Journal of Statistics Applications & Probability

Author Country (or Countries)



In order to assess individual’s risk of a disease, semi-parametric proportional hazards Cox models are mostly used, while fewer studies have used parametric models. The aim of the present work was to compare semi-parametric and parametric statistical methods regarding their goodness of fit. To investigate the research hypothesis, characteristics of the 3042 participants of the ATTICA epidemiological study, were used; 2583 of them were found in the 10-year follow-up (2011-2012) and 317 (15.7%) developed a cardiovascular disease event. Three multivariable models, adjusted for the same set of risk factors were compared regarding their performance, using the Bayesian Information Criterion (BIC). All models were adjusted for: age, gender, physical activity level, Body Mass Index, smoking, hypertension, diabetes mellitus, hypercholesterolemia and adherence to the Mediterranean dietary pattern (assessed with MedDietScore). The semi-parametric Cox proportional hazard model had the worst performance as compared with the parametric survival models under the Weibull distribution. Between the two other parametric models, the Weibull model had the best performance (BIC =1386.488) as compared with the model with the exponential distribution (BIC =1729.724) (p for Harell?s C ?0.001). It appears that parametric models in relation to semi-parametric Cox proportional hazard models have better performance, while parametric model with Weibull distribution had the best performance among the parametric models.

Suggested Reviewers