Abstract
Many studies in the field of haemophilia and other coagulation deficiencies require analyses of bleeding frequencies. In haemophilia, the association of bleeding frequency with factor VIII (FVIII) activity levels is known from experience, but significant results are lacking. Bleeding frequencies in haemophilia are highly skewed count data, with large proportions of zeros. Both the skewness and the high amount of zeros pose a problem for standard (linear) modelling techniques. This study investigated the optimal analysing strategy for bleeding data by using the association of residual clotting factor level and number of joint bleeds in moderate and mild patients treated on demand as example. In total, 433 patients with moderate (27%) and mild (73%) haemophilia A treated on demand were included in this study. One year of self-reported data on joint bleed frequency and baseline clotting factor activity were analysed using Poisson, negative binomial, zero-inflated Poisson, and zero-inflated negative binomial distributions. Multivariate regression analysis using negative binomial distribution provided the optimum data analytical strategy. This model showed 18% reduction [Rate ratio (RR) 0.82; 95%confidence interval (CI) 0.77-0.86] of bleeding frequency with every IU dL(-1) increase in residual FVIII activity. The actual association is expected to be higher because of exclusion (30 out of 463 patients) of patients on prophylaxis (baseline FVIII levels 0.01-0.06 IU mL(-1)). The best way to analyse low frequency bleeding data is using a negative binomial distribution.