Joint Modelling of Longitudinal and Competing Risks Survival Data

Abstract

AbstractBiomedical studies may collect longitudinal and survival data in follow-up studies. In randomised controlled trials for malaria treatment, longitudinal parasite count and hemoglobin level and survival outcomes, time to fever resolution or time to parasite clearance, are recorded. The longitudinal and survival data are analysed separately, yet longitudinal outcomes may be important predictors in the survival process. Standard survival analysis methods cannot handle such longitudinal outcomes. In such studies, survival competing risks are possible; thus analysis should consider survival, longitudinal and competing risks. In joint modelling, options for modelling dependence are a key issue as well as choice of random effects distribution. The example used in this work was from sub-Saharan Africa.Joint modelling framework, mixed-effects models and Cox-specific models for analysis of longitudinal and survival data were applied to malaria dataset from Malawi Liverpool Wellcome Trust. Longitudinal outcomes considered were hemoglobin level and parasite count, while survival outcomes were time to treatment failure due to severe malaria and time to withdrawal (due to adverse effects and protocol violation).Different survival outcomes observed were severe malaria (4.95%) and withdrawal (10.89%). The longitudinal outcomes were not associated with the risks of severe malaria and withdrawal in the Cox model. The true hemoglobin level and age were associated with the risk of withdrawal (p = 0.0111) and (p = 0.0305), respectively, in the joint model, and the separate models were opted to fit the data.When an association between longitudinal and survival outcomes is of interest, joint models can be considered over separate methods. However, where there is no association, separate models for survival and longitudinal data analysis can be used.KeywordsSurvival modelsLongitudinal dataCompeting risksJoint modelsSevere malariaRandomised controlled trialsEfficacyCox-specific modelHemoglobin levelParasite countWithdrawalMixed-effectsTimeBiomedical studyEventCensoringParameterRandom effectsEstimationBICProfileCovariatesFollow-upRelative riskTreatmentPackageErrorFrameworkDataResearchPredictingAssociationPatients