Validation of interstitial lung disease diagnosis codes in electronic health records and integrated open claims among cancer patients in the United States

 Interstitial lung disease (ILD) is a potential adverse effect of chemo- and immunotherapeutic agents used in the treatment of malignancy. The validity of ILD coding in U.S. claims data among patients with cancer remains unclear. 

We conducted a validation study to test the performance of claims-based algorithms in identifying patients with ILD among those who have been diagnosed with lung, breast, ovarian, endometrial, cervical, or colorectal cancer. We used electronic health record (EHR) data from the Mass General Brigham Health Care System linked with Medicare, Medicaid, and commercial claims data from 01/01/2017 to 12/31/2024. The base algorithm required two ICD-10-CM codes for ILD, each 7 to 180 days apart. Several nested algorithms additionally required ILD coding by a pulmonologist, ILD-relevant procedure, excluding autoimmune diseases, and requiring chemo/immunotherapy treatment were defined. The positive predictive values (PPVs) of each claims-based algorithm were calculated against the referent standard established through chart review by pulmonologists. 

 The cohort consisted of 360 claims-identified patients. The PPV of the base algorithm was 59% (95% confidence interval: 53%–64%). The algorithm additionally requiring ILD-relevant procedure with chemo- or immunotherapy administration [text omitted in the provided image]–71%). The PPV of the base algorithm among patients with colorectal cancer was the lowest – 38% (24%–52%). The PPVs improved meaningfully when the index diagnosis of ILD was coded by a pulmonologist, ranging from 78–88% across different algorithms. 

 We found that, in patients diagnosed with lung, breast, gynecologic, or colorectal cancer, claims-based ILD diagnoses demonstrated moderate to low alignment with a reference standard by expert chart review. The PPV can be improved when ILD is coded by a pulmonologist, though at the cost of substantially fewer cases of ILD identified.