COMPLETE BLOOD COUNT-BASED PREDICTIVE MODELLING FOR DIABETES MELLITUS AMONG MALAYSIAN WOMEN: IMPLICATIONS FOR PUBLIC HEALTH SCREENING

Abstract

Diabetes mellitus is a major global public health concern, with a rising prevalence among women. Early prediction using routine clinical data, such as complete blood count (CBC) parameters, provides a cost-effective strategy for identifying individuals at risk. This study compared the effectiveness of classical binary logistic regression and Bayesian binary logistic regression in predicting diabetes mellitus among females using CBC parameters. Data from 537 female patients were analysed, incorporating CBC variables and diabetes status. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), Hosmer-Lemeshow goodness-of-fit test, posterior predictive checks, and sensitivity analyses. Bayesian diagnostics, including trace plots and credible intervals (Crls), were used to ensure model reliability. The Bayesian model outperformed the classical approach, demonstrating higher AUC, improved calibration, and more informative uncertainty estimation. Significant predictors included age, mean corpuscular volume (MCV), mean corpuscular haemoglobin concentration (MCHC), white blood cell count (WBC), haematocrit (HCT), platelet count, and haemoglobin (Hb). The Bayesian model produced stable posterior estimates and robust convergence, supporting its suitability for small or correlated datasets. These findings highlight the potential of CBC parameters as early indicators of diabetes risk among women and underscore the value of Bayesian modelling in informing gender-sensitive screening strategies and interventions.