Abstract

The famous Cox proportional hazard model is applied in most medical research studies that involve time to event data like mass oral azithromycin for the reduction of childhood mortality. Literature has shown that this model overestimates estimates in the presence of time varying covariates. This thesis compared the Cox proportional hazard model to Extended Cox model that account for time varying covariates. The models were applied to model the effect of age, weight, and treatment in relation to death as the outcome of interest. The study findings showed that the Cox PH model overestimated the effect of the covariates to the hazards of a participant dying as it did not take into consideration the presence of time dependent covariates in the data as compared to the extended Cox model. Kaplan Meir survival curves were plotted to compare survival in the two study arms (Placebo and Azithromycin drug groups). The hazard of death was associated with covariates age, weight, and treatment-received in both study arms. The results from the Cox model showed that the expected hazard is 13.56 (4.89, 37.64) times higher in a person who is one year older than another. For the variable weight, the expected hazard is 0.763(0.64, 0.90) times lower risk reduction in the drug group as compared to placebo group. For the variable treatment received, the expected hazard is 0.03 (0.02, 0.08) times lower risk reduction in the drug group as compared to placebo group for those treated than those who did not receive treatment. Whereas in the extended Cox model, the main model showed that the expected hazard is 3.74(0.53, 26.35) times higher in a person who is one year older than another, the expected hazard is 21.01(6.82, 64.72) times higher for those who did not receive drug than those who received drug (p=<0.001). The time varying covariates model showed that the expected hazard is 1.24(0.84, 1.83) times higher in a person who is one year older than another. The expected hazard is 1.85(1.48, 2.31) times higher in those who did not receive drug than those who received drug (p=<0.001). The Extended Cox model was a better model when studying data that involves time varying covariates to avoid reporting overestimated estimates.

More details

School	: School of Natural and Applied Sciences
Issued Date	: 2022

Download full document