Association between nintedanib adherence trajectory and healthcare use among idiopathic pulmonary fibrosis patients

This was a descriptive, non-interventional cohort study that used the administrative data and patient sample from our earlier study [24]. The data source for this study was 100% fee-for-service Medicare data from 2013 to 2019 to allow for a 360-day baseline period and a 360-day follow-up period. The enrollment file contains monthly information on individuals’ enrollment in each part of Medicare, demographic information, residential location, and date of death. Claims were available for inpatient hospital, skilled nursing facility, and outpatient facility services (Part A), physician and other professional services (Part B) including emergency department visits, and outpatient prescription drugs (Part D). Part D claims include standardized prescription-level information, including drug names, National Drug Codes (NDCs), strength, quantity, days’ supply, and fill date.

The study cohort consisted of community-dwelling older adults (≥ 66 years) with IPF who started treatment with nintedanib between 10/01/2014 and 12/31/2018. The index date was defined as the date of the first prescription fill for nintedanib. Patients were required to have continuous coverage in Medicare Parts A, B and D for 12 months prior to the index date to preclude no prior antifibrotic use and for 12 months following index date to enable the capture of medication adherence history. Patients were required to qualify for Medicare based on age and were excluded if they used skilled nursing facilities, used long-term care facilities or hospice care (because patterns of medical care use are atypical among patients in institutionalized and end-of-life settings), and were dual-eligible for Medicaid and Medicare during the baseline and follow-up periods. Additional details, including sample selection criteria, are available in the original publication [24]. This study was determined to be exempt from review by the WCG Institutional Review Board.

Inpatient hospitalization and total medical spending outcomes were measured during the year following nintedanib initiation. Inpatient hospitalization was measured in three ways: (i) as a binary indicator for whether a patient had at least one inpatient hospitalization claim; (ii) as the count of days that a patient was hospitalized; and (iii) as the elapsed time from the index date to the first hospitalization (censored at end of follow-up). Total medical costs were calculated as the sum of total amounts paid (by all payers and patients) for all medical services that occurred during the 360-day period. Types of medical services covered by Medicare include inpatient facility, outpatient facility, skilled nursing facility, home healthcare, hospice, durable medical equipment, and clinician office visits (and other services covered under the Part B benefit). Two versions of each outcome were constructed: the all-cause, which included all claims, and the IPF-related, which included only claims that had an IPF diagnosis code in any position.

The principal explanatory variable of interest was the five-category measure of nintedanib adherence trajectory group membership (Fig. 1). This variable indicates the specific adherence trajectory to which a patient was most likely to belong based on the predicted probabilities of trajectory membership derived from the best-fitting GBTM in our prior analysis [24].

Covariates were selected from the Medicare data based on their plausibility as confounders and were measured from the index claim and during the 360-day baseline period. Patient demographics included age on index date and indicators for female sex and non-Hispanic white race and ethnicity. Contextual measures included the year of the index date, residential Census region, and Social Deprivation Index score. The Social Deprivation Index combines ZIP Code-level Census data on poverty, education, housing, car ownership and employment into a score between 0 and 100, with higher values indicating more social deprivation [27]. Clinical measures included the count of 20 comorbid conditions found to be predictive of 1-year mortality; [28] indicators for four IPF-related comorbidities: gastroesophageal reflux disease, hypoxia, pulmonary hypertension, and sleep apnea; indicators for receipt of three IPF-related medical services: high-resolution CT scan, lung biopsy, and oxygen therapy/supplemental oxygen; an indicator for whether the prescriber for the index nintedanib claim was a pulmonologist; and four measures of baseline all-cause healthcare use: indicators for any emergency department visit and any inpatient hospital admission, total prescription drug spending, and total medical care (non-pharmacy) spending. All dollar-denominated measures were adjusted to 2019 US dollars using the Medical Care component of the US Consumer Price Index (retrieved from the US Bureau of Labor Statistics on August 1, 2022).

Baseline characteristics were summarized as means and standard deviations for continuous measures or counts and proportions for categorical measures overall and stratified by nintedanib adherence trajectory. Means and proportions were compared for joint equality across adherence trajectories with ANOVA and χ [2] tests, respectively.

A regression modeling framework was employed to compare outcomes between pairs of nintedanib adherence trajectories. The high adherence trajectory was the referent. Unadjusted outcomes were compared first using an “empty” linear regression model specification with no covariates. Adjusted outcomes were compared using a regression model specification that included covariates and had a functional form based on its observed distribution in the data. The binary “any hospitalization” outcomes were modeled with logistic regression. The non-negative “hospital days” outcomes were modeled with a zero-inflated quasi-maximum likelihood Poisson model to account for the non-trivial proportion of beneficiaries with no hospitalizations during follow-up. The “total medical costs” outcomes were modeled using a GLM with a log link and a gamma family. Average marginal effects and their 95% confidence intervals (CIs) were calculated using the delta method. Unadjusted and covariate-adjusted Cox proportional hazards models were estimated to enable pairwise comparison of hospitalization risk across nintedanib adherence trajectories relative to the high adherence trajectory, as expressed by hazard ratios. Standard errors were made robust to heteroskedasticity of unknown form.

To help place the findings in context, a supplemental descriptive analysis was conducted to characterize aspects of the use of both antifibrotic medications available in the study period, nintedanib and pirfenidone, during the one-year follow-up period stratified by nintedanib adherence trajectory group. Antifibrotic use was measured as the proportions of patients discontinuing nintedanib, using pirfenidone, and using pirfenidone conditional on discontinuing nintedanib. PDCs were characterized over the entire one-year follow-up for nintedanib alone, pirfenidone alone, and nintedanib and pirfenidone combined. Means and proportions were compared for joint equality across adherence trajectories with ANOVA and χ [2] tests, respectively.

Data preparation was performed using SAS 9.4 (SAS Institute Inc., Cary, NC), and data analyses were conducted using Stata 17 (StataCorp, College Station, TX). Statistical significance was based on 2-sided tests with α = 0.05.