Modelling survival data of thalassaemia patients in Pakistan

Fatimid Foundation with its centers serves as a charitable organization for millions of blood disease carriers in Pakistan. This retrospective survival study is an analysis of the Thalassaemia patients registered in Multan centre who are followed up to nine years to assess the gender risk of death.

Data on 120 patients of Thalassaemia during 1994-2002 was analyzed. The standard Kaplan-Meier and Nelson-Aalen procedures were used to compare the survival function of the male and female patients of Thalassaemia. The statistical significance was also assessed using log rank test. The Cox Proportional Hazards (PH) model using a forward selection procedure was used to identify the potential factors associated with the increase risk of death.

By ignoring the censoring, the average survival time of males and females were calculated to be 1308.1 and 1574.7 days respectively. The average hazard rate for the female group was 0.00033 and for the male group it was 0.00061. The median survival time was estimated from the Kaplan-Meier survival curves (Males: 1400 days, Females: 1785 days). Under the censoring mechanism, the mean survival time for males was calculated to be 1465.82 days with standard error of 146 days, where as for females it was calculated to be 1855.26 with standard error of 151.82 days. When considering death as event of interest Under the Cox modelling approach, we found a male patient of Thalassaemia was 1.484 times more likely to die when compared to his female counterpart. When considering the iron over load as the event of interest, we found a male patient of Thalassaemia was 1.753 fold more likely to reach excessive Haemochromatosis level compared to female patient.

The analysis reinforces that female group prognosis is better than the male group. Further, our analysis of real survival data set shows that the high status of Haemochromatosis level is associated with the increased risk of death in patients with Thalassaemia with other possible confounders.