Smoking-related bias of standardized mortality ratios in rheumatoid arthritis: A modeling study

Abstract

Objective

Standardized mortality ratios (SMRs) for rheumatoid arthritis (RA) are age- and sex-matched to the general population, but may be biased because smoking is more common in the RA group. This modeling study used national mortality data on smokers and non-smokers to estimate the effect on SMRs of the higher smoking prevalences typically found in RA.

Methods

Data from the United States National Health Interview Surveys 1999–2004 were used to create hypothetical cohorts with an age-sex composition typical of patients with RA (age 30 to 79; 70 % women). The reference cohort had the smoking prevalence of the general population (21.8 % current smokers). Additional cohorts were created that had higher proportions of smokers, approximating the prevalence of smoking commonly present in RA, with smoking relative risks of 1.25, 1.5, 1.75, and 2.0 compared to the reference cohort. SMRs were computed on 2000 replicate samples in which mortality over 10 years and 15 years was compared between the higher-smoking simulated RA cohorts and the reference cohort.

Results

The reference cohort had a prevalence of current smoking of 21.8 %. In a hypothetical RA cohort with a higher smoking prevalence, equal to a smoking relative risk of 2.0 compared to the general population, the median SMR for RA was 1.23 at 10 years and 1.17 at 15 years. At a smoking prevalence equivalent to a relative risk of 1.25, the median SMR for RA was 1.07 at 10 years and 1.04 at 15 years. Results were similar for SMRs based on relative risks that compared ever smokers to never smokers. Differences in smoking intensity between the hypothetical RA groups and reference cohorts had small effects on SMRs.

Conclusions

SMRs in RA may be inflated by even small increases in the prevalence of smoking relative to the general population. In these cases, an SMR benchmark of 1.0 to represent equal mortality outcomes would be too strict.