Purpose: The aim was to identify a mathematical model that, when fitted with the survival time distribution of a Hodgkin's disease population, would provide a reliable estimate of expected survival at diagnosis for any new Hodgkin patient. This model would be based upon a multivariable selection of the best prognostic factors evaluable at diagnosis and its forecast could be of assistance in the choice of treatment.
Methods: The study sample consisted of the 5,023 patients whose basic clinical information was collected into the IDHD. These were people treated with standard protocols over the last two decades in 18 institutions. Several survival time distributions (exponential, Weibull, Gompertz, log-logistic and log-normal) were investigated to find the one that best fit the data and to relate its parameters to patient prognostic characteristics.
Results: The log-normal model provided the best fit for the data. The most statistically significant prognostic covariates were stage, age, histotype, B symptoms, serum albumin, sex and involved area distribution. Mediastinal, extranodal or bone marrow involvement, erythrocyte sedimentation rate, hemoglobin, serum alkaline phosphatase and lactate dehydrogenase did not add significant information. An equation containing these seven variables was derived to estimate median survival. Five distinct prognostic classes were identified by four cut-off values for this estimate.
Conclusions: Direct use of estimated median survival or allocating each patient into one of the five prognostic classes allows better tailoring of clinical strategies according to prognostic characteristics, more accurate patient stratification and evaluation of results in clinical trials and metaanalyses. Instructions are given for using this tool for both clinical and investigational purposes.