Ischaemic Heart Disease Treatment Cost Analysis Using Generalised Linear Model (GLM)

Abstract

Ischaemic Heart Disease (IHD) is the most commonly diagnosed cardiovascular condition among patients with heart diseases. The global prevalence of IHD has risen significantly, from 112 million cases in 1990 to 254 million cases in 2021, imposing a substantial economic burden on healthcare systems worldwide. In Malaysia, the Ministry of Health (MOH) reported a total IHD treatment expenditure of RM389 million in 2017, equivalent to 38.4% of the total treatment cost of all cardiovascular diseases in public hospitals. Despite this burden, episode-level cost models derived from hospital administrative data, a prerequisite for specific resource planning, Diagnosis-Related Group (DRG) tariff setting and high-cost pathway identification, are still sparse in the Malaysian literature. This gap limits evidence-based healthcare planning and resource allocation. This study estimates the cost of single episode IHD treatment for patients admitted to Hospital Canselor Tuanku Muhriz (HCTM) using a Generalised Linear Model (GLM) framework. Using a retrospective dataset of 7,581 atherosclerotic cardiovascular disease cases, three regression models are compared: the Gamma GLM with a logarithm link function, Gaussian GLM with a logarithm link function, and log-transformed Multiple Linear Regression (MLR). Model selection is based on Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), deviance, Pearson chi-square statistics, and Quantile-Quantile (Q-Q) plots. The Gamma GLM demonstrated the best fit across all evaluation criteria, consistent with the theoretical appropriateness of the Gamma distribution for right-skewed, positive-valued cost data, where variance is proportional to the square of the mean. The key drivers of IHD treatment costs identified are the type of treatment received and the severity level of the patient. Patients undergoing Coronary Artery Bypass Grafting (CABG) incurred costs approximately 7.8 times higher than those managed without surgery, while severe patients incurred costs approximately 1.8 times higher than mild patients. Significant interactions between variables were found and led to a substantial differences in treatment costs. Most notably, the PCI x CABG interaction reduced the expected combined cost multiplier by 93.4% while PCI x Moderate severity level generated a significantly higher additional cost. This suggests that treatment type and patient severity level should be jointly modelled. The results provide an empirical evidence for policy interventions aimed at high-cost treatment pathways and patient risk classification.