Abstract

This paper was inspired by the experience of the Authors research group composed by oncologist veterinarians and a biostatistician to evaluate treatments and prognostic factors with the aim to help veterinarians involved in longitudinal studies into evaluating and writing prognostic results.

Longitudinal studies are commonly analysed by techniques for survival data, taking into account for the time elapsed from the beginning of observation and the occurrence of an event related to treatment effect or disease course. The presence of incomplete follow-up information for some subjects requires specific descriptive and inferential statistical methods. Two literature datasets were analysed to show statistical models implementation techniques and to discuss statistical issues: I) A multicentre clinical trial on remission maintenance of children with acute Lymphoblastic leukaemia and II) A randomized clinical trial on advanced inoperable lung cancer. Data sets concerned studies on “humans”, nevertheless the peculiar data structure allowed to discuss some aspects which are common to survival analysis studies, regardless on subject’s characteristics. Log-rank test was used to compare survival curves for treatments and the relationship between Log-Rank test and univariate Cox model results was explained. As the evaluation of prognostic impact cannot be based only on p-values, the strength of the association between treatments and prognosis was estimated to take into account for the clinical relevance of results. On the second data set, beside of treatment, other clinical variables were available and a multivariate Cox model was applied. Model implementation was discussed concerning the coding of categorical variables and the relationship between continuous variables and model response. Suggested rules for the maximum number of covariates to be included in order to obtain reliable results were cited. Finally, the predictive ability of the model was discussed based on a measure of the area under ROC curve specific for survival data.

Keywords: Survival analysis; Prognosis; Interpretation of model results