Our study asked what features of Peer Reviewed Orthopaedic Research Program (PRORP) grants and grant recipients were associated with successful grant outcomes, with success defined as publication of results within 5 years of receipt of funding.

Using data from all PRORP grants from 2009 to 2017, we built machine-learned predictive models to estimate publication within 5 years. Features included in the analysis were principal investigator characteristics (sex, degree, and institution type) and grant characteristics (research grant mechanism, primary and secondary research topics, and amount awarded). We evaluated model performance using calibration plots and then determined the models' discriminatory ability by estimating the area under the receiver operator curve and c-statistic. Then we used Brier scores to obtain an overall assessment of each model's accuracy. We ultimately selected 1 model for administrative use based on its performance measures.

The Bayesian generalized linear model performed best (area under the curve, 0.82 [95% CI, 0.68-0.95]; Brier score, 0.17 [95% CI, 0.10-0.23]) and therefore was selected. It showed administrative utility in decision curve analysis. Awardee features most strongly associated with publication were previous award winner (no), principal investigator specialty (anesthesiology), sex (male), degree (PhD), and institution type (university). Grant features most strongly associated with publication were primary topic (stem cell therapy and chemoprevention), secondary topic (biologics), research type (in vitro study and animal validation), award mechanism (translational research), award amount > median amount ($598,000), and award years 2010-2011.

This model can aid the PRORP in identifying awardees who will successfully publish, and funding agencies and policymakers likewise can use it to apportion grants in a manner that promotes diversity across the applicant pool.

This study aimed to develop and validate a clinical score for the prediction of critical care entrance in children with dengue.

We conducted a retrospective cohort study using admissions from January 2019 to August 2021, at Hospital Infantil Napoleón Franco Pareja, in Cartagena, Colombia. We included all children 18 years or younger, with a positive immunoglobulin M or nonstructural protein 1 laboratory test and admitted for follow-up at the emergency department. We selected variables retrospectively collected on emergency admission for feature selection. We assessed discrimination and calibration in the development dataset, using 1000 bootstrap replications for internal validation. Data from 2019 to 2020 were used for development and 2021 for temporal validation. We report the c -statistic for discrimination with 95% confidence intervals (CIs), as well as the calibration intercept and slope.

One thousand three hundred eighty-five patients were included for development and internal validation. In temporal validation with 519 additional patients, the c -statistic was 0.82 (95% CI: 0.77-0.87), with a calibration slope of 0.98 (95% CI: 0.77-1.18). We selected the 50 th percentile of the distribution of predicted probability of critical care entrance (5%) as a threshold value for increased alert at emergency admission, missing 10% of all cases that need to enter critical care (sensitivity of 90% with 95% CI of 82-95, and specificity of 48% with 95% CI of 41-50).

Our validated model can be useful to predict critical care entrance in children with dengue. We recommend the validation and potential recalibration of our score in other clinical settings.

Current protocols to triage life support use scores that are biased and inaccurate.

To determine if adding age to triage protocols used in disaster scenarios improves the identification of critically ill patients likely to survive.

Observational cohort study from March 1, 2020, to March 1, 2022, at 22 hospitals in three networks, divided into derivation (12 hospitals) and validation cohorts (ten hospitals). Participants were critically ill adults (90% COVID-19 positive) who would have needed life support during an overwhelming case surge. Life support was defined as vasoactive medications for shock, invasive or noninvasive mechanical ventilation, or oxygen therapy with Pao2/Fio2 less than 200.

The primary outcome was death in the intensive care unit. We fit logistic regression models using a modified Sequential Organ Failure Assessment (SOFA) score with and without age in the derivation cohort and assessed predictive performance in the validation cohort using area under the receiver operating characteristic curves (AUCs) and compared observed and predicted mortality.

The final analysis contained 7,660 patients with 16,711 life-support episodes. In the validation cohort, the AUC for age plus SOFA was significantly higher than the AUC for SOFA alone (0.66 vs. 0.54; p < 0.001). SOFA score substantially overpredicted mortality (13% predicted vs. 5% observed) for younger patients (< 40 yr) and underestimated mortality (14% predicted vs. 31% observed) for older patients (> 80 yr). In contrast, age plus SOFA had good calibration overall and across age groups. The addition of age improved but did not eliminate differences between observed and predicted mortality across racial-ethnic groups.

Age-inclusive triage better identifies ICU survivors than SOFA alone and is more equitable. Incorporating age into prioritization algorithms could save more lives in a crisis scenario.

To develop and validate a computed tomography-based radiomics nomogram for cancer-specific survival (CSS) prediction in curatively resected colorectal cancer (CRC), and its performance was compared with the American Joint Committee on Cancer (AJCC) staging and clinical-pathological models.

A total of 794 patients with curatively resected CRC from a prospective cancer registry program were included and randomly divided into the training (n = 556) and validation (n = 238) cohorts. A radiomics signature (RS) predicting CSS was constructed with a hybrid automatic machine learning strategy, and the prognostic value was assessed with Kaplan-Meier (KM) survival analysis. The performance of the established models was assessed by the discrimination, calibration, and clinical utility.

A 10-feature-based RS with independent prognostic value was developed. KM survival curves showed that high-risk patients defined by RS had a worse CSS than the low-risk patients (log-rank P<0.001). The radiomics nomogram integrating the RS and clinical-pathological factors had the optimal performance in predicting CSS in terms of Harrell's concordance index (0.803 [95% confidence interval: 0.761-0.845] for the primary cohort, 0.772 [95% confidence interval: 0.702-0.841] for the validation cohort), time-dependent receiver operating curves (time-ROC) (the area under the time-ROC curves [AUC] at three years were 84.06±2.86 and at five years were 86.35±2.19 in the primary cohort, the AUC at three years were 77.6±4.76, and at five years were 84±3.66 in the validation cohort), calibration curves and decision curve analysis, in comparison with the AJCC staging model, clinical-pathological model, and the RS alone.

