MODELLING ASSOCIATION BETWEEN COMPETING RISKS SURVIVAL AND LONGITUDINAL OUTCOMES IN RANDOMIZED CONTROLLED TRIAL FOR MALARIA INTERVENTIONAL STUDIES USING JOINT MODELS

Abstract

Biomedical studies may collect longitudinal and survival data in follow-up malaria studies. In randomized controlled trials in malaria interventional studies the longitudinal and survival data are analyzed separately (mixed-effect and Cox Models), yet the longitudinal outcomes may be important predictors in the survival outcomes. Standard methods for survival analysis, cannot be considered with such longitudinal outcomes. In such studies, survival process may also include multiple events (competing risks), implying that three blocks, survival, longitudinal and competing risks need to be considered in the analysis. In order to assess the association between the malaria longitudinal and the survival outcomes collected in biomedical studies, joint modelling framework was considered to combine the three blocks in the analysis. Joint models were also compared to separate models. Different survival outcomes observed were severe malaria (4.95%), withdrawal (10.89%) and censored (84.16 %). The time dependent haemoglobin level and parasite count were not associated with the risks of severe malaria and withdrawal in the extended-time dependent Cox model. The true longitudinal markers parasite counts and haemoglobin levels were associated with the risk of severe malaria (p <0.0001) and (p=0.029) respectively and had no effects on the risk of withdrawal in the joint models as these markers change with time. Generally the separate models were the best fit to the malaria dataset than the joint models due to lack of association between the survival outcomes and longitudinal outcomes in the cause specific time dependent hazard model.