Identifying clinical trial candidates using AI predictions of treatment change: A pilot implementation study.

Abstract

1515 Background: Precision oncology clinical trials often struggle to accrue due to the difficulty of identifying patients whose tumors meet complex tumor genomic matching criteria at propitious moments when they need new treatment. We conducted a pilot of deploying an artificial-intelligence (AI) model to find such patients and facilitate clinical trial matching workflows at a large academic cancer center. Methods: An AI model was trained to predict initiation of new palliative intent systemic therapy within 30 days of a given imaging report (CT, MRI, bone scan, or PET-CT), using the text of that report and prior reports for each patient. The model architecture and performance were previously published (Kehl et al, JCO-CCI 2021 5:622-630). From April to December 2022, the model was applied prospectively each time patients with solid tumors that had undergone next-generation sequencing underwent an imaging study. Model output was linked to the MatchMiner tool, which matches patients to clinical trials using tumor genomics (Klein et al, NPJ Precis Oncol 2022 6(1):69). The output consisted of lists of patients with tumor genomic matches to specific trials, sorted in order of the likelihood of changing treatment within 30 days. This information was provided to an oncology nurse navigator (ONN) charged with coordinating recruitment to early-phase targeted and immunotherapy trials. For 9 high priority trials, the ONN reviewed records for each patient flagged as likely to change treatment and contacted treating oncologists when patients appeared potentially eligible. The proportions of records leading to oncologist contact, and reasons for oncologist non-contact, were analyzed. Results: 2093 patients had tumors with genomic matches to the 9 trials of interest; 492 patients (24%) were deemed “likely” to change treatment by our AI model at least once during the pilot. The treating oncologist for 66 patients (3% of the total; 13% of 492 “likely” patients) was contacted after ONN review for potential eligibility. Reasons for not contacting treating oncologists included cases where they had already decided to continue current treatment (21%); the trial had no slots at the time (14%); or the patient was ineligible on nurse navigator review (12%); or had already been evaluated for the trial (7%), started new treatment (18%), or enrolled in hospice (5%). Of the 66 patients whose oncologists were contacted, 9 patients had a consult regarding early phase trials; 4 consented to participate; and all 4 received protocol treatment. Conclusions: Identifying potential precision clinical trial candidates by using AI to find patients likely to change treatment is feasible, but many other factors also impact potential trial enrollment. A prospective study is planned to evaluate the impact of sharing clinical trial information directly with treating oncologists when our model predicts a high probability of treatment change.

Ähnliche Arbeiten