The radiomics nomogram integrating the RS and clinical-pathological factors could be a valuable individualized predictor of the CSS for curatively resected CRC patients.

Periprosthetic joint infection (PJI) is an uncommon, but serious complication in total joint arthroplasty. Personalized risk prediction and risk factor management may allow better preoperative assessment and improved outcomes. We evaluated different data-driven approaches to develop surgery-specific PJI prediction models using large-scale data from the electronic health records (EHRs).

A large institutional arthroplasty registry was leveraged to collect data from 58,574 procedures of 41,844 patients who underwent at least one primary and/or revision hip and/or knee arthroplasty between 2000 and 2019. The registry dataset was augmented with additional clinical, procedural, and laboratory data from the EHRs for more than 100 potential predictor variables. The main outcome was PJI within the first year after surgery. We implemented both traditional and machine learning methods for model development (lasso regression, relaxed lasso regression, ridge regression, random forest, stepwise regression, extreme gradient boosting, neural network) and used 10-fold cross-validation to calculate measures of model performance in terms of discrimination (c-statistic) and calibration.

All models discriminated similarly in predicting PJI risk, with negligible differences of less than 0.08 between the best- and worst-performing models. The relaxed and fully relaxed lasso models using the Cox model structure outperformed the other models with concordances of 0.787 in primary hip arthroplasty and 0.722 in revision hip arthroplasty, with the number of predictors ranging from nine to 41. The concordances with the relaxed lasso models were 0.681 in primary and 0.699 in revision knee arthroplasty, with a higher number of predictors in the models. Predictors included in the models varied substantially across the four surgical groups.

The incorporation of additional data from the EHRs offers limited improvement in PJI risk stratification. Furthermore, improvement in PJI risk prediction was modest with the machine learning approaches and may not justify the added complexity.

Posttransplant diabetes mellitus (PTDM) is associated with significant morbidity and mortality in liver transplant recipients (LTRs). We used the Organ Procurement and Transplantation Network (OPTN) database to compare the incidence of developing PTDM across the United States and develop a risk prediction model for new-onset PTDM using OPTN region as well as donor-related, recipient-related, and transplant-related factors. All US adult, primary, deceased donor, LTRs between January 1, 2007, and December 31, 2016, with no prior history of diabetes noted , were identified. Kaplan-Meier estimators were used to calculate the cumulative incidence of PTDM, stratified by OPTN region. Multivariable Cox proportional hazards models were fitted to estimate hazards of PTDM in each OPTN region and build a risk prediction model, through backward selection. Cumulative incidence of PTDM at 1 year, 3 years, and 5 years after transplant was 12.0%, 16.1%, and 18.9%, respectively. Region 3, followed by regions 8, 2, and 9, had the highest adjusted hazards of developing PTDM. Inclusion of OPTN region in a risk prediction model for PTDM in LTRs (including recipient age, sex, race, education, insurance coverage, body mass index, primary liver disease, cold ischemia time, and donor history of diabetes) modestly improved performance (C-statistic = 0.60). In patients without pre-existing, confirmed diabetes mellitus, the incidence of PTDM in LTRs varied across OPTN regions, with the highest hazards in region 3, followed by regions 8, 2, and 9. The performance of a novel risk prediction model for PTDM in LTRs has improved performance with the inclusion of the OPTN region. Vigilance is recommended to centers in high-risk regions to identify PTDM and mitigate its development.

A detrimental association between radiation-induced lymphopenia (RIL) and oncologic outcomes in patients with esophageal cancer has been established. However, an optimal metric for RIL remains undefined but is important for the application of this knowledge in clinical decision-making and trial designs. The aim of this study was to find the optimal RIL metric discerning survival.

Patients with esophageal cancer treated with concurrent chemoradiation therapy (CRT; 2004-2022) were selected. Studied metrics included absolute lymphocyte counts (ALCs) and neutrophil counts-and calculated derivatives-at baseline and during CRT. Multivariable Cox regression models for progression-free survival (PFS) and overall survival (OS) were developed for each RIL metric. The optimal RIL metric was defined as the one in the model with the highest c-statistic.

Among 1339 included patients, 68% received photon-based and 32% proton-based CRT (median follow-up, 24.9 months). In multivariable analysis, the best-performing models included "ALC in week 3 of CRT" (corrected c-statistic 0.683 for PFS and 0.662 for OS). At an optimal threshold of <0.5 × 103/μL (ie, grade ≥3 RIL), ALC in week 3 was significantly associated with PFS (adjusted hazard ratio, 1.64; 95% CI, 1.27-2.13) and OS (adjusted hazard ratio, 1.56; 95% CI, 1.15-2.08), with 5-year PFS of 29% vs 40% and OS of 38% vs 51%, respectively.

Reaching grade ≥3 RIL in week 3 of CRT for esophageal cancer is the strongest RIL metric to distinguish survival outcomes. We suggest that this metric should be the target for lymphopenia-mitigating strategies and propose this metric to be included in future trials.

