Data set and sample selection

We performed a retrospective cohort study with health insurance claims data of the Allgemeine Ortskrankenkasse (AOK) provided by the AOK Research Institute. AOK provides statutory health insurance for roughly 32 percent of the German population [16]. Membership is open to anyone regardless of factors such as professional affiliation, income, age or comorbidities [17].

The initial data set included all individuals insured with an ICD-10 diagnosis of various ILDs, including IPF [J84.1], other fibrosing ILDs [J84.0, J84.8, J84.9, D48.1], sarcoidosis [D86.0-D86.9], drug-associated ILDs [J70.2-J70.4], pneumoconiosis [J62.0-J62.8, J63.0-J63.8], radiation-associated pneumonitis [J70.1], eosinophilic pneumonia [J82], hypersensitivity pneumonitis [J67.9] and connective tissue-associated ILD [J99.1] from January 1, 2013 to December 31, 2018. Survival information was available until the end of 2019. For our analysis, we selected a subsample with at least one IPF diagnosis [J84.1] combined with at least one prescription of pirfenidone (tradename: “Esbriet”) or nintedanib (tradename: “Ofev”) in the patient-individual observation period. Furthermore, selected patients needed to be at least 40 years old at the date of the therapy initiation.

To identify relevant patients for our analysis we used the ATC-Codes “L04AX05” for pirfenidone and “L01XE31” for nintedanib. For the latter, only “Ofev” was considered via the national product codes (Pharmazentralnummer—PZN), as nintedanib under the tradename “Vargatev” is also licensed for treating distinct forms of non-small-cell lung cancer. Furthermore, the license of nintedanib for systemic sclerosis-associated interstitial lung disease and progressive fibrosing interstitial lung disease is not considered in this study as the approval took place after the study period.

We subsequently excluded patients, who were not continuously insured with the AOK and those with a baseline period (pre-observational) or an outcome period (post-observational) of less than one year. Hence, we dropped cases with a first prescription before January 1, 2014 or after December 31, 2017.

Therapy initiation was set as the date the patient redeemed the first drug prescription with either pirfenidone or nintedanib. To perform an intention-to-treat analysis group assignment was based on the first prescription.

Outcome variables

We compared pirfenidone and nintedanib-treated patients regarding two-year all-cause mortality, one-year all-cause as well as respiratory-related hospitalization, and one year overall and respiratory-related health care costs. All outcomes were calculated starting at the date of the treatment initiation.

Respiratory-related hospitalizations included all hospital visits with the following ICD-10 codes as primary diagnosis [18]: IPF [J84.1], respiratory infection [A481, B250, J09-J22, J40], pneumothorax [J93], pulmonary embolism [I26], pulmonary hypertension and right heart disease [I50, I270, I272, I278, I279], respiratory insufficiency [J96], and other chronic and acute lung diseases [J40-J47].

All-cause and respiratory-related health care costs in the year after therapy initiation were calculated based on outpatient physician costs, inpatient costs, and pharmaceutical costs, which could be directly obtained from the claims data. Respiratory-related costs in the inpatient and outpatient sector were based on cases with the diagnoses mentioned above. For respiratory-related pharmaceutical costs we filed prescriptions of pirfenidone, nintedanib, glucocorticoids, corticosteroids, immunosuppressants, acetylcysteine, sildenafil, and antihypertensives for pulmonary arterial hypertension. The corresponding ATC-Codes are presented in Additional file 1: Table S1. Outpatient physician costs were available on a quarterly basis only. Therefore, we redistributed costs incurred in the quarter of the treatment initiation proportionally to the time before and after the date of treatment initiation. Accordingly, this was also carried out for the last observation quarter. Inpatient costs could be determined on a daily base, but if a hospital stay exceeded the one-year follow-up time, costs were also distributed proportionally to in hospital days within the observation period. Pharmaceutical costs were calculated based on the day they were retrieved by the patient.

