AI Algorithm Flags 3% Misdiagnosis Rate in Lung Cancer Cases
Caris Life Sciences published findings in JAMA Network Open showing that an AI algorithm identified 123 cases of misdiagnosed lung squamous cell carcinoma out of nearly 4,000 cases submitted for molecular profiling. The tumors were actually metastatic cancers that had spread to the lung from other sites - skin, head and neck, bladder, and thymus - rather than primary lung cancers.
The misclassifications matter because they led to different first-line treatment recommendations under established clinical guidelines. Patients initially thought to have early-stage lung cancer were actually dealing with metastatic disease requiring different therapy.
How the AI Model Worked
The AI system analyzed gene expression, genomic alterations, and other molecular signals simultaneously. Squamous cell carcinoma looks similar under a microscope regardless of where it originated, making traditional pathology unreliable for distinguishing primary lung tumors from metastases.
The algorithm flagged cases where molecular patterns - such as UV-induced mutations in skin cancers or human papillomavirus signals - suggested a different tissue origin. In roughly three-quarters of the misdiagnosed cases, the corrected diagnoses aligned with known clinical or imaging findings, validating the AI's assessment.
Scale and Consistency
A key advantage was that the AI screened every case, not just those suspected of misdiagnosis. Many diagnostic errors persist because clinicians lack suspicion or incomplete clinical context. The algorithm operated as a consistent quality-control checkpoint independent of individual experience or vigilance.
At population scale, a 3.1% misdiagnosis rate translates to thousands of patients annually receiving incorrect treatment for this single cancer subtype. Dr. Matthew Oberley, chief clinical officer at Caris Life Sciences, said the findings suggest that systematic AI deployment could reduce errors and ensure patients receive therapies matched to their disease's true biology.
Implications for Care Settings
Standardizing tumor-origin evaluation across institutions could reduce diagnostic variability, particularly benefiting community hospitals and regions with limited subspecialty pathology expertise. The approach could democratize access to advanced diagnostic capabilities beyond academic medical centers.
Accurate tumor classification is foundational to selecting appropriate therapy, enrolling patients in clinical trials, and predicting outcomes. As AI for Healthcare becomes more integrated into routine workflows, it may uncover previously unrecognized disease patterns and support the shift toward precision oncology at scale.
Your membership also unlocks:
