Treatment patterns, healthcare resource utilization, and costs among patients with idiopathic pulmonary fibrosis treated with antifibrotic medications in US-based commercial and Medicare Supplemental claims databases: a retrospective cohort study

This observational retrospective cohort analysis utilized de-identified US administrative claims data covering October 1, 2014, to June 30, 2018, from three IBM MarketScan Databases: the Commercial Claims and Encounters (Commercial), the Medicare Supplemental and Coordination of Benefits (Medicare), and the Early View Databases [22]. The commercial database includes healthcare data of employees and their dependents covered under a variety of fee-for-service and managed care health plans. The Medicare database contains the healthcare experience of retirees with Medicare supplemental insurance paid for by employers. The Commercial and Medicare databases provided data access for the period of October 1, 2014, through September 30, 2017. Each database captures the inpatient medical, outpatient medical, and outpatient prescription drug data for its respective covered population, and together form a nationally representative sample of insured individuals living in the US.

The Early View database includes the same components as the Commercial and Medicare databases for the period of October 1, 2017, through June 30, 2018. Claims are available in the Early View database within 45 days of the end of the service month, and previous analysis has shown that over 97% of drug claims are fully adjudicated within the 30 days of the service date making it appropriate to use for treatment pattern analysis but not for HRU and cost analyses.

Study variables were based on enrollment information, International Classification of Diseases, 9th and 10th Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) codes, Current Procedural Terminology 4th edition codes, Healthcare Common Procedure Coding System codes, and National Drug Codes. All study data were accessed with protocols compliant with US patient confidentiality requirements, including the Health Insurance Portability and Accountability Act of 1996 regulations (HIPAA). The databases used in the study are fully de-identified and compliant with the HIPAA. Research using the MarketScan Research Databases falls under Title 45 of the Code of Federal Regulations Part 46.101(b)(4) exemption from Institutional Review Board review because the databases contain only information that cannot be used to identify study subjects. As the data were de-identified, consent from patients was not sought.

The study population consisted of patients with at least one outpatient pharmacy claim for pirfenidone or nintedanib from October 1, 2014, to June 30, 2018. The first such claim during the study period was set as the index date. Patients were required to be at least 40 years old on the index date, have continuous enrollment with medical and prescription coverage for at least 12 months before the index date (baseline period), and have at least one non-diagnostic inpatient or outpatient claim with an diagnosis code for IPF (ICD-9-CM: 516.3 and 516.31; ICD-10-CM: J84.112) between the baseline period and June 30, 2018. Patients were excluded if they had evidence of other interstitial lung diseases (hypersensitivity pneumonitis, diffuse connective tissue disease, rheumatoid arthritis and other inflammatory polyarthropathies, radiation fibrosis, pneumoconiosis, asbestosis, silicosis or talcosis, berylliosis and other inorganic dusts, unspecified pneumoconiosis, sarcoidosis) between the index date and June 30, 2018.

Three datasets were constructed from the set of patients who met the above criteria: the treatment patterns dataset, the variable-length HRU and cost while persistent dataset, and the 12-month fixed-length HRU and cost dataset. Patients in each dataset were stratified into two cohorts based on the antifibrotic received on the index date: pirfenidone or nintedanib.

The treatment patterns dataset contained all eligible patients and followed them from 12 months before their index date until the earliest of the following: inpatient death, end of continuous enrollment, or June 30, 2018 (end of the Early View data). The variable-length HRU and cost while persistent dataset comprised patients who had an index date on or before September 30, 2017 (end of non-Early View data). Patients were followed from 12 months before their index date until the earliest of the following: inpatient death, end of continuous enrollment, treatment discontinuation (defined below), or September 30, 2017. The 12-month fixed-length HRU and cost dataset comprised patients who had an index date on or before September 30, 2016 and at least 12 months of continuous enrollment in non-Early View data. Patients were followed from 12 months before their index date through 12 months after their index date, regardless of the length of time they persisted on their index medication.

Demographic characteristics, including age, sex, and geographic region, were measured on the index date. Baseline clinical characteristics were measured during the baseline period and included: the Deyo-Charlson Comorbidity Index (CCI) [23], the CCI excluding codes related to chronic obstructive pulmonary disease (COPD), supplemental oxygen use, opioid prescriptions, and claims with diagnosis codes for cardiovascular disease (including stroke specifically), COPD, cancer (including lung cancer specifically), dyspnea, and recent pneumonia (3 months preceding the index date). Patients who were treated at an Interstitial Lung Disease (ILD) Center of Excellence as identified by the zip code of the facility were also flagged. Centers of Excellence were defined as medical centers with specific expertise in the treatment of pulmonary fibrosis, as recognized by the Pulmonary Fibrosis Foundation.

Overall, 1455 patients with IPF met the selection criteria (Fig. 1). Of these, 799 patients received pirfenidone, and 656 received nintedanib. The mean patient age at index was 70.9 years, and 31.3% of patients were female (Table 1). Baseline characteristics were generally similar between the two cohorts; however, there were some differences in regional distribution and fewer pirfenidone users had a baseline diagnosis of COPD than nintedanib users (44.9% vs. 50.8%, P = 0.027).

The mean length of follow-up data available for treatment pattern analysis was 480.2 days (median: 403 days) among pirfenidone-treated patients and 446.6 days (median: 383 days) among nintedanib-treated patients (P = 0.064). Descriptively, the mean (SD) PDC in all patients and in both treatment groups was 0.66 (0.35) (Table 2). Over the variable-length follow-up, 48.8 and 50.5% of patients were considered adherent in the pirfenidone and nintedanib cohorts, respectively (P = 0.532).

