346 Background: Data elements required for clinical trial eligibility assessments are often unstandardized and incomplete in EHR making clinical research participant identification a challenging and biased process. Moreover, the absence of new patient records and disease codes results in monitoring large patient populations until records are completed. These issues can lead to missed opportunities for timely patient identification. N-Power Medicine (NPM) developed a registry-enhanced oncology network platform for unbiased, and real-time quality data collection. We report from this platform on completeness rates of key variables for eligibility assessment in NSCLC and CRC clinical trials. Methods: NPM’s platform consists of 3 parts: patient registry (Kaleido), onsite and remote staff to consent patients and ensure data completeness, quality and standardization, and technology to structure and process the data. Patients with NSCLC and CRC from 3 community oncology practices across different US regions were enrolled in Kaleido and visited the clinic between February and May 2024. Data for 11 key variables were abstracted. To minimize the need for extensive monitoring until a diagnosis was confirmed, we developed a machine learning (ML) model for lung cancer using non-C34 ICD10 codes to construct multiple classes of tunable predictive models. Results: A total of 728 (NSCLC=281, CRC=447) patients were evaluated. The male-to-female ratio was 1 for NSCLC and 1.3 for CRC and 90% of NSCLC and 68% of CRC patients were over 60 years of age. The platform achieved 99% completeness for critical data points such as diagnosis date, histology, metastatic status, recent systemic anti-cancer treatments, ECOG performance status and other cancer diagnoses with diagnosis date. 47% of NSCLC and 43% of CRC patients had metastatic disease at the time of evaluation. Over 90% of patients with metastasis had their current therapy line and relevant genomic test results successfully documented. Imaging-based cancer status completeness was 95% for NSCLC and 89% for CRC. The ML model would identify over 50% of all patients that would eventually receive a lung cancer diagnosis, with over 80% specificity up to 3 weeks before a diagnosis. Conclusions: NPM’s platform demonstrated high completeness rates for the majority of essential data elements, improving the inclusion of all eligible patients in clinical trials. The predictive ML model shows promise in identifying suitable patients before an ICD10 code is assigned, reducing the need for extensive monitoring and ensuring timely completeness of necessary assessments for eligibility. Importantly, these advancements were achieved without adding financial or resource burdens to clinical workflows.
Read full abstract