MD Anderson Researchers Publish AI Study Predicting Lung Inflammation Risk Before Immunotherapy

A chest scan taken before lung cancer treatment may hold a warning about a dangerous side effect. In newly published research, an AI model from UT MD Anderson estimated which patients were more likely to develop lung inflammation after immunotherapy. It performed similarly in a hospital test and an external dataset, but its use in care still needs prospective study.

The warning hidden in a routine scan

The condition is pneumonitis, a potentially life-threatening inflammation of the lungs. According to MD Anderson, it occurs in about 10% of lung cancer patients receiving immunotherapy. Identifying higher-risk patients before symptoms appear could help clinicians choose whom to monitor more closely.

That is the proposed role of CIPHER, the model described in the Journal for ImmunoTherapy of Cancer study. It reads chest CT scans obtained before treatment and looks for patterns associated with patients who later developed pneumonitis. The researchers say it outperformed approaches based on clinical risk factors or radiomics, which extracts measured features from images.

First learn lung tissue, then look for risk

The team first trained CIPHER to recognize patterns in lung tissue, using more than 590,000 CT image slices from 2,500 people with lung cancer. It did not begin by learning directly from confirmed pneumonitis cases. Researchers then assessed whether the patterns it had learned could distinguish patients who developed the condition after immunotherapy.

What the second dataset adds

Area under the curve, or AUC, measures how well a model separates patients who later developed the condition from those who did not. It is not the percentage of patients correctly diagnosed. The external test matters because its patient population, CT scanners and imaging protocols differed from those in the MD Anderson test.

The model's predictions remained significant after researchers accounted for age, smoking history, tumor type and prior radiation to the chest. Patients CIPHER classified as high-risk also tended to develop pneumonitis sooner after starting immunotherapy. That timing might help target early follow-up, though no monitoring plan was tested here.

A risk score is not a care plan

The distinction between prediction and clinical benefit is the next hurdle. MD Anderson says larger prospective studies, following patients forward, are needed to determine whether CIPHER can fit into clinical workflows. A score that separates past cases is promising; deciding how to respond to that score before symptoms appear requires further testing.

The researchers also plan to assess whether the model works for other cancers treated with immunotherapy. They may test whether adding other biological markers improves its predictions. Neither extension is established by this lung cancer result.