Factors affecting the survival of prediabetic patients: comparison of Cox proportional hazards model and random survival forest method

• Published in: BMC medical informatics and decision making Vol. 24; no. 1; pp. 246 - 11
• Main Authors: Sharafi, Mehdi, Mohsenpour, Mohammad Ali, Afrashteh, Sima, Eftekhari, Mohammad Hassan, Dehghan, Azizallah, Farhadi, Akram, Jafarnezhad, Aboubakr, Zakeri, Abdoljabbar, Looha, Mehdi Azizmohammad
• Format: Journal Article
• Language: English
• Published: England BioMed Central Ltd 03-09-2024; BioMed Central; BMC
• Subjects: Adult; Afternoon sleep; Aged; Cohort analysis; Comparative analysis; Cox regression model; Demographic variables; Demographics; Demography; Depression; Diabetes; Diabetes mellitus; Diabetes mellitus (non-insulin dependent); Diabetes Mellitus, Type 2 - epidemiology; Diabetes Mellitus, Type 2 - mortality; Disease; Error analysis; Feature selection; Female; Glucose; Hazard identification; Health care; Humans; Investigations; Iran - epidemiology; Machine learning; Male; Mental depression; Middle Aged; Patient outcomes; Plasma; Prediabetic; Prediabetic state; Prediabetic State - epidemiology; Prognosis; Proportional Hazards Models; Prospective Studies; Questionnaires; Random Survival Forest; Risk Factors; Statistical models; Statistics; Survival; Survival analysis; Type 2 diabetes; Variables; Iran; Survival analysis; Prediabetic; Depression; Cox regression model; Non-alcoholic fatty liver disease; Afternoon sleep; Random Survival Forest
• ISSN: 1472-6947; 1472-6947
• DOI: 10.1186/s12911-024-02648-3

The worldwide prevalence of type 2 diabetes mellitus in adults is experiencing a rapid increase. This study aimed to identify the factors affecting the survival of prediabetic patients using a comparison of the Cox proportional hazards model (CPH) and the Random survival forest (RSF).
This prospective cohort study was performed on 746 prediabetics in southwest Iran. The demographic, lifestyle, and clinical data of the participants were recorded. The CPH and RSF models were used to determine the patients' survival. Furthermore, the concordance index (C-index) and time-dependent receiver operating characteristic (ROC) curve were employed to compare the performance of the Cox proportional hazards (CPH) model and the random survival forest (RSF) model.
The 5-year cumulative T2DM incidence was 12.73%. Based on the results of the CPH model, NAFLD (HR = 1.74, 95% CI: 1.06, 2.85), FBS (HR = 1.008, 95% CI: 1.005, 1.012) and increased abdominal fat (HR = 1.02, 95% CI: 1.01, 1.04) were directly associated with diabetes occurrence in prediabetic patients. The RSF model suggests that factors including FBS, waist circumference, depression, NAFLD, afternoon sleep, and female gender are the most important variables that predict diabetes. The C-index indicated that the RSF model has a higher percentage of agreement than the CPH model, and in the weighted Brier Score index, the RSF model had less error than the Kaplan-Meier and CPH model.
Our findings show that the incidence of diabetes was alarmingly high in Iran. The results suggested that several demographic and clinical factors are associated with diabetes occurrence in prediabetic patients. The high-risk population needs special measures for screening and care programs.