Statistical Methods for Individual-Level Data in Cohort Mortality Studies of Rheumatic Diseases

Abstract

In the field of rheumatic diseases, prospective individual-level data on (prevalent) cohorts of patients over a long period of time are collected. One of the many uses of such data is to investigate mortality. Excess mortality, time trends (temporal variation) in mortality rates, life expectancy and life years lost of patients, and risk factors for mortality may all be of interest. In this article, we show how careful application of Poisson and Cox relative risk regression models can be used to tackle these questions, illustrating the methods with cohort data from patients with systemic lupus erythematosus and psoriatic arthritis. For external comparisons of a cohort with a standard (or reference) population, Poisson regression can be used to generalize the usual standardized mortality ratio (SMR) analyses. For assessing the pattern of (excess) mortality over time, unadjusted and adjusted “rolling average” SMRs are developed and shown to provide further insight concerning an observed decline in SMR over time for psoriatic arthritis patients. An extension of the traditional life expectancy and years of life lost calculations is also derived. For comparisons within a cohort, we demonstrate how Cox regression models can incorporate time measured on a variety of scales to allow the identification of a calendar time decline in mortality risk for lupus patients which is demonstrably independent of possible declines with disease duration and/or time in clinic and of other disease related explanatory variables.