AI Study Reveals Hidden Misdiagnoses in Lung Cancer, Raising Stakes for Precision Treatment
A new study published in JAMA Network Open by Caris Life Sciences highlights how artificial intelligence is uncovering clinically significant misdiagnoses in lung cancer—findings that could directly impact treatment decisions and patient outcomes.
The research demonstrates that integrating AI into routine cancer diagnostics can identify cases where tumors initially classified as lung squamous cell carcinoma (SCC) actually originated elsewhere in the body.
AI Identifies Misdiagnosed Cancer Cases
The study analyzed nearly 4,000 cases previously diagnosed as primary lung squamous cell carcinoma. Using Caris’ proprietary GPSai algorithm alongside molecular profiling and clinical data, researchers found that 123 cases (approximately 3.1%) were misdiagnosed.
Instead of originating in the lung, these tumors had metastasized from other primary sites, including:
- Skin (cutaneous cancers)
- Head and neck cancers
- Urothelial cancers
- Thymic cancers
According to Matthew Oberley, distinguishing between primary lung SCC and metastatic tumors is particularly challenging, as they often appear similar under traditional pathology methods.
AI Enhances Diagnostic Accuracy
The AI model’s findings were validated using multiple independent data sources, including genomic signatures, immunohistochemistry, and clinical histories. In roughly 75% of cases, the corrected diagnoses aligned with existing clinical or imaging evidence, reinforcing the reliability of the AI-driven approach.
Experts say this highlights AI’s strength in analyzing complex, multidimensional datasets—a capability that surpasses traditional diagnostic techniques reliant on visual interpretation and limited biomarkers.
Unlike conventional methods, AI integrates:
- Gene expression data
- Genomic alterations
- Molecular and biological signals
This allows it to detect subtle patterns that may otherwise go unnoticed.
Impact on Treatment Decisions
One of the most significant findings is the potential impact on patient care. In the majority of misdiagnosed cases, correcting the tumor’s origin would have led to different first-line treatment recommendations under established clinical guidelines.
Even a relatively small error rate can have major consequences, particularly when early-stage lung cancer is incorrectly diagnosed instead of metastatic disease.
At scale, researchers warn, such misclassifications could affect thousands of patients annually, leading to suboptimal or inappropriate treatment.
AI as a Continuous Quality Control Tool
The study also highlights AI’s role as an always-on diagnostic safety net.
Unlike traditional workflows, which rely on clinician suspicion, the AI model was applied systematically to every case. This enabled it to flag discrepancies regardless of prior assumptions, helping uncover errors that might otherwise persist.
AI can also detect molecular “signatures” tied to specific cancer origins, such as:
- UV-related mutations in skin cancers
- Virus-associated signals like HPV
These indicators are often overlooked in routine diagnostic processes.
Advancing Precision Oncology
Researchers believe widespread adoption of AI-driven diagnostics could significantly improve the consistency and accuracy of cancer care.
Key potential benefits include:
- Standardizing diagnostic quality across healthcare systems
- Supporting clinicians in resource-limited settings
- Enabling more precise treatment selection
- Improving eligibility for targeted therapies and clinical trials
By accurately identifying tumor origin, AI could accelerate the broader shift toward precision oncology, where treatments are tailored to the biological characteristics of each patient’s disease.
The Bigger Picture
While the misdiagnosis rate identified in the study may appear modest, its implications are far-reaching. As AI becomes more integrated into clinical workflows, it has the potential not only to correct existing errors but also to uncover previously undetected disease patterns.
Ultimately, experts say, the technology represents a powerful tool for improving diagnostic accuracy, enhancing patient outcomes, and driving a more data-driven future for cancer care.