Lung adenocarcinoma (LUAD) has a heterogeneous prognosis and controversial postoperative treatment protocols. We aim to develop and validate a multimodal analysis framework that integrates CT images with H&E-stained whole-slide images (WSIs) to enhance risk stratification and predict adjuvant chemotherapy benefit in LUAD patients. We retrospectively collected data from 1039 resectable LUAD patients (stage I-III) across four centres, forming a training dataset (n = 303), two testing datasets (n = 197 and n = 228) for survival analysis, and a feature testing dataset (n = 311) for interpretability analysis. We extracted 487 tumour/peritumour radiomics features from CT images and 783 multiscale pathomics features from WSIs, characterising the shape of tumour (CT) and cancer nuclei (WSIs), as well as the intensity and texture of tumour/peritumour regions (CT) and tumour regions/epithelium/stroma (WSIs). A survival support vector machine (SVM) was employed to establish a radiopathomics signature using the optimal set of multimodal features, including 2 tumour radiomics features, 3 peritumour radiomics features, and 4 nuclei heterogeneity pathomics features. The radiopathomics signature outperformed both radiomics and pathomics signatures in predicting disease-free survival (DFS) (C-index: training dataset, 0.744 vs. 0.734 and 0.692; testing dataset 1, 0.719 vs. 0.701 and 0.638; testing dataset 2, 0.711 vs. 0.689 and 0.684), demonstrating greater robustness compared to the state-of-the-art deep learning integration approaches. It provided additional prognostic information beyond clinical risk factors (C-index of clinical plus radiopathomics vs. clinical models: training dataset, 0.763 vs. 0.676; testing dataset 1, 0.739 vs. 0.676; testing dataset 2, 0.711 vs. 0.699, p < 0.001). Compared to low-risk patients categorised by the radiopathomics signature, high-risk patients achieved comparable DFS when receiving adjuvant chemotherapy (training dataset, HR = 1.53, 95 % CI 0.85-2.73, p = 0.153; testing dataset 1 and 2, HR = 1.62, 95 % CI 0.92-2.85, p = 0.096), but had significantly worse DFS when only observed after surgery (training dataset, HR = 4.46, 95 % CI 2.82-7.05, p < 0.001; testing datasets 1 and 2, HR = 3.52, 95 % CI 2.26-5.49, p < 0.001), indicating the predictive value of the radiopathomics signature for adjuvant chemotherapy benefit (interaction p < 0.05). Further interpretability analysis revealed that the radiopathomics signature was associated with various prognostic/treatment-related biomarkers, including differentiation, immune phenotypes, and EGFR status. The multimodal integration framework offered a cost-effective approach for LUAD characterisation by leveraging complementary information from radiological and histopathological imaging. The radiopathomics signature demonstrated robust prognostic capabilities, providing valuable insights for postoperative treatment decisions.

The objective of this study was to validate the PredictAI models for predicting major bleeding (MB) in patients with active cancer and venous thromboembolism (VTE) with anticoagulant (ACO) therapy, within 6 months after primary VTE, using an independent cohort of patients from the TESEO database.

This study conducted an external validation of the PredictAI models using the international, prospective TESEO registry from July 2018 until October 2021. Data from 40 Spanish and Portuguese hospitals recruiting consecutive cases of cancer-associated thrombosis under anticoagulant treatment and without missing values regarding the model outcome or predictors were used. Patients with baseline MB or unknown MB status during follow-up were excluded for the validation analysis. Logistic regression (LR), decision tree (DT), and random forest (RF) approaches were used to validate the models.

Included patients from the TESEO cohort (2179 patients) had similar key demographics and clinical characteristics to the PredictAI cohort (21,227 patients). During the 6-month follow-up period, 10.9% (n = 2314) and 5.9% (n = 129) of patients experienced at least one MB event in the PredictAI and TESEO cohorts, respectively. Hemoglobin, metastasis, age, platelets, leukocytes, and serum creatinine were described as predictors for MB in PredictAI; the external validation results in TESEO showed statistical significance by LR and RF approaches, with ROC-AUC values of 0.59 and 0.56, respectively (both p < 0.05).

PredictAI models for predicting MB in anticoagulant-treated cancer patients within the first 6 months following VTE diagnosis have been externally validated. These models may be considered as a tool to guide objective decisions regarding the indication or extension of anticoagulant therapy in this population.

Aims/Background Idiopathic pulmonary fibrosis (IPF) is associated with an increased mortality risk. However, the factors that contribute to this risk remain unknown. This study aimed to systematically review existing predictive models for IPF-related mortality and to evaluate prognostic factors associated with patient outcomes. Methods A comprehensive literature search was conducted on PubMed, Cochrane Library, Web of Science, and Embase for studies on IPF mortality risk prediction models published between 1 January 1984 and 15 November 2024. Two independent reviewers screened, extracted, and cross-checked the data. The risk of bias and model applicability were also evaluated. Results A total of 17 risk prediction models were identified. The area under the receiver operating characteristic (ROC) curve (AUC) ranged from 0.728 to 0.907, while the model validation results ranged from 0.750 to 0.920. The concordance index (C-index) of 10 studies was more than 0.7, indicating good predictive performance. This study encompassed a total of 17 risk prediction models incorporating between 3 and 8 combined prognostic variables, with the most frequently included predictors being forced vital capacity as a percentage of the predicted value (FVC%pred), carbon monoxide diffusion capacity as a percentage of the predicted value (DLCO%pred), gender, age, six-minute walk test (6MWT) results, and dyspnea severity. Conclusion Current IPF mortality risk prediction models remain in an exploratory phase, with a generally high risk of bias. Furthermore, the lack of external validation in some models limits their generalizability. Future research should focus on improving the applicability of the model to enhance clinical application.

Develop a multivariable model to identify children with diabetic ketoacidosis (DKA) and/or hyperglycemic hyperosmolar state (HHS) at increased risk of adverse outcomes and apply it to analyze adverse outcomes during and after the COVID-19 pandemic.

Retrospective review of clinical data from 4565 admissions (4284 with DKA alone, 31 [0.7%] only HHS, 250 [5.4%] hyperosmolar DKA) to a large academic children's hospital from January 2010 to June 2023. Data from 2010-2019 (N = 3004) were used as a training dataset, and 2020-2021 (N = 903) and 2022-2023 (N = 658) data for validation. Death or intensive care unit stays > 48 hours comprised a composite "Adverse Outcome" group. Risks for this composite outcome were assessed using generalized estimating equations.

There were 47 admissions with Adverse Outcomes (1.5%) in 2010-2019, 46 (5.0%) in 2020-2021, and 16 (2.4%) in 2022-2023. Eight patients died (0.18%). Maximum serum glucose, initial pH, and diagnosis of type 2 diabetes most strongly predicted Adverse Outcomes. The proportion of patients with type 2 diabetes was highest in 2020-2021. A multivariable model incorporating these factors had excellent discrimination (area under receiver operator characteristic curve [AUC] of 0.948) for the composite outcome in the training dataset, and similar predictive power (AUC 0.960 and 0.873) in the 2020-2021 and 2022-2023 validation datasets, respectively. In the full dataset, AUC for death was 0.984.

