The following is a summary of “Integrated non-invasive diagnostics for prediction of survival in immunotherapy,” published in the July 2024 issue of Oncology by Yeghaian et al.
Integrating diverse non-invasive diagnostic sources holds significant promise for enhancing the predictive accuracy of Artificial Intelligence (AI) models, a critical step toward their clinical validation and implementation. This study investigates the potential of integrating data from chest CT imaging, routine laboratory blood tests, and clinical parameters to retrospectively predict 1-year survival in patients with advanced non-small cell lung cancer, melanoma, and urothelial cancer undergoing immunotherapy.
A total of 475 patients were included, 444 had longitudinal chest CT scans, and all 475 had longitudinal laboratory data available. AI models were individually trained on each diagnostic modality, followed by a model-agnostic approach to integrate the prediction probabilities from each modality into a unified decision.
Integrating data from multiple diagnostic modalities resulted in a noticeable enhancement in predictive performance. The combined model using both CT imaging and laboratory data achieved the highest area under the curve (AUC) of 0.83 (95% CI: 0.81-0.85, p < 0.001). In comparison, models trained solely on laboratory data or CT imaging independently yielded AUCs of 0.81 and 0.73, respectively.
In this retrospective cohort study, integrating diverse non-invasive diagnostic modalities significantly improved the predictive capabilities of AI models. This finding underscores the potential of combining chest CT imaging, laboratory blood tests, and clinical parameters to enhance prognostic accuracy in patients undergoing immunotherapy for advanced cancers.
Source: sciencedirect.com/science/article/pii/S2590018824000200