Covariates and stabilized inverse probability of treatment weighting

Stabilized inverse probability of treatment weighting (IPTW) based on propensity scores [19, 20] was used to adjust for differences in covariates between both groups, as no randomization was performed. The advantage of IPTW in comparison to k:1 matching is that all eligible patients retain in the analysis [21]. Covariates included in the estimation of the propensity scores were selected a priori based on clinical expertise and pre-existing literature. These covariates comprised age, gender, and residential area in four district types (major city, urban, rural, remote rural) [22]. Furthermore, area deprivation according to the well-established “German Index of Multiple Deprivation” from the year 2010 (GIMD 2010) was incorporated, which usually serves as a proxy for socioeconomic background if corresponding individual data is not available [23, 24]. Additionally, the time (in years) between the first IPF diagnosis (left-censored at 2013) and treatment initiation was taken into account. Comorbidities were considered via the Elixhauser Index [25] using the ICD-10 coding algorithm of Quan et al. [26]. To capture IPF-relevant comorbidities more precisely, we modified the Elixhauser Index as described by Schwarzkopf et al. [27]. Accordingly, we separated pulmonary hypertension and lung cancer from the corresponding Elixhauser categories and analyzed them separately. The remaining Elixhauser categories were included in the propensity score model, if they had a prevalence of at least 5% in at least one treatment group. In addition, we included other IPF-specific comorbidities not covered by the Elixhauser Index [27], namely gastro-oesophageal reflux disease [K21], obstructive sleep apnoea syndrome [G47.3], ischemic heart disease [I20-I25], and thromboembolism [I80, I26]. To avoid false-positive comorbidity diagnoses, patients had to have at least two confirmed outpatient physician diagnosis in two separated quarters or one primary inpatient diagnosis in the year before therapy initiation. We accounted for the intake of IPF-related or comorbidity-specific drugs in the six months prior to therapy initiation. IPF-related drugs included immunosuppressants, acetylcysteine, glucocorticoids, and corticosteroids. Comorbidity-specific drugs followed the prescription patterns described by Schwarzkopf et al. [27] with corresponding ATC-Codes presented in Additional file 1: Table S1. Furthermore, we considered the use of health care services before therapy initiation by reflecting IPF-specific hospitalization (IPF as primary diagnosis), respiratory-related hospitalization, and all-cause hospitalization in the three months before therapy initiation by corresponding binary variables. Also included were the number of contacts to outpatient pulmonologists and the number of overall contacts to outpatient physicians in the year before therapy initiation.

Standardized Mean Differences (SMD) were used to assess balance of the covariates between the groups after IPTW, with differences less than 0.1 indicating a good balance [28].

Statistical analysis

Mortality was analyzed by using IPTW-weighted mortality rates per 100 person-years and Kaplan–Meier plots, while hospitalization was investigated with IPTW-weighted hospitalization rates per 100 person-years and cumulative probability curves. Additionally, we calculated weighted Cox Proportional Hazard models with sandwich estimators of the variance for mortality and hospitalization to obtain hazard ratios. In addition, we investigated the three most common reasons for all-cause and respiratory-related hospitalizations. To compare health care costs for the year after treatment initiation, we calculated IPTW-weighted means with bootstrapped 95% confidence intervals (CI). Weighted group differences were also calculated with bootstrapped 95% confidence intervals.

We performed two sensitivity analyses (SA). For SA1, we excluded patients with a first pirfenidone prescription before 2015. Thus, we cover only the period, when both drugs were approved and excluded pirfenidone-treated patients with treatment initiation before nintedanib was available. For SA2, we considered treatment discontinuation, which was defined as a treatment gap of more than 60 days, with day 60 as discontinuation date [15]. Hence, treatment discontinuation became an additional censoring event in the time-to-event analyses. Additionally, we investigated how many of the patients discontinuing the treatment switched medication within the treatment gap range of 60 days. In the cost analysis we excluded costs after the discontinuation date. To account for the discontinued observation time in the cost analysis, we calculated costs per month.

Results were defined as significant if the calculated CIs of the differences did not contain “1” in the Cox models and the “0” in the cost analyses. All analyses were conducted with R-Software version 4.0.3.

Source link