Type 2 diabetes and severity of initial hyperglycemia and acidosis are independent risk factors for Adverse Outcomes and explain the higher frequency of Adverse Outcomes during the COVID-19 pandemic. Risks decreased in January 2022 to June 2023.

Cardiac rehabilitation (CR) has been proven to reduce mortality and morbidity in patients with cardiovascular disease. Machine learning (ML) techniques are increasingly used to predict healthcare outcomes in various fields of medicine including CR. This systemic review aims to perform critical appraisal of existing ML-based prognosis predictive model within CR and identify key research gaps in this area.

A systematic literature search was conducted in Scopus, PubMed, Web of Science, and Google Scholar from the inception of each database to January 28, 2024. The data extracted included clinical features, predicted outcomes, model development, and validation as well as model performance metrics. Included studies underwent quality assessments using the IJMEDI and Prediction Model Risk of Bias Assessment Tool checklist.

A total of 22 ML-based clinical models from 7 studies across multiple phases of CR were included. Most models were developed using smaller patient cohorts from 41 to 227, with one exception involving 2280 patients. The prediction objectives ranged from patient intention to initiate CR to graduate from outpatient CR along with interval physiological and psychological progression in CR. The best-performing ML models reported area under the receiver operating characteristics curve between 0.82 and 0.91, with sensitivity from 0.77 to 0.95, indicating good prediction capabilities. However, none of them underwent calibration or external validation. Most studies raised concerns about bias. Readiness of these models for implementation into practice is questionable. External validation of existing models and development of new models with robust methodology based on larger populations and targeting diverse clinical outcomes in CR are needed.

To effectively address Alzheimer's disease, it is crucial to understand its earliest manifestations, underlying mechanisms and early markers of progression. Recent findings of very early brain activation anomalies highlight their potential for early disease characterization and predicting future cognitive decline. Our objective was to evaluate the value of brain activation-both individually and in combination with structural and neuropsychological measures-for predicting cognitive change. The study included 105 individuals from the Consortium for the Early Identification of Alzheimer's Disease-Quebec cohort who exhibited subjective cognitive decline or mild cognitive impairment. Cognitive decline was assessed by calculating the slope of Montreal Cognitive Assessment scores using regression models across successive assessments, and individuals were characterized as either decliners or stable based on clinically reliable change. We evaluated cognitive decline predictions using unimodal models for each class of predictors and multimodal models that combined these predictors. Functional activation emerged as a strong predictor of cognitive change (R²=52.5%), with 87.6% accuracy and 98.7% specificity, performing comparably to structural and neuropsychological measures. Although the unimodal functional model exhibited high specificity, indicating that functional abnormalities frequently predict future decline, it had low sensitivity (60%), meaning that the absence of abnormalities does not rule out future decline. Multimodal models provided greater explanatory power than unimodal models and greater sensitivity than the functional model. These findings highlight the potential role of early brain activation anomalies in the early detection of future cognitive changes, offering valuable insights for clinicians and researchers in assessing cognitive decline risk and refining clinical trial criteria.

Coronary artery disease (CAD) is a leading cause of morbidity and mortality worldwide, and accurately predicting individual risk is critical for prevention. Here we aimed to integrate unmodifiable risk factors, such as age and genetics, with modifiable risk factors, such as clinical and biometric measurements, into a meta-prediction framework that produces actionable and personalized risk estimates. In the initial development of the model, ~2,000 predictive features were considered, including demographic data, lifestyle factors, physical measurements, laboratory tests, medication usage, diagnoses and genetics. To power our meta-prediction approach, we stratified the UK Biobank into two primary cohorts: first, a prevalent CAD cohort used to train predictive models for cross-sectional prediction at baseline and prospective estimation of contributing risk factor levels and diagnoses (baseline models) and, second, an incident CAD cohort using, in part, these baseline models as meta-features to train a final CAD incident risk prediction model. The resultant 10-year incident CAD risk model, composed of 15 derived meta-features with multiple embedded polygenic risk scores, achieves an area under the curve of 0.84. In an independent test cohort from the All of Us research program, this model achieved an area under the curve of 0.81 for predicting 10-year incident CAD risk, outperforming standard clinical scores and previously developed integrative models. Moreover, this framework enables the generation of individualized risk reduction profiles by quantifying the potential impact of standard clinical interventions. Notably, genetic risk influences the extent to which these interventions reduce overall CAD risk, allowing for tailored prevention strategies.

The extent to which high-sensitivity cardiac troponin can predict cardiovascular disease (CVD) is uncertain.

We aimed to quantify the potential advantage of adding information on cardiac troponins to conventional risk factors in the prevention of CVD.

We meta-analyzed individual-participant data from 15 cohorts, comprising 62,150 participants without prior CVD. We calculated HRs, measures of risk discrimination, and reclassification after adding cardiac troponin T (cTnT) or I (cTnI) to conventional risk factors. The primary outcome was first-onset CVD (ie, coronary heart disease or stroke). We then modeled the implications of initiating statin therapy using incidence rates from 2.1 million individuals from the United Kingdom.

Among participants with cTnT or cTnI measurements, 8,133 and 3,749 incident CVD events occurred during a median follow-up of 11.8 and 9.8 years, respectively. HRs for CVD per 1-SD higher concentration were 1.31 (95% CI: 1.25-1.37) for cTnT and 1.26 (95% CI: 1.19-1.33) for cTnI. Addition of cTnT or cTnI to conventional risk factors was associated with C-index increases of 0.015 (95% CI: 0.012-0.018) and 0.012 (95% CI: 0.009-0.015) and continuous net reclassification improvements of 6% and 5% in cases and 22% and 17% in noncases. One additional CVD event would be prevented for every 408 and 473 individuals screened based on statin therapy in those whose CVD risk is reclassified from intermediate to high risk after cTnT or cTnI measurement, respectively.

