Abstract

Background: In survival analysis studies the interest is time taken to experience an event of interest. However, the probability of encountering the event of interest is commonly altered in studies where subjects experience an event other than that of interest. The standard survival time analysis methods, such as Kaplan-Meier method and the standard Cox model, fall short of differentiating different causes when competing risks are present. This is overcome by using statistical models that account for competing risks. The aim of the study was to compare and discuss estimates from nonparametric Cumulative Incidence Function, cause-specific hazards and subdistribution hazards in modeling time a patient suffering from infectious diseases spent in hospital until discharged. Death in hospital was identified as a competing risk. Methods: The nonparametric CIF was applied to the data to estimate the probability that a death or hospital discharge has occurred before a given day. In addition, the cause-specific hazards modeled the effect of HIV status, age and patient’s sex in relation to death or being discharged from hospital. The subdistribution hazards which does not assume independence between events was also used to compare results with the cause-specific hazards. Test of assumptions and model diagnostics followed. Results: Of 829 patients suffering from infectious diseases, 438 (52.4%) were females.452 (54.5%) patients were HIV positive, 116 (14.0%) were HIV negative and 261 (31.5%) had unknown HIV status. The nonparametric CIF, like the rest of models, showed that the HIV positive had a lower probability of being discharged in hospital than the HIV negative. The cause-specific hazard of hospital discharge for males was 0.73 (p<0.001). This meant that male patients were 27% less likely to be discharged from hospital compared to females. The subdistribution hazards estimates were close to those by cause-specific hazards. This suggested that the estimation of the hazards of encountering the event discharge was not affected much by the event death. Conclusions and Recommendation: It is important to follow up cause-specific hazards with subdistribution hazards as it provides a check for the effect competing events on the estimation of probability of occurrence of event of interest. The nonparametric CIF turned out a better estimator of patient’s cumulative incidence than the compliment of Kaplan-Meier.

More details

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

Download full document