Survival Models Based on Clinical and Demographic Covariates of Haemodialysis Patients
Ramkumar Thandiakkal Balan *
Department of Mathematics and Statistics, University of Dodoma, Kikuyu Avenue, Dodoma, Tanzania.
*Author to whom correspondence should be addressed.
Abstract
Background and Objective: Patients with chronic kidney disease (CKD) in Tanzania may avoid modern medical care because of limited knowledge and fear of treatment costs, which may increase mortality even among younger patients. Survival models for CKD and end-stage kidney disease (ESKD) patients based on clinical, demographic and laboratory covariates may help physicians intervene earlier and prolong patients' survival. This study aimed to develop parametric and non-parametric survival models for haemodialysis patients and to determine the covariates influencing mortality among dialysis patients. Although machine-learning models are increasingly used and can be efficient, they may not always be appropriately interpreted or implemented in some clinical situations.
Materials and Methods: This retrospective cohort study included 171 dialysis patients admitted to Muhimbili Hospital, Dar es Salaam, in 2015 and followed up to 2018. The hospital is a primary referral health centre for kidney diseases, and its nephrology department manages many ESRD patients, mainly through dialysis and transplantation. Basic prevalence was determined, and patient survival time was analysed using five parametric models. The Cox proportional hazards model and Kaplan-Meier model were used to identify significant survival patterns related to smoking, alcohol intake and HIV status.
Results: Of the 171 patients, 148 survived for 0-500 days, 20 survived for 501-1000 days, and only 3 patients survived for more than 1000 days. The factors associated with survival were sex, the number of dialysis sessions, blood transfusion and alcohol consumption. The lognormal distribution was the best parametric fit for the data, and the estimated average survival time was 268 days. The Cox proportional hazards model identified alcohol intake and the number of dialysis sessions as significant covariates. The Kaplan-Meier curve and mortality-rate curve showed significant differences according to smoking status, alcohol consumption and HIV infection, and these differences were supported by log-rank tests.
Conclusion: CKD and dialysis treatment were more common among males in Tanzania, and few patients survived beyond three years of treatment and follow-up. The number of dialysis sessions, lack of hygienic blood transfusion and alcohol intake were associated with mortality among CKD patients undergoing dialysis.
Keywords: Covariate detection, survival modelling, mortality factors, binary logistic regression, dietary habits, parametric survival models, chronic kidney disease (CKD), end-stage kidney disease (ESKD), haemodialysis