Precision Medicine · us
AI helps identify gaps in lung cancer testing as community healthcare evaluation finds completion rate rises to 90%
The first hurdle in precision treatment may be whether testing is completed. A before-and-after comparison of an AI-assisted notification workflow showed improved completion of lung cancer biomarker testing, but whether AI deserves the credit still requires more rigorous validation.
Personalizing lung cancer treatment requires healthcare teams to first obtain sufficient information about the tumor. However, any interruption between arranging testing and receiving results can leave treatment decisions without important clues. An evaluation in a community healthcare system found that after a workflow was introduced to use AI to help identify testing gaps and notify physicians, the biomarker testing completion rate among patients with non-small cell lung cancer rose from 71.1% to 90%, a difference of 18.9 percentage points.
The findings were presented at the 2026 ASCO Quality Care Symposium. The study compared two periods before and after implementation: a retrospective analysis of 471 adult patients in 2024, followed by a prospective evaluation of 1,115 patients between 2025 and 2026. Patients with stage IB through IV disease, as well as those with unknown stage, were included.
The system is intended to help track care processes. The tool, called Tempus Next, imports clinical data nightly and uses AI-assisted information extraction alongside clinical data review to identify patients with outstanding testing, then sends notifications to physicians. The tests tracked in the study varied by disease stage, including EGFR, ALK and PD-L1 testing for patients with early-stage disease, and multigene panel testing and other tests for those with advanced disease or unknown stage.
Patients with unknown stage had the largest increase in completion rate, from 49.6% to 76.8%. The results for this group also suggest that incomplete data and incomplete testing may be issues healthcare teams need to address together; an increase in the overall completion rate still does not mean that every patient group has received equally adequate testing.
Gaps in follow-up action remained after notifications were sent. During the study, physicians received notifications concerning 120 patients, of whom 24 subsequently completed testing, 83 had no testing record, and another 13 were no longer eligible for care. The absence of a record cannot be directly equated with confirmed lack of testing, but these figures show that identifying gaps is only a starting point, with continued follow-up and coordination still needed.
This study offers a preliminary signal of process improvement, but it cannot establish a causal relationship between AI and the increase in completion rate. Differences in patient populations, testing practices or other care measures between periods could all have affected the results. Existing reporting also did not provide sufficient data to determine whether patient survival or treatment outcomes improved. The key next step is to use more rigorous comparative designs to determine whether these notifications can consistently lead to appropriate testing and truly help clinical decision-making.