EVALUATING CLUSTER EFFECTS IN MALARIA SURVIVAL ANALYSIS WITH A SIMULATED EXTENDED COX MODEL

Abstract

Malaria remains a significant global health challenge, particularly in tropical regions. Accurate analysis of patient survival data is essential for understanding disease progression and evaluating the effectiveness of interventions. However, traditional survival analysis often overlooks clustering effects from factors like location, healthcare or family relationship. This study examines how unshared heterogeneity in treatment regimens and reporting time affect malaria patient survival analysis. A simulated dataset, following a Weibull distribution for typical malaria treatment duration (3-7days) was generated to assess the extended Cox model's ability to handle clustering. Three cluster sizes (20, 10, 5 observations) and varying total clusters (25, 50, 100) were used to mimic a 500-patient malaria dataset from Keffi General Hospital, Nigeria, considering shared treatment similarities within clusters. Cluster effects were introduced through a normally distributed random variable. Model 2, with 10 observations per cluster, performed best based on constant hazard, low AIC, and BIC. This suggests that 50 clusters of 10 observations each effectively capture the malaria data's underlying structure. The analysis of simulated covariates revealed that male patients had 15% higher risk of death compared to females. Additionally, younger patients (0-5years), patients with blood types A, B, or AB (particularly type A), and those with increasing body temperatures were identified as high-risk groups. This study underscores the importance of considering clustering effects in analyzing malaria time-to-event data, especially for clustered datasets; a sample size of 500, divided into 50 clusters of 10 patients each, seems optimal for analyzing real-world malaria datasets using the extended Cox model.