Measurement of cardiac troponin results in a modest improvement in the prediction of first-onset CVD that may translate into population health benefits if used at scale.

Rib fractures compromise approximately 40% of all fractures in the United States. Despite their prevalence, the relationship between rib fractures, solid organ injuries, and immune responses remains poorly understood. This exploratory study investigates the immunological profile associated with pulmonary and renal complications in rib fracture patients using data from our Surgical Critical Care Initiative Clinical Data Repository. The aim is to correlate distinct cytokine/chemokine profiles with high-energy rib fracture patterns, such as first rib fracture, and associated complications, potentially providing predictive biomarkers for patient outcomes.

Clinical and demographic data on patients with rib fractures were extracted from Surgical Critical Care Initiative Clinical Data Repository. Patients were categorized based on the presence or absence of complications. A comprehensive panel of 46 inflammation and tissue repair biomarkers was measured using the Meso Scale Discovery platforms. Principal component analysis was used to reduce the dimensionality of the cytokine data. Statistical and machine learning models assessed the association between biomarker patterns, rib fracture localization, and complications. Logistic regression models with high discriminative performance were developed for first rib fractures (high energy transfer), lung injury, and pneumonia.

Among 150 rib fracture patients, 73 had complications. Cytokine analysis revealed two distinct clusters: Cluster 1, associated with pro-inflammatory responses and tissue repair, and Cluster 2, linked with anti-inflammatory responses, angiogenesis, and immunometabolism. Predictive models demonstrated strong validity (area under the curve, >0.90) and identified key variables such as the cytokine IL2Ra, significantly associated with acute kidney injury, acute lung injury, and pulmonary complications post-rib fractures, particularly first rib fractures.

IL2Ra release is significantly correlated with high-energy transfer injuries like first rib fractures, indicating a bidirectional relationship between these fractures and the immune response. Furthermore, a hierarchical relationship exists among clinical complications, with kidney and lung injuries frequently preceding pneumonia. These findings underscore the potential utility of integrating immunological markers into clinical decision-support frameworks for personalized therapeutic interventions.

Prognostic; Level III.

To assess the ability of clinical inflammatory models to predict tumor regression grade (TRG) in response to neoadjuvant chemoradiotherapy (NCRT) and survival in patients with locally advanced rectal cancer (LARC).

We retrospectively analyzed 161 patients with LARC who underwent NCRT followed by total mesorectal excision at Beijing Hospital between May 2007 and March 2022. By using logistic and Cox regression analyses, we developed prediction models for TRG in response to NCRT and overall survival (OS), respectively.

Multivariable logistic regression analysis indicated that variations in neutrophil, lymphocyte, and monocyte counts and pre-NCRT (preneoadjuvant chemoradiotherapy) CA19 - 9 levels independently predicted TRG in response to NCRT (all P < 0.05). Multivariate Cox regression analysis revealed that clinical tumor (cT) stage, pre-NCRT platelet count, CA19 - 9 level, number of lymph node metastases, and TRG could independently predict OS (all P < 0.05). On the basis of these results, we developed models to predict TRG and OS, respectively. The final predictive model for predicting the response to NCRT had areas under the curve (AUCs) of 0.783 and 0.809 in the training and testing cohorts, respectively; for predicting the 5-year OS rate, the AUC rates were 0.842 and 0.930 in the training and test sets, respectively. The calibration and decision curves showed favorable performance in our prediction models.

We combined inflammatory markers with tumor characteristics and successfully developed clinical prediction models for TRG in response to NCRT and OS in patients with LARC. Our findings offer insights for optimizing treatment in patients with LARC.

Recent advances in deep learning have shown promising results in medical image analysis and segmentation. However, most brain MRI segmentation models are limited by the size of their data sets and/or the number of structures they can identify. This study evaluates the performance of 6 advanced deep learning models in segmenting 122 brain structures from T1-weighted MRI scans, aiming to identify the most effective model for clinical and research applications.

A total of 1510 T1-weighted MRIs were used to compare 6 deep learning models for the segmentation of 122 distinct gray matter structures: nnU-Net, SegResNet, SwinUNETR, UNETR, U-Mamba_BOT, and U-Mamba_ Enc. Each model was rigorously tested for accuracy by using the dice similarity coefficient (DSC) and the 95th percentile Hausdorff distance (HD95). Additionally, the volume of each structure was calculated and compared between normal controls (NCs) and patients with Alzheimer disease (AD).

U-Mamba_Bot achieved the highest performance with a median DSC of 0.9112 (interquartile range [IQR]: 0.8957, 0.9250). nnU-Net achieved a median DSC of 0.9027 [IQR: 0.8847, 0.9205], and had the highest HD95 of 1.392 [IQR: 1.174, 2.029]. The value of each HD95 (<3 mm) indicates its superior capability in capturing detailed brain structures accurately. Following segmentation, volume calculations were performed, and the resultant volumes of NCs and patients with AD were compared. The volume changes observed in 13 brain substructures were all consistent with those reported in existing literature, reinforcing the reliability of the segmentation outputs.

This study underscores the efficacy of U-Mamba_Bot as a robust tool for detailed brain structure segmentation in T1-weighted MRI scans. The congruence of our volumetric analysis with the literature further validates the potential of advanced deep learning models to enhance the understanding of neurodegenerative diseases such as AD. Future research should consider larger data sets to validate these findings further and explore the applicability of these models in other neurologic conditions.

A single-center retrospective cohort study.

To develop a predictive scoring system for bone union after conservative treatment of lumbar spondylolysis and assess its internal validity.

Lumbar spondylolysis, a common stress fracture in young athletes, is typically treated conservatively. Predicting bone union rates remains a challenge.

This study included patients aged 18 years or younger with lumbar spondylolysis undergoing conservative treatment. A multivariable logistic regression analysis was used to develop a scoring system containing 6 factors: sex, age, lesion level, main side stage of the lesion, contralateral side stage of the lesion, and spina bifida occulta. The predictive scoring system was internally validated from the receiver operating characteristic (ROC) curve using bootstrap methods.

