Pirfenidone in rheumatoid arthritis-associated interstitial lung disease – Authors' reply

Regarding the forced vital capacity (FVC) and percent predicted FVC (FVC%) secondary outcomes in our TRAIL1 study,1Solomon JJ Danoff SK Woodhead FA et al.Safety, tolerability, and efficacy of pirfenidone in patients with rheumatoid arthritis-associated interstitial lung disease: a randomised, double-blind, placebo-controlled, phase 2 study.Lancet Respir Med. 2023; 11: 87-96Summary Full Text Full Text PDF PubMed Google Scholar we wish to clarify the perceived inconsistency between the statistical model results and the raw data summaries, particularly for usual interstitial pneumonia. Although outcomes were secondary and subgroup analyses were not pre-planned, the statistical model estimated larger declines in placebo compared with pirfenidone for FVC and FVC%, a pattern not visually apparent in the raw data plots. Actual visit dates often differed from intended visit dates. The protocol allowed visits to deviate by a varying number of weeks. Allowable visit windows were expanded further during the COVID-19 pandemic. Nevertheless, 125 visits among 59 participants occurred outside the prespecified window and were included in all efficacy analyses. These deviations were not surprising given the follow-up duration and pandemic interruption. In the raw data plots, values were grouped and graphed as if they occurred on the intended visit day—subtitled, the visit label. In contrast, analyses in figures 2 and 3 use actual visit dates to generate estimated trends, described in the statistical analysis section, with markers drawn at intended visit weeks for convenience. With longitudinal data, underlying trends are often not visually apparent when evaluating averaged data at individual times. Such summaries ignore within-subject correlation, and longitudinal methods are needed for valid inference. If within-subject correlation is ignored, as when drawing conclusions from crude raw data plots, especially with small sample sizes in the usual interstitial pneumonia subgroups, then inferences are likely to be incorrect. It was deemed necessary a priori to perform the predictive modelling described in the statistical analysis section. Besides treatment assignment, the model included baseline value and predominant high-resolution CT pattern as covariates. Figure 2 shows estimated trends from the prespecified model. Usual interstitial pneumonia and non-usual interstitial pneumonia results in figure 3 were derived post hoc from the full model with a sample size of 123 participants, as described. In contrast, graphs of raw subgroup averages do not account for covariates. The raw data graphs show that the number of observations diminished at later visits, which is common in longitudinal trials. The 48 participants with usual interstitial pneumonia with week 52 data were probably not representative of the 81 participants with usual interstitial pneumonia who started the trial. It is a mistake to equate in importance the observed grouped averages at later visits with those at earlier visits. Figures 2 and 3 show increasing standard errors, signifying decreasing confidence as weeks progressed and available data diminished. Finally, all efficacy analyses were conducted according to intention-to-treat principles, including analyses that generated figures 2 and 3, and raw data summaries, as titled. IOR was awarded a grant from Genentech–Roche on behalf of the TRAIL1 network. All other authors declare no competing interests. Safety, tolerability, and efficacy of pirfenidone in patients with rheumatoid arthritis-associated interstitial lung disease: a randomised, double-blind, placebo-controlled, phase 2 studyDue to early termination of the study and underpowering, the results should be interpreted with caution. Despite not meeting the composite primary endpoint, pirfenidone slowed the rate of decline of FVC over time in patients with RA-ILD. Safety in patients with RA-ILD was similar to that seen in other pirfenidone trials. Full-Text PDF