In this historical cohort study, clinically relevant outcomes were compared between patients with IPF newly treated with pirfenidone (n = 804) and nintedanib (n = 509), using the French national claims dataset. After adjustment on confounding factors, at one year, nintedanib use was associated with a higher all-cause mortality (HR, 1.8; 95% CI, 1.3–2.6), a higher risk of acute respiratory-related hospital admission (HR, 1.3; 95% CI, 1.0–1.7) and a lower risk of treatment discontinuation (HR, 0.7; 95% CI, 0.6–0.9) compared to pirfenidone use.
Comparisons were adjusted for several variables known to be associated with mortality in patients with IPF. Consistent with previous studies, age [2, 21, 22], Charlson comorbidity index [23], use of supplemental oxygen [24] and hospital admissions in the year prior to treatment initiation [25] were associated with increased risks of all-cause mortality. Number of hospital admissions in the year prior to treatment initiation, time between IPF diagnosis and treatment initiation, and use of supplemental oxygen were also associated with acute respiratory-related hospitalisations in the study period, similar to previous studies [2, 22, 24, 25]. Adjustment on variables made it possible to assess the influence of each confounding factors on outcomes. The use of propensity score method was not necessary in this study as there were enough events to be able to include all the variables in multivariable analyses.
As this study was based on claims data, lung-function parameters and their changes with time, which are predictors of outcome in patients with IPF, were not available. However, both drugs have the same indication, prescription and reimbursement modalities in France, and both study groups had similar characteristics including time to treatment initiation. It is therefore unlikely that the observed differences in outcome could be explained by differences in disease severity at the time of antifibrotic treatment initiation.
As prescription of antifibrotic medications is strictly controlled in France by the use of a dedicated prescription form for the sole indication of IPF, restriction of prescription to pulmonologists and requirement of diagnosis confirmation following multidisciplinary discussion, we are confident that the study population consisted of patients with a ‘true’ diagnosis of IPF. Moreover, the off-label prescription of antifibrotic drugs is expected to be very rare. Indeed, in the large international PASSPORT registry, only three of 1009 patients enrolled (and none for the subgroup from France) had a diagnosis other than IPF, despite the fact that this registry allowed the inclusion of subjects with other indications [26]. Therefore, although the diagnosis was not confirmed individually for each patient, we are confident that prescription of antifibrotic medications was limited to patients with IPF. An algorithm based on age and ICD-10 codes was nevertheless used to exclude subjects who may have had a differential diagnosis, especially connective tissue disease, sarcoidosis and pneumoconiosis. The median age in our population (73 years) and the large male-to-female predominance are consistent with a diagnosis of IPF in the vast majority of our patients.
To date, no head-to-head randomised controlled trials have been conducted to compare pirfenidone and nintedanib. Indirect comparisons (network meta-analyses) have compared the two treatments, generally showing no clear difference in outcome between the two drugs, although with some discrepancy [27–29] and methodological limitations. The network meta-analysis performed by Fleetwood et al. [29] was however the only one that could assess outcomes at 52 weeks after initiation in all treatment groups, suggesting a lower, but not significant all-cause mortality with pirfenidone compared to nintedanib (HR = 0.74; 95% CI, 0.27–1.95). In a previous analysis based on insurance claims data in the US, Dempsey et al. [30] found (in a study population of comparable size to ours) a decreased risk of all-cause mortality in patients receiving an antifibrotic drug, but no significant difference in all-cause mortality between drugs (nintedanib versus pirfenidone, HR for overall mortality, 1.14; 95% CI, 0.79–1.65; p = 0.471). The reasons for the discrepancy between the study by Dempsey and ours can only be speculated upon. Dempsey et al. [30] used data from OptumLabs Data Warehouse, which includes data from both Medicare and commercially insured subjects, e.g. presumably a group of patients with a smaller fraction of low-income or unemployed individuals, as compared to the French NHS data that cover the entire population. In addition, both drugs became available at the same time in the US, while in France nintedanib was reimbursed 30 months later than pirfenidone, potentially influencing physician choice of the medication. Although both pirfenidone and nintedanib were approved and reimbursed in the US for the same indication (i.e. treatment of IPF, irrespective of disease severity), co-payments may vary and influence treatment initiation, continuation or choice of medication. Of note, in the Dempsey study, the discontinuation rate was high (45.1%) but no information was given about between-drug differences in the rate of discontinuation.
The choice of initiating pirfenidone or nintedanib could, in theory, be related to patient characteristics, disease history, comorbidities, concomitant medications, physician experience and patient choice based on possible adverse events. The earlier availability of pirfenidone compared to nintedanib in France could have impacted how the drugs were prescribed during the study period. To limit the impact of known or unknown confounders, our study included only first-time users, and the models were adjusted by date of first prescription and time between the date of IPF diagnosis and treatment initiation. Of note, in our study, two patients received nintedanib in the context of a TUA (i.e. with potentially a more severe disease), but they made only 0.4% of the nintedanib sample, which therefore could not impact the results.
Overall, more than 15% of the patients stopped therapy early after treatment initiation (within three months) and 50% of patients stopped their treatment at one year. Comparative analysis of treatment discontinuation showed that female patients were more likely to interrupt therapy, in agreement with a previous study conducted in patients with chronic obstructive pulmonary disease [31]. The use of supplemental oxygen, a proxy of disease severity, was associated with a greater risk of treatment discontinuation, in contrast to other studies showing fewer interruptions with increasing disease severity [31]. Pirfenidone treatment was associated with a higher risk of drug discontinuation at 12 months than nintedanib. Treatment discontinuation could have been due to tolerability issues, which tend to occur within three months of treatment initiation [32], or to a switch from pirfenidone to nintedanib. However, results from this study represent on-treatment cumulative incidence up to one year and therefore are unlikely to be influenced by a switch in medication.
This study had some limitations mainly related to its observational design and data source. The ICD-10 code used to confirm the presence of pulmonary fibrosis in the study population (J84.1) is not specific to IPF, and ICD codes and age were mostly used to exclude other forms of fibrotic interstitial lung diseases among patients initiating an antifibrotic drug. Although imperfect, such an approach has already been used in the French healthcare dataset in a study that further validated the definition algorithm for acute respiratory-related hospitalisations [33]. Risk factors for outcomes were identified from proxies based on hospital admissions, visits, concomitant medications and hospital discharge codes; however, data on lung function and body mass index, both major determinants of IPF prognosis, were not available, as is often the case for claims data. While the effects of confounding factors were taken into account as thoroughly as possible, residual confounding effects may still be present due to the study design. Nonetheless, the consistency of our findings with regard to all-cause mortality and acute respiratory-related hospitalisations, along with identifying factors known to impact these outcomes, such as disease severity or prior hospital admissions, supports the robustness of our findings.
Among observational studies of antifibrotics, this study stands out because it was population-based, conducted with a dataset recording the health care consumption of almost all French citizens (i.e. a population of 60 + million inhabitants, without distinction of age, gender, ethnicity, residency, incomes, psychosocial status, etc., i.e. from a large country with universal healthcare coverage and a single unified healthcare system), and because it records individual and comprehensive data for the population since 2006.
In summary, this comparative analysis of outcomes following initiation of antifibrotic therapy for IPF showed that pirfenidone treatment was associated with lower all-cause mortality, fewer acute respiratory-related hospitalisation and more frequent treatment discontinuation compared to nintedanib. Beyond well-known limitations of observational comparisons, these real-world data suggest potential differences in outcomes between IPF medications, that warrant further exploration with different datasets and approaches.