The final analysis included 301 patients with 416 lesions, with an overall bone union rate of 80%. On multivariable analysis, the main and contralateral stages were identified as factors associated with bone union. The predictive scoring system was developed from the main side stage score (prelysis, early=0, progressive stage=1) and the contralateral side stage score (none=0, prelysis, early, progressive stage=1, terminal stage=3). The area under the curve was 0.855 (95% confidence interval, 0.811-0.896) for the ROC curve, showing good internal validity. The predicted bone union rates were generally consistent with the actual rates.

A simple predictive scoring system was developed for bone union after conservative treatment of lumbar spondylolysis, based on the stage of the lesion on the main and contralateral sides. The predicted bone union rate was ~90% for a total score of 0-1 and ≤30% for a score of 3-4. This system demonstrated good internal validity, suggesting its potential as a useful tool in clinical decision-making for the management of spondylolysis.

Artificial intelligence (AI)-based clinical decision support (CDS) has the potential to augment high-stakes clinical decisions in the emergency department (ED). However, its current usage and translation to implementation remains poorly understood. We asked: (1) What is the current landscape of AI-CDS for individual patient care in the ED? and (2) What phases of development have AI-CDS tools achieved?

We performed a scoping review of AI for prognostic, diagnostic, and treatment decisions regarding individual ED patient care. We searched five databases (MEDLINE, EMBASE, Cochrane Central, Scopus, Web of Science) and gray literature sources from January 1, 2010, to December 11, 2023. We adhered to guidelines from the Joanna Briggs Institute and PRISMA Extension for Scoping Reviews. We published our protocol on Open Science Framework (DOI 10.17605/OSF.IO/FDZ3Y).

Of 5168 unique records identified, we selected 605 studies for inclusion. The majority (369, 61%) were published in 2021-2023. The studies ranged over a variety of clinical applications, patient populations, and AI model types. Prognostic outcomes were most commonly assessed (270, 44.6%), followed by diagnostic (193, 31.9%) and disposition (115, 19%). Most studies remained in the earliest phase of preclinical development (572, 94.5%) with few advancing to later phases (33, 5.5%).

By thoroughly mapping the landscape of AI-CDS in the ED, we demonstrate a rapidly increasing volume of studies covering a breadth of clinical applications, yet few have achieved advanced phases of testing or implementation. A more granular understanding of the barriers and facilitators to implementing AI-CDS in the ED is needed.

Early warning system (EWS) scores are implemented on surgical wards to identify patients at high risk of postoperative clinical deterioration, but its predictive value in older patients is unclear. This study assessed the prognostic value of EWS scores to predict severe postoperative complications in older patients compared to younger patients.

This study utilized data from the TRACE study. EWS scores were routinely measured on postoperative days one (POD1) and three (POD3). The cohort was divided by age: < 70 years and ≥ 70 years. Performance measures of EWS scores on POD1 and POD3 were assessed to predict severe postoperative complications. Missed event rates (proportion of events not detected by the EWS threshold) and nonevent rates (proportion of EWS values above the threshold without an adverse event) were calculated.

Among 4866 patients, 39.3% were ≥ 70 years old. Severe complications occurred in 6.1% of older compared to 5.8% of younger patients (P = 0.658). EWS scores on POD1 and POD3 did not differ between age groups. For severe complications, EWS showed moderate discrimination in both older (POD1: C-statistic 0.65 (95%CI 0.59-0.70); POD3: 0.63 (95%CI 0.57-0.69)) and younger patients (POD1: 0.68 (95%CI 0.65-0.72); POD3: 0.65 (95%CI 0.61-0.70)). Overall, calibration was good. For EWS score ≥ 3, the missed event rate was at least 69% and nonevent rate 75%.

Predicted performance of the EWS score was moderate among older and younger patients. A limitation of the EWS score is the high rate of missed events and nonevents.

The objective of this study was to identify a female-specific prognostic biomarker for peripheral artery disease (PAD) and develop a prediction model for 2-year major adverse limb events (MALE). Patients with/without PAD were recruited (n=461). Plasma concentrations of 68 circulating proteins were measured and patients were followed for 2 years. The primary outcome was MALE (composite of vascular intervention, major amputation, or acute/chronic limb threatening ischemia). We trained a random forest model using: 1) clinical characteristics, 2) female-specific PAD biomarker, and 3) clinical characteristics and female-specific PAD biomarker. Galectin-9 was the only protein to be significantly elevated in females compared to males in the discovery/validation analyses. The random forest model achieved the following AUROC's: 0.72 (clinical features), 0.83 (Galectin-9), and 0.86 (clinical features + Galectin-9). We identified Galectin-9 as a female-specific PAD biomarker and developed an accurate prognostic model for 2-year MALE using a combination of clinical features and plasma Galectin-9 levels.

Despite apparently curative treatment, many patients with colorectal cancer develop subsequent metastatic disease. Current prognostic models are criticised because they are based on standard staging and omit novel biomarkers. Improved prognostication is an unmet need.

To improve prognostication for colorectal cancer by developing a baseline multivariable model of standard clinicopathological predictors, and to then improve prediction via addition of promising novel imaging, genetic and immunohistochemical biomarkers.

Prospective multicentre cohort.

Thirteen National Health Service hospitals.

Consecutive adult patients with colorectal cancer.

Collection of prespecified standard clinicopathological variables and more novel imaging, genetic and immunohistochemical biomarkers, followed by 3-year follow-up to identify postoperative metastasis.

Best multivariable prognostic model including perfusion computed tomography compared with tumour/node staging. Secondary outcomes: Additive benefit of perfusion computed tomography and other biomarkers to best baseline model comprising standard clinicopathological predictors; measurement variability between local and central review; biological relationships between perfusion computed tomography and pathology variables.

