Background: Postural instability and gait difficulties are key symptoms of Parkinson's disease (PD), elevating the risk of falls substantially. Falls afflict 35% to 90% of PD patients, representing a major challenge in managing the condition. Accurate prediction of fall risk and identification of contributing factors are essential for timely interventions.Objectives: Our objective was to develop and validate a machine learning (ML) algorithm across multiple centers in Italy to accurately forecast fall risk and identify related factors using routinely collected clinical data.Methods: Patient data from two Italian centers (N = 251) were divided into a training cohort (N = 164) for ML model development and a validation cohort (N = 87). External validation was conducted on a subset of PPMI study patients (N = 65). We compared the performance of logistic regression (LR) and Support Vector Classifier (SVC) models trained on clinical data. The Shapley Additive exPlanations (SHAP) method was employed to examine the predictive power of individual variables.Results: In the training set, SVC outperformed LR slightly (AUC: LR = 0.779 ± 0.054, SVC = 0.792 ± 0.056). However, LR demonstrated better prediction accuracy in both internal (AUC: LR = 0.753, SVC = 0.733) and external validation cohorts (AUC: LR = 0.714, SVC = 0.676). SHAP analysis on the LR model revealed associations between fall risk and both motor and non‐motor variables.Conclusions: ML‐based models effectively estimate fall risk across different clinical centers, enabling tailored interventions to enhance PD patients' quality of life. Challenges persist in predicting falls in US‐based patients due to demographic and healthcare system differences.

Background: Predicting prognosis in people with multiple sclerosis (pwMS) at early disease stages still remains an unmet need. Machine learning (ML) strategies demonstrated good reliability when applied for prediction in medicine. This study aimed at developing a predictive algorithm comparing different ML approaches, by using routine demographic, clinical and radiological data from a large multicentric cohort of newly diagnosed pwMS. Methods: Demographic, clinical, radiological and biochemical data were retrospectively collected at three Italian MS centers at baseline and four timepoints thereafter (6, 12, 24, and 36 months). Data from the first evaluation and subsequent 2‐year follow‐up were analyzed, comparing different ML models (Random Forest, Extra Trees, XGBoost, Logistic Regression and Support Vector Classifier) to predict progression independent of relapse activity (PIRA) at year 3. To understand how features impacted the selected model's output, a ML explainability analysis was performed on the whole cohort and on specific subsets of patients, those aged under 45 and those NEDA‐3 at the 2‐year follow‐up. Results: Data from 719 pwMS (age 34.6 ± 11.2 years); female sex 501 (70%) were analyzed. Ninety‐two pwMS (13%) developed PIRA at year 3. Random Forest achieved the highest score, with a test set area under the ROC curve (AUC) of 0.75 ± 0.06. Features with the highest predictive impact were Expanded Disability Status Scale at 24 months, age at symptom onset and disease duration at baseline. Conclusion: Our results showed the feasibility of applying ML techniques to predict short‐term PIRA in newly diagnosed pwMS by using routine clinical practice data, paving the way for tailored and personalized approaches.

Background/Objectives: Neovascular age-related macular degeneration (nAMD) is a retinal disorder leading to irreversible central vision loss. The pro-re-nata (PRN) treatment for nAMD involves frequent intravitreal injections of anti-VEGF medications, placing a burden on patients and healthcare systems. Predicting injections needs at each monitoring session could optimize treatment outcomes and reduce unnecessary interventions. Methods: To achieve these aims, machine learning (ML) models were evaluated using different combinations of clinical variables, including retinal thickness and volume, best-corrected visual acuity, and features derived from macular optical coherence tomography (OCT). A “Leave Some Subjects Out” (LSSO) nested cross-validation approach ensured robust evaluation. Moreover, the SHapley Additive exPlanations (SHAP) analysis was employed to quantify the contribution of each feature to model predictions. Results: Results demonstrated that models incorporating both structural and functional features achieved high classification accuracy in predicting injection necessity (AUC = 0.747 ± 0.046, MCC = 0.541 ± 0.073). Moreover, the explainability analysis identified as key predictors both subretinal and intraretinal fluid, alongside central retinal thickness. Conclusions: These findings suggest that session-by-session prediction of injection needs in nAMD patients is feasible, even without processing the entire OCT image. The proposed ML framework has the potential to be integrated into routine clinical workflows, thereby optimizing nAMD therapeutic management.

Diabetic retinopathy (DR) screening commonly relies on fixed follow-up intervals, although progression risk differs across patients. We developed a preliminary image–clinical framework to support personalized follow-up recommendations from retinal fundus images and systemic risk factors. Public DR datasets were harmonized into a binary task distinguishing absence of DR from DR of any grade. An ImageNet-pretrained ResNet50 and a foundation model-based feature extraction pipeline were compared. The selected image model was integrated with literature-derived severe retinopathy progression curves and clinical risk modifiers to estimate personalized cumulative risk and assign follow-up intervals using a predefined acceptable risk threshold. The ResNet50-based model was selected for subsequent analyses. In the target cohort, the integrated model assigned 90.0% of patients to follow-up within 12 months. Clinical adjustment substantially modified image-only recommendations, generally shifting patients toward shorter intervals. These preliminary findings support the feasibility of combining image-derived estimates of baseline DR status with clinical modifiers to inform personalized screening intervals. Larger longitudinal studies are needed to validate calibration, clinical utility, and safety before real-world implementation.