Serious infection following anti-tumor necrosis factor alpha therapy in patients with rheumatoid arthritis: lessons from interpreting data from observational studies

Abstract

Objective: In a recent observational study, we found that the risk of serious infection following anti-tumor necrosis factor alpha (anti-TNFalpha) therapy in patients with rheumatoid arthritis (RA) was not importantly increased compared with the background risk in routinely treated RA patients with similar disease severity. Observational data sets are, however, subject to a number of important biases related to selection factors for the timing of starting and stopping therapy. Infection risk is also likely to vary with duration of therapy. This study was undertaken to examine the influences of these biases and of the method of analysis on the risk of infection.

Methods: We compared the risk of serious infection in 8,659 patients treated with anti-TNFalpha with that in 2,170 patients treated with traditional disease-modifying antirheumatic drugs (DMARDs) recruited to the British Society for Rheumatology Biologics Register. We applied a number of statistical models in which we varied the length of the followup period by using different definitions of the date of discontinuation of treatment and different lag periods of risk following drug cessation.

Results: When the at-risk period was defined as "receiving treatment", the adjusted incidence rate ratio comparing patients receiving anti-TNFalpha therapy with patients receiving DMARD therapy was 1.22 (95% confidence interval [95% CI] 0.88-1.69). Limiting followup to the first 90 days, however, revealed an adjusted incidence rate ratio of 4.6 (95% CI 1.8-11.9). Rates of infection were increased in the 90 days immediately following drug discontinuation and beyond, explained by selection factors for drug discontinuation.

Conclusion: These findings show that overall, the way in which UK rheumatologists select patients for starting and discontinuing anti-TNFalpha therapy explains our previous finding of no increase in risk. However, there may be important increases in true risk, notably early in the course of treatment, that would become more evident depending on the definition of at-risk period.

Figure 1

Patterns of constancy of risk of infection while receiving treatment (on drug) and after discontinuation of treatment (off drug). a, Increased risk at start of therapy. b, Constant risk with ongoing drug exposure. c, Increasing risk with cumulative drug exposure. d, Combination of the risk patterns shown in a–c. e, Ongoing constant risk for set lag window after discontinuation of treatment (drug stop). f, Linear decrease in risk back to baseline. g, Nonlinear decrease in risk back to baseline. h, Differing durations of risk windows, based on the half-life of each drug.

Figure 2

Influence of the definition of the at-risk period on an infection being attributed to therapy. a, Duration of treatment as the at-risk period. The at-risk window during which events can be attributed to the drug runs from 0 to 15 months. Serious infection A (at 14 months) is therefore attributed to the drug, but serious infections B (at 17 months) and C (at 23 months) are not. b, Duration of treatment plus a lag window of 3 months beyond discontinuation of the drug as the at-risk period. Serious infections A and B are attributed to the drug, but serious infection C is not. c, Start of treatment to the end of followup as the at-risk period. Serious infections A, B, and C are all attributed to the drug. d, Date of the first missed dose as the drug discontinuation date (stop date). If the stop date is incorrectly defined as the date of last dose given, rather than the first dose missed, event A will not be attributed to the drug.

Figure 3

Cumulative incidence of infections (a) and risk of serious infection (b), by drug. The y-axis shows the hazard, or risk, of serious infection. ETA = etanercept; INF = infliximab; ADA = adalimumab; DMARD = disease-modifying antirheumatic drug.