Related Experiment Video
Updated: Oct 25, 2025
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
Validation of a claims-based algorithm for identifying non-infectious pneumonitis in patients diagnosed with lung
Shane S Neibart1, Daniella E Portal1, Jyoti Malhotra2
1Rutgers Robert Wood Johnson Medical School, New Brunswick, New Jersey, USA.
Purpose:
Non-infectious pneumonitis (NIP) is a common complication of treatments for lung cancer. We know of no existing validated algorithm for identifying NIP in claims databases, limiting our ability to understand the morbidity and mortality of this toxicity in real-world data.
Methods:
Electronic health records (EHR), cancer registry, and administrative data from a National Cancer Institute-designated comprehensive cancer center were queried for patients diagnosed with lung cancer between 10/01/2015-12/31/2020. Health insurance claims were searched for ICD-10-CM codes that indicate an inpatient or outpatient diagnosis with possible NIP. A 20-code (Algorithm A) and 11-code (Algorithm B) algorithm were tested with and without requiring prescription with corticosteroids. Cases with a diagnosis of possible NIP in the 6 months before their first lung cancer diagnosis were excluded. The algorithms were validated by reviewing the EHR. The positive predictive value (PPV) for each algorithm was computed with 95% confidence intervals (CI).
Results:
Seventy patients with lung cancer had a diagnosis code compatible with NIP: 36 (51.4%) inpatients and 34 (48.6%) outpatients. The PPV of Algorithm A was 77.1% (95% CI: 65.6-86.3). The PPV of Algorithm B was 86.9% (95% CI: 75.8-94.2). Requiring a documented prescription for a systemic corticosteroid improved the PPV of both Algorithm A and Algorithm B: 92.5% (95% CI: 79.6-98.4) and 100.0% (95% CI: 90.0-100.0), respectively.
Conclusions:
This study validated ICD-10-CM and prescription-claims-based definitions of NIP in lung cancer patients. All algorithms have at least reasonable performance. Enriching the algorithm with corticosteroid prescription records results in excellent performance.
