Stratified impact analysis of intrinsic phenotypes and therapeutic interventions on the clinical prognosis of cross-disease validation: right treatment for right patient in precision radiotherapy

Purpose: Align with precision radiotherapy campaign of selecting the “right treatment” for “right patient”, this study aims to analyze the stratified impact of intrinsic phenotypes and therapeutic interventions on the local recurrence (LR) of hypopharyngeal squamous cell carcinoma (HSCC) and non-small cell lung cancer (NSCLC) patients and demonstrate the clinical associations between planomics features and treatment outcomes. Methods: A total of 76 HSCC patients and 123 NSCLC patients treated with radiotherapy were retrospectively analyzed. Patients were randomly divided into the training, validation, and test sets respectively. From each patient, 1316 radiomics, 1316 dosiomics, and 2951 planomics features were extracted. Three predictive models based on the LogisticRegression with elastic net regularization were developed: (1) Intrinsic Phenotypic model (radiomics + clinical), (2) Therapeutic Interventional model (dosiomics + planomics), and (3) Hybrid model (all features). The models' performances were assessed using the area under the curve (AUC) in the test set. Results: After feature selection, 6 radiomics, 6 dosiomics, and 5 planomics features were retained for the HSCC cohort, whereas the NSCLC cohort retained 1 clinical factor, 11 radiomics, 8 dosiomics, and 9 planomics features. In the test set, the AUC for the Intrinsic Phenotypic model, Therapeutic Interventional, and Hybrid models for the HSCC cohort were 0.607, 0.719, and 0.800 respectively. For the NSCLC cohort, the corresponding AUCs were 0.898, 0.908, and 0.935 respectively. The Hybrid model consistently outperformed the other models across both cohorts. Conclusions: Planomics has been proven as a predictor for LR based on clinical outcomes. The stratified analysis on two diseases consistently demonstrated that the LR was more dependent on the therapeutic interventions than the intrinsic phenotypes. The hybrid models integrating intrinsic phenotypes and therapeutic interventions provided the most accurate predicting results, assisting personalized precision medicine.