Multi-center prospective study evaluating an AI-enabled clinical decision support tool to improve biomarker testing in early-stage NSCLC.

Abstract

8044 Background: Non-adherence to guideline-concordant biomarker testing in non-small cell lung cancer (NSCLC) can limit access to targeted therapies and adversely impact survival. We evaluated an AI-enabled clinical decision support (AI-CDSS) program comprising: (1) education around baseline testing rates; (2) continuous monitoring to generate real-time alerts for eligible patients with missing biomarker testing; and (3) longitudinal feedback via dashboards. Here, we report the effectiveness of this program in identifying and closing biomarker testing gaps for patients with early-stage NSCLC. Methods: In this descriptive study, we analyzed patients with confirmed NSCLC across 6 geographically and socioeconomically diverse US community health systems. The AI-CDSS identified early-stage patients eligible for biomarker testing (eNSCLC as AJCC 8th edition Stg IB-IIIB (T3, N2) with planned curative intent treatment). Biomarker testing included EGFR, ALK, and PD-L1. We compared testing adherence between a baseline period (BL: 24 months through 3 months prior to the health system-specific roll-out) and a post-launch period (PL: roll-out through Oct 2025). The AI-CDSS was implemented on a rolling basis across health systems (BL from Feb 2022 - Dec 2024 and PL from Feb 2024 - Oct 2025). Testing rates were calculated as the proportion (%) of patients with testing completed within 90 days of pathologic diagnosis in each period. The improvement in test rates (absolute lift) is calculated as the difference in PL - BL testing percentages in the two periods. Results: A total of 662 patients with eNSCLC (270 BL and 392 PL) were included in the analysis. Patients were predominately white (85%), had a history of smoking (88%), with a median age of 70 years at diagnosis. The stage distribution was as follows: Stage III (34%), Stage II (37%), Stage IB (25%), and Stage IB or IIA [indeterminate] (5%). The absolute lift in biomarker testing within 90 days of pathologic diagnosis before vs after intervention was 18% for EGFR, 24% for ALK, and 13% for PDL1 biomarkers. Among patients with molecular testing who received adjuvant treatment, 89% were on guideline-concordant adjuvant treatment. Conclusions: Implementation of an AI-CDSS was associated with clinically meaningful improvements in rates of biomarker testing for eNSCLC and resulted in high concordance with guideline-directed adjuvant therapy. Appropriate and timely biomarker testing is essential for perioperative treatment planning. This study provides preliminary evidence that AI can use complex electronic health records to provide real-time interventions that can promote guideline-concordant care. Testing gap results. Biomarker Baseline N Baseline Test Rate Post Launch N Post Launch Test Rate Absolute Lift EGFR 264 49% 392 67% 18% ALK 270 43% 389 67% 24% PD-L1 270 59% 389 72% 13%