Abstract

Background The Cox proportional hazards regression model has become the most used tool in the analysis of censored survival data. However, some features of the Cox model may cause problems for the analyst or an interpreter of the data. They include the restrictive assumption of proportional hazards for covariate effects, and “loss” (non-estimation) of the baseline hazard function induced by conditioning on event times. Compared with Cox Proportional Hazard model, parametric models are different in the way they exploit the information contained in the data. Parametric models specify how the hazard varies over time (hazard shape), which may provide insights into and guidance on how best to compare outcomes. This study compares the goodness of fit of Cox proportional hazard model and parametric survival models in modelling the household environmental and social economical determinants of under-five child mortality Methods The study used the 2010 Malawi Demographic Health survey data. The Cox Proportional hazard model was used and the proportional hazard (PH) assumption was assessed using both the graphical method and by adding time-dependent covariate in the Cox model. Good-ness of fit of the Cox PH model was also assessed using Cox Snell residuals. The parametric proportional hazard as well as accelerated failure time models was also used. The Weibull, lognormal, log-logistic, exponential and Gompertz model were fit and to find the most appropriate model, these models were compared using Akaike Information Criterion (AIC) and the goodness of fit for all the parametric models was assessed using Cox-Snell residuals. Results The Cox Proportional hazard model violated the assumption of proportionality and was not fitting the data well. The lognormal model was found to fit the data well and since this model is expressed in terms of accelerated failure time model, the violation of the proportion hazards assumption was overcome. Mother education, father education, house hold size, source of cooking fuel, type of toilet facility and wealth index are found to be significantly associated with child mortality. On the other hand, area of residence, source of water and access to electricity are found to be not significantly associated with child mortality. Conclusion The results obtained from the Cox PH model are not as effective as those obtained from the parametric AFT model since the PH assumption was found to be violated in the Cox PH model. Log-normal AFT model is found to be the most appropriate parametric model to be used in the analysis of child survival. Hence, researchers in child mortality using survival analysis can use the log-normal model as this will give them the more accurate and efficient results.

More details

School	: School of Natural and Applied Sciences
Issued Date	: 2012

Download full document