Between 2011 and 2016, 448 participants were recruited; 122 (27%) were withdrawn, leaving 326 (226 male, 100 female; mean ± standard deviation 66 ± 10.7 years); 183 (56%) had rectal cancer. Most cancers were locally advanced [≥ T3 stage, 227 (70%)]; 151 (46%) were node-positive (≥ N1 stage); 306 (94%) had surgery; 79 (24%) had neoadjuvant therapy. The resection margin was positive in 15 (5%); 93 (28%) had venous invasion; 125 (38%) had postoperative adjuvant chemotherapy; 81 (25%, 57 male) developed recurrent disease. Prediction of recurrent disease by the baseline clinicopathological time-to-event Weibull multivariable model (age, sex, tumour/node stage, tumour size and location, treatment, venous invasion) was superior to tumour/node staging: sensitivity: 0.57 (95% confidence interval 0.45 to 0.68), specificity 0.74 (95% confidence interval 0.68 to 0.79) versus sensitivity 0.56 (95% confidence interval 0.44 to 0.67), specificity 0.58 (95% confidence interval 0.51 to 0.64), respectively. Addition of perfusion computed tomography variables did not improve prediction significantly: c-statistic: 0.77 (95% confidence interval 0.71 to 0.83) versus 0.76 (95% confidence interval 0.70 to 0.82). Perfusion computed tomography parameters did not differ significantly between patients with and without recurrence (e.g. mean ± standard deviation blood flow of 60.3 ± 24.2 vs. 61.7 ± 34.2 ml/minute/100 ml). Furthermore, baseline model prediction was not improved significantly by the addition of any novel genetic or immunohistochemical biomarkers. We observed variation between local and central computed tomography measurements but neither improved model prediction significantly. We found no clear association between perfusion computed tomography variables and any immunohistochemical measurement or genetic expression.

The number of patients developing metastasis was lower than expected from historical data. Our findings should not be overinterpreted. While the baseline model was superior to tumour/node staging, any clinical utility needs definition in daily practice.

A prognostic model of standard clinicopathological variables outperformed tumour/node staging, but novel biomarkers did not improve prediction significantly. Biomarkers that appear promising in small single-centre studies may contribute nothing substantial to prognostication when evaluated rigorously.

It would be desirable for other researchers to externally evaluate the baseline model.

This trial is registered as ISRCTN95037515.

This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 09/22/49) and is published in full in Health Technology Assessment; Vol. 29, No. 8. See the NIHR Funding and Awards website for further award information.

Currently, there is a deficiency in a strong risk prediction framework for precisely evaluating the likelihood of severe postoperative complications in patients undergoing elective hepato-pancreato-biliary surgery subsequent to experiencing breakthrough infection of coronavirus disease 2019 (COVID-19). This study aimed to find factors predicting postoperative complications and construct an innovative nomogram to pinpoint patients who were susceptible to developing severe complications following breakthrough infection of COVID-19 after undergoing elective hepato-pancreato-biliary surgery.

This multicenter retrospective cohort study included consecutive patients who underwent elective hepato-pancreato-biliary surgeries between January 3 and April 1, 2023 from four hospitals in China. All of these patients had experienced breakthrough infection of COVID-19 prior to their surgeries. Additionally, two groups of patients without preoperative COVID-19 infection were included as comparative controls. Surgical complications were meticulously documented and evaluated using the comprehensive complication index (CCI), which ranged from 0 (uneventful course) to 100 (death). A CCI value of 20.9 was identified as the threshold for defining severe complications.

Among 2636 patients who were included in this study, 873 were included in the reference group I, 941 in the reference group II, 389 in the internal cohort, and 433 in the external validation cohort. Multivariate logistic regression analysis revealed that completing a full course of COVID-19 vaccination > 6 months before surgery, undergoing surgery within 4 weeks of diagnosis of COVID-19 breakthrough infection, operation duration of 4 h or longer, cancer-related surgery, and major surgical procedures were significantly linked to a CCI > 20.9. A nomogram model was constructed utilizing CCI > 20.9 in the training cohort [area under the curve (AUC): 0.919, 95% confidence interval (CI): 0.881-0.957], the internal validation cohort (AUC: 0.910, 95% CI: 0.847-0.973), and the external validation cohort (AUC: 0.841, 95% CI: 0.799-0.883). The calibration curve for the probability of CCI > 20.9 demonstrated good agreement between the predictions made by the nomogram and the actual observations.

The developed model holds significant potential in aiding clinicians with clinical decision-making and risk stratification for patients who have experienced breakthrough infection of COVID-19 prior to undergoing elective hepato-pancreato-biliary surgery.

Risk stratification of COVID-19 patients can support therapeutic decisions, planning and resource allocation in the hospital. In times of high incidence, a prognostic model based on data efficiently retrieved from one source can enable fast decision support.

A model was developed to identify patients at risk of developing severe COVID-19 within one month based on their age, sex and imaging features extracted from the thoracic computed tomography (CT). The model was trained on publicly available data from the Study of Thoracic CT in COVID-19 (STOIC) challenge and validated on unseen data from the same study and an external, multicentric dataset. The model, trained on data acquired before any variant of concern dominated, was assessed separately on data collected at later stages of the pandemic when the delta and omicron variants were most prevalent.

A logistic regression based on handcrafted features was found to perform on par with a direct deep learning approach, and the former was selected for simplicity. Volumetric and intensity-based features of lesions and healthy lung parenchyma proved most predictive, in addition to patient age and sex. The model reached an area under the curve of 0.78 on the challenge test set and 0.74 on the external test set. The performance did not drop for the subset acquired at a later stage of the pandemic.

A logistic regression utilizing features from thoracic CT and its metadata can provide rapid decision support by estimating short-term COVID-19 severity. Its stable performance underscores its potential for real-world clinical integration. By enabling rapid risk stratification using readily available imaging data, this approach can support early clinical decision-making, optimize resource allocation, and improve patient management, particularly during surges in COVID-19 cases. Furthermore, this study provides a foundation for future research on prognostic modelling in respiratory infections.

