(B-230) Optimizing the Identification and Classification of Non-Small Cell Lung Cancer (NSCLC): Validation and Implementation of a Patient Cohort Algorithm in Claims Databases

Background: Administrative claims data lack diagnostic codes to identify lung cancer cases by histology subtypes and staging. Although the application of case-finding algorithms has minimized misclassification, improved accuracy, and led to robust findings, this requires regular updates to incorporate new treatment approvals and guidelines.

Objectives: This study updates and validates a case-finding algorithm to classify lung cancer patients (NSCLC vs. SCLC) and determine NSCLC stage (early vs. advanced) using medication and procedure-based methods in administrative claims data.

Methods: We refined existing case-finding and staging algorithms defined by Patel et al, JCO. 2024 (reference algorithm) in administrative claims data, by using updated clinical guidelines and insights from an Electronic Medical Records (EMR) database, ConcertAI RWD360® (CAI). This algorithm was validated in CAI using a study population diagnosed with lung cancer between January 1, 2015, and December 31, 2023 and receiving any lung cancer treatment per latest clinical guidelines. Histology information using procedure codes in CAI data was used to differentiate NSCLC vs. SCLC, and staging information was used to distinguish early (stage I-IIIA) vs. advanced (a) (stage IIIB - IV) NSCLC patients. Validation metrics included sensitivity and specificity; the positive predictive values (PPV) of algorithms were used to compare against the reference. After validation of the algorithm using CAI data, the algorithm was implemented in IQVIA PharMetrics® Plus database in treated lung cancer patients to setup a cohort of treated aNSCLC patients between 2015-2023.

Results: The updated and validated case-finding algorithm showed high sensitivity (87.4%) and specificity (93.8%) in identifying NSCLC patients in CAI. The PPV for the new algorithms compared against the reference algorithms were similar: case-finding (0.986 vs. 0.976) and staging (0.929 vs. 0.861), respectively. The cohort setup in the PharMetrics® Plus database (N=729,205) had an eligible lung cancer cohort of N=25,874, of which 13.02% were uncategorized by the algorithm. Among the categorized patients (N=22,506), this new algorithm identified 75.7% (n=17,034) as NSCLC, with majority of the patients grouped as advanced-stage; 24.31% (n=5,472) were classified as SCLC.

Conclusions: The new algorithm showed strong performance metrics in distinguishing lung cancer patient type and stage and could potentially enhance real-world application in administrative claims data. This methodology will continue to be refined with further updates and analyses and can serve as a useful tool in setting up standing cohorts of aNSCLC patients in various administrative claims datasets.