Reply: Survival and course of lung function in the presence or absence of antifibrotic treatment in patients with idiopathic pulmonary fibrosis

Reply to S. Suissa and D. Assayag, and J. Borchardt,

We thank S. Suissa and D. Assayag, as well as J. Borchardt, for their interest in our study and for giving us the opportunity to respond to their comments regarding the potential impact of immortal time bias and lead time bias on the observed difference between idiopathic pulmonary fibrosis (IPF) patients treated with or without antifibrotic drugs under real-life conditions. As a key finding of the INSIGHTS-IPF Registry follow-up analysis, we had reported that survival was significantly higher in IPF patients receiving antifibrotic therapy when compared with propensity-matched IPF patients not receiving such drugs [1].

At the time of enrolment in our study, 266 patients were on antifibrotic therapy (either pirfenidone or nintedanib) and 32 additional patients newly started antifibrotic therapy at some point during follow-up. Consequently, we reported the patient characteristics at INSIGHTS-IPF enrolment for patients who were ever treated with antifibrotic therapy (n=298) and for those who were never treated with antifibrotic therapy (n=290). The mean time since IPF diagnosis was similar in patients who were treated with antifibrotic therapy (2.0±3.0 years) and patients who were not treated with antifibrotic therapy (2.1±3.9 years). Importantly, the observation time started for each patient at registry enrolment as described in the data analysis section. The observation period was divided into intervals with and without antifibrotic therapy based on the exact start and stop dates of antifibrotic treatment. At least two treatment periods were assigned to patients who started antifibrotic therapy during follow-up (n=32); namely, a treatment period without antifibrotic therapy, and another with antifibrotic therapy after treatment initiation. The episode before the start of antifibrotic therapy was included in the mortality and other outcome analyses of the “no antifibrotics” group, in order to avoid the proposed “immortal time bias”. Clearly, we should have expanded a bit more on this important methodological detail in our paper. Lung function data and 6-min walk test results were analysed by weighted linear mixed models to account for the possibility of more than one treatment episode for a single patient (additional cluster variable), and to account for the longitudinal study design, which was based on observed values, as explained in the paper. Mortality was analysed by Cox model with time-dependent exposure. We thank S. Suissa and D. Assayag for giving us the opportunity to provide these clarifications to the readers.

Another challenge in the analysis of treatment data in cohort studies is the risk of bias by indication. This is the risk that the exposure is associated with the outcome, i.e. antifibrotic therapy is initiated in patients who have a more severe course of the disease and may be at higher risk for mortality, or, alternatively, antifibrotic therapy may be withheld from patients with very advanced disease. Therefore, a propensity score was calculated and all analyses were weighted with an inverse probability weight based on the propensity score. Patients receiving antifibrotic treatment were on average 1.1 years older than controls at treatment initiation, as highlighted by J. Borchardt, but the difference was not statistically significant (p=0.148). Moreover, age was included as a predictor in the propensity score model in order to balance the age distribution between the two groups as we described in the methods section.

In summary, we were fully aware of the risk of several (typical) types of biases which might have compromised our analysis due to the observational nature of the data, and we discussed these biases in the limitations section of our article, including the immortal time bias (“healthy survivor bias” in our article). As suggested by S. Suissa and D. Assayag, we applied a proper time-dependent method for data analysis to mitigate the risk of immortal time bias, along with other methods to avoid other types of bias, and therefore are confident that our results did not exaggerate the observed survival difference between treated and untreated patients.

Shareable PDF