Some clinical algorithms incorporate an individual's race, ethnicity, or both as an input variable or predictor in determining diagnoses, prognoses, treatment plans, or risk assessments. Inappropriate use of race and ethnicity in clinical algorithms at the point of care may exacerbate health disparities and promote harmful practices of race-based medicine. Using database analysis primarily, we identified 42 risk calculators that use race and ethnicity as predictors, five laboratory test results with reference ranges that differed based on race and ethnicity, one therapy recommendation based on race and ethnicity, 15 medications with race- and ethnicity-based initiation and monitoring guidelines, and five medical devices with differential racial and ethnic performances. Information on these clinical algorithms is freely available at https://www.clinical-algorithms-with-race-and-ethnicity.org/ . This resource aims to raise awareness about the use of race and ethnicity in clinical algorithms and track progress toward eliminating their inappropriate use. The database is actively updated to include clinical algorithms that were missed and additional characteristics of these algorithms.

Objective.Machine learning's (MLs) ability to capture intricate patterns makes it vital in neural engineering research. With its increasing use, ensuring the validity and reproducibility of ML methods is critical. Unfortunately, this has not always been the case in practice, as there have been recent retractions across various scientific fields due to the misuse of ML methods and validation procedures. To address these concerns, we propose the first version of the neural engineering reproducibility and validity essentials for ML (NERVE-ML) checklist, a framework designed to promote the transparent, reproducible, and valid application of ML in neural engineering.Approach.We highlight some of the unique challenges of model validation in neural engineering, including the difficulties from limited subject numbers, repeated or non-independent samples, and high subject heterogeneity. Through detailed case studies, we demonstrate how different validation approaches can lead to divergent scientific conclusions, highlighting the importance of selecting appropriate procedures guided by the NERVE-ML checklist. Effectively addressing these challenges and properly scoping scientific conclusions will ensure that ML contributes to, rather than hinders, progress in neural engineering.Main results.Our case studies demonstrate that improper validation approaches can result in flawed studies or overclaimed scientific conclusions, complicating the scientific discourse. The NERVE-ML checklist effectively addresses these concerns by providing guidelines to ensure that ML approaches in neural engineering are reproducible and lead to valid scientific conclusions.Significance.By effectively addressing these challenges and properly scoping scientific conclusions guided by the NERVE-ML checklist, we aim to help pave the way for a future where ML reliably enhances the quality and impact of neural engineering research.

Undifferentiated arthritis (UA) often develops into rheumatoid arthritis (RA), but predicting disease progression from seronegative UA remains challenging because seronegative RA often does not meet the classification criteria. This study aims to build a machine learning (ML) model to predict the progression from seronegative UA to RA using clinical and laboratory parameters.

KURAMA cohort (training dataset) and ANSWER cohort (validation dataset) were utilized. Patients with seronegative UA were selected based on specific inclusion and exclusion criteria. Clinical and laboratory parameters, including demographic data, acute phase reactants, autoantibodies, and physical examination findings, were collected. Various ML models, including a Feedforward Neural Network (FNN), were developed and compared. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and other metrics. SHapley Additive exPlanations (SHAP) values were computed to interpret the importance of variables.

KURAMA cohort included 210 patients with seronegative UA, of whom 57 (27.1%) progressed to RA. The FNN model demonstrated the highest predictive performance with an AUC of 0.924 and a sensitivity of 80.7% in the training dataset. Validation with ANSWER cohort (140 patients; 32.1% progressed to RA) showed an AUC of 0.777, sensitivity of 77.8%. MMP-3 had the highest impact on the model.

The FNN model exhibited robust performance in predicting the progression of RA from seronegative UA and maintained substantial sensitivity in an independent validation cohort. This model using only clinical and laboratory parameters has potential for predicting RA progression in patients with seronegative UA.

Severe pneumonia has a poor prognosis and high mortality. Current severity scores such as Acute Physiology and Chronic Health Evaluation (APACHE-II) and Sequential Organ Failure Assessment (SOFA), have limited ability to help clinicians in classification and management decisions. The goal of this study was to analyse the clinical characteristics of severe pneumonia and develop a machine learning-based mortality-prediction model for patients with severe pneumonia.

Consecutive patients with severe pneumonia between 2013 and 2022 admitted to Beijing Chaoyang Hospital affiliated with Capital Medical University were included. In-hospital all-cause mortality was the outcome of this study. We performed a retrospective analysis of the cohort, stratifying patients into survival and non-survival groups, using mainstream machine learning algorithms (light gradient boosting machine, support vector classifier and random forest). We aimed to construct a mortality-prediction model for patients with severe pneumonia based on their accessible clinical and laboratory data. The discriminative ability was evaluated using the area under the receiver operating characteristic curve (AUC). The calibration curve was used to assess the fit goodness of the model, and decision curve analysis was performed to quantify clinical utility. By means of logistic regression, independent risk factors for death in severe pneumonia were figured out to provide an important basis for clinical decision-making.

A total of 875 patients were included in the development and validation cohorts, with the in-hospital mortality rate of 14.6%. The AUC of the model in the internal validation set was 0.8779 (95% CI, 0.738 to 0.974), showing a competitive discrimination ability that outperformed those of traditional clinical scoring systems, that is, APACHE-II, SOFA, CURB-65 (confusion, urea, respiratory rate, blood pressure, age ≥65 years), Pneumonia Severity Index. The calibration curve showed that the in-hospital mortality in severe pneumonia predicted by the model fit reasonably with the actual hospital mortality. In addition, the decision curve showed that the net clinical benefit was positive in both training and validation sets of hospitalised patients with severe pneumonia. Based on ensemble machine learning algorithms and logistic regression technique, the level of ferritin, lactic acid, blood urea nitrogen, creatine kinase, eosinophil and the requirement of vasopressors were identified as top independent predictors of in-hospital mortality with severe pneumonia.

A robust clinical model for predicting the risk of in-hospital mortality after severe pneumonia was successfully developed using machine learning techniques. The performance of this model demonstrates the effectiveness of these techniques in creating accurate predictive models, and the use of this model has the potential to greatly assist patients and clinical doctors in making well-informed decisions regarding patient care.