MITOF: An Integrated AI Framework for Forecasting and Structured Reporting in Clinical Immunosuppressive Therapy Management

Abstract

This retrospective framework study presents Medical Integrated Therapy Optimisation and Forecasting (MITOF), an integrated clinical artificial-intelligence framework for immunosuppressive therapy in renal transplantation, in which patient risk stratification, dosage optimisation, dosage forecasting and structured clinical reporting are combined into a single workflow with role-based views for doctors, patients and hospital management. The framework was validated on Romanian retrospective clinical datasets, with most components grounded in a renal-transplant cohort of 22,885 transplant follow-up hospitalisation records between 2007 and 2022. The predictive components were validated retrospectively against recorded outcomes: the survivability classifier reached an accuracy of 92.9% and an area under the curve of 0.843, the LSTM forecaster outperformed an ARIMA baseline on irregular tacrolimus series, and the dosage model was most stable with the Huber loss and the Adam optimiser. We are careful about what each result establishes: the dosage model reproduces clinical practice rather than proven optimality, and the quality and usefulness of the generated report, together with the clinical value of the platform, will be evaluated through a planned clinician reader study and a prospective pilot. The main contribution is the integration itself, with role-based reporting and value-level traceability, validated module by module on Romanian transplant data. At the platform level, MITOF is organised around three clinical-AI requirements: clinical reviewability, role-specific abstraction and value-level provenance, so that model outputs remain inspectable rather than acting as autonomous decisions.