This retrospective study included patients from the Syapse Learning Health Network (LHN), a longitudinal database that captures data from cancer patients treated in community-based integrated care delivery networks across 25 states in the USA, including 457 hospitals and over 1300 oncologists. Each patient’s records include data from various sources: electronic health records (EHR), laboratory and radiology/imaging systems, computerized order entry systems, and hospital-based cancer registries. The study population included all patients who were 18 years or older at time of initial breast cancer diagnosis, had evidence of mBC (stage IV at diagnosis or physician-assessed progression/recurrence), and had initiated a HER2-directed therapy in the metastatic setting between September 25, 1998, and February 22, 2020. Patients with locally advanced disease who had not developed metastases were not included in this study. Patients were followed from their index date (initiation of HER2-directed therapy in the metastatic setting) until the earliest of their first incident ILD event (defined below) or date of death, last contact, or study end on May 22, 2020. This study end date allowed for a minimum of three months of potential follow-up when data analysis and curation started. The date of death was determined by a validated composite mortality score [22].

To allow for six months capture of drug prescribing in routine care, the following HER2-directed therapies, all approved by the US FDA prior to December 2019, were included: trastuzumab, trastuzumab-anns, trastuzumab-dkst, trastuzumab-dttb, trastuzumab-pkrb, trastuzumab-qyyp, ado-trastuzumab emtansine, neratinib, lapatinib, and pertuzumab. Patients treated with trastuzumab deruxtecan or tucatinib during the study period were excluded as these agents were approved after December 2019.

ILD was defined by the presence of at least one of 64 lung conditions identified through review of clinician notes, imaging narratives, pathology reports, bronchoscopy results, and relevant International Classification of Diseases (ICD) codes recorded in patient charts (Supplemental Table 1). Chronic obstructive pulmonary disease (COPD), while documented separately, was excluded from the list of terms defining ILD. Given the lack of a functional definition of ILD, in an attempt to include all the relevant terms and case presentations in this analysis, the standardized Medical Dictionary for Regulatory Activities (MedDRA) queries for ILD were used and augmented with any term considered to be indicative of a potential pulmonary inflammatory or fibrotic event based on expert clinical input (Supplemental Table 1). This broad operational definition of ILD was used to capture any possible occurrence of interstitial lung changes, document the heterogeneous nature of ILD, and capture the earliest signs of the ILD. To reduce potential misclassification, records for patients with pleural effusion alone on or after the patient’s diagnosis of mBC were reviewed in detail, and patients with evidence of malignant pleural effusion unaccompanied by a second qualifying lung condition were not considered to have ILD. Patients with lung metastases who did not present with pleural effusion or other ILD lung conditions were also not considered to have ILD. Patients with ILD identified exclusively in the follow-up (post-index) period were classified as having incident ILD. Patients with any evidence of ILD in the baseline period were considered to have prevalent ILD.

HER2-directed therapy use in the neoadjuvant and adjuvant settings, use of potentially ILD-inducing therapies (Supplemental Table 2) [16], and radiation to the thoracic region were assessed in the one-year pre-index period. Age, race, and performance status were documented at metastatic disease diagnosis date, and body mass index (BMI) was assessed at index date. Performance status measures were mapped to Eastern Cooperative Oncology Group (ECOG) performance status if alternative performance measures were documented. Smoking history was assessed at the initial breast cancer diagnosis date and each patient was categorized as an “ever” or “never” smoker. Additionally, pre-index data were used to assess baseline patient demographic and clinical characteristics, including identifying prevalent ILD and the presence of associated preexisting lung conditions such as COPD, pulmonary embolism, lung cancer, or lung metastasis. “Rapid” initiation of HER2-directed therapy was defined as starting HER2-directed therapy within 1 month from the date of first metastasis.

Relevant clinical attributes were extracted from the EHRs of patients with evidence of ILD. These data included: occurrence of ILD-specific lung diagnosis, signs and symptoms and their timing, and anticancer treatment regimen at ILD onset. In addition, an interval spanning six months before the first recorded ILD event until six months after the last recorded ILD event was used to search for hospitalizations, emergency room (ER) visits, supplemental oxygen, and mechanical ventilation. The use of steroid treatment for ILD management was also assessed.

The distribution of baseline patient demographic and clinical characteristics was evaluated for the population overall and additionally stratified by ILD status per three mutually exclusive groups: (1) the “none” group includes patients without prevalent or incident ILD; (2) the “prevalent” group includes patients with prevalent ILD, excluding those with incident ILD; and (3) the “incident” group includes patients with incident ILD, excluding those with prevalent ILD. The distributions of categorical variables were summarized as frequencies and percentages, while medians and interquartile ranges (IQR: 25th percentile, 75th percentile) were reported for continuous variables.

Cumulative incidence curves were used to characterize the distribution of time to the first incident ILD event for the subset of patients without prevalent ILD. Patients who were diagnosed with mBC in or after 2010, a time period in which death data were available, were considered for all time-to-event analyses. Results were reported separately for ILD incidence for two scenarios: anytime after the initiation of HER2-directed therapy and ILD incidence during HER2-directed therapy. In the first scenario, patients were followed for incident ILD from index date to the earliest of first recorded ILD event, death, last contact, or end of study.

To account for variable duration of HER2-directed therapy in the second scenario, patients were followed from their index date to the earliest of first recorded ILD, first HER2-directed therapy discontinuation or termination, date of death, date of last contact or study end. Gaps in HER2-directed therapy of 30 days or less were bridged, creating a contiguous treatment period. Cumulative incidence estimates are reported overall, and additionally stratified by smoking status and presence/absence of associated preexisting lung conditions. Additionally, Cox proportional hazard models were used to estimate associations between pre-index demographic and clinical factors and incident ILD risk while on HER2-directed therapy. These associations were assessed among all patients, as well as restricted to those without prevalent ILD, in the event that subsequent ILD may reflect unresolved, existing disease in patients with prevalent ILD. Models included the following covariates: age, race, body mass index (BMI), smoking status, associated preexisting lung conditions, history of thoracic radiation excluding radiation to intact breast, potentially ILD-inducing systemic therapy, time between mBC diagnosis and index date, ECOG performance score, and presence/absence of prevalent ILD. The proportional hazards assumption was assessed graphically using Schoenfeld residuals.

All data analyses were performed in R version 3.6.1. The level of significance was set at 0.05.