The mean (SD) duration of persistence with index medication was significantly longer in the pirfenidone vs. nintedanib cohort in an comparison unadjusted for differences in patient characteristics (305.2 [314.7] vs. 264.4 [282.9] days; P = 0.010) (Table 2). Approximately 40% of patients in both cohorts discontinued their index medication (40.8% vs. 43.3%, P = 0.338). Among those with discontinued, 17.8% of patients re-initiated pirfenidone, and 14.0% re-initiated nintedanib (P = 0.197). The mean (SD) time to re-initiation of index medication was not significantly different between pirfenidone-treated and nintedanib-treated patients (150.4 [107.4] vs. 167.9 [142.2] days; P = 0.484). Among the 10.8% of pirfenidone patients and 8.8% of nintedanib patients who switched to the alternate antifibrotic, the time between the index and switching date was shorter for pirfenidone patients (251.4 [184.7] vs. 366.9 [227.5] days; P = 0.001).

In models weighted to account for differences in baseline characteristics, no significant difference in odds of adherence (OR [95% CI]) was observed between patients treated with pirfenidone and patients treated with nintedanib (1.07 [0.88–1.30]; P = 0.513). However, indexing on pirfenidone was associated with a longer time to discontinuation or switching than indexing on nintedanib (HR [95% CI]: 0.84 [0.71–0.99], P = 0.032) (Fig. 2). This association was stronger in the subgroup of patients who did not switch therapies (n = 1311; HR [95% CI]: 0.75 [0.63–0.91], P = 0.003).

Among the 697 pirfenidone users and 551 nintedanib users who indexed before September 30, 2017, the mean length of follow-up while persistent was 260.7 days (median, 179 days) in the pirfenidone cohort and 229.6 days (median, 134 days) in the nintedanib cohort (P = 0.025). While persistent with antifibrotic therapy, a significantly greater proportion of patients treated with pirfenidone than patients treated with nintedanib had at least one all-cause inpatient hospitalization (24.7% vs. 18.0%; P = 0.004) or respiratory-related inpatient hospitalization (13.9% vs. 8.9%; P = 0.006) (Table 3). The differences between pirfenidone and nintedanib cohorts in the number of all-cause inpatient hospitalizations (0.06 [0.18] vs. 0.05 [0.16]; P = 0.208) and respiratory-related inpatient hospitalizations (0.03 [0.14] vs. 0.02 [0.11]; P = 0.166) during the persistent period were not statistically significant.

While patients were persistent with therapy, the predicted all-cause healthcare costs based on the weighted multivariable models were comparable between pirfenidone and nintedanib treated patients ($11,272 vs. $11,987 PPPM, P = 0.115), but predicted respiratory-related costs were significantly lower in patients treated with pirfenidone than patients treated with nintedanib ($9015 vs. $10,167 PPPM, P < 0.001) (Fig. 3). While patients were persistent with therapy, predicted drug costs were significantly lower in patients treated with pirfenidone than patients treated with nintedanib ($7359 vs. $8686 PPPM; P < 0.001).

Among the 348 pirfenidone users and 266 nintedanib users with at least 12 months of post-index non-Early View data, 12-month all-cause and respiratory-related healthcare utilization rates were generally similar between cohorts (Table 4). Among patients treated with pirfenidone or nintedanib, roughly one-quarter had at least one all-cause inpatient hospitalization (26.1% vs. 24.4%; P = 0.629). While there was no statistically significant difference in the proportions of patients with at least one respiratory-related inpatient hospitalization, the mean (SD) length of stay for respiratory-related hospitalizations was significantly shorter in patients treated with pirfenidone than patients treated with nintedanib (4.1 [2.9] vs. 6.3 [5.4] days; P = 0.033).

In multivariable models, the annual total all-cause and respiratory-related healthcare costs among the subset of patients with 12 months of post-index enrollment were comparable between patients treated with pirfenidone and patients treated with nintedanib (Fig. 4). However, patients treated with pirfenidone had significantly lower adjusted annual mean drug costs than those treated with nintedanib ($68,850 vs. $77,033 PPPY, P = 0.007).

There are several limitations that should be considered when interpreting the findings of this study. First, as with any analysis of healthcare claims data, the adherence and persistence results reflect patterns in prescription fills (and not necessarily use) of the index medications; therefore, they may not fully capture usage patterns of antifibrotic therapy among patients with IPF. Second, unadjusted results for healthcare utilization did not account for potential differences in patient characteristics between treatment groups. Third, cost comparisons were adjusted for some potential confounders; however, residual confounding may exist due to the lack of clinical data included in administrative claims data or other unobserved differences between cohorts. Because patients were selected based on initiation of pirfenidone vs. nintedanib and clinical notes are not available, we were not able to control for years since initial diagnosis. Measures of disease severity, such as results of pulmonary function tests and GAP stage, were also not available. Fourth, only a small number of patients had 12 months of non-Early View follow-up data, so the analysis may be underpowered to detect differences in HRU and costs. Fifth, the proportions of patients with specific utilization types measured during the persistent period did not account for variable follow-up time. Sixth, prior to the approval of these medications there was no effective treatment for IPF; therefore, patients in this analysis may be sicker on average than future patients who have access to these medications upon initial diagnosis. Patients who did not survive 12 months after index medication initiation did not meet the continuous enrollment criteria for the fixed 12-month follow-up outcomes. Finally, analysis was restricted to patients with commercial or Medicare Supplemental insurance and findings may not extend to patients with other types of insurance or no insurance.