BRAIN Journal Interactive Companion

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

Medical Integrated Therapy Optimisation and Forecasting connects anonymised transplant records, retrospective AI modules, and structured role-based reports for immunosuppressive therapy review.

Authors: Rareș Arvinte, Diana Trandabăț, and Dan Cristea.

Take the article quiz

Source rows -> model outputs -> clinician review

Reviewable platform signalMeasured values, predictions, and generated explanations remain separate.

Article Console

Integrated support for transplant follow-up

MITOF addresses a practical clinical informatics problem: transplant follow-up evidence is spread across files, laboratory analyses, medication records, prescriptions, and short epicrisis texts.

01Retrospective cohort

22,885 transplant-cohort admissions

Four linked CSV datasets cover January 2007 to July 2022. After excluding non-day admissions, 22,885 kidney-transplant day-admission records supported dosage estimation, forecasting, and structured reporting.

02Three-layer model

Data, analysis, reporting

The data layer prepares records, the analysis layer creates reviewable signals, and the reporting layer adapts the same evidence for doctors, patients, and management. Research validation is a separate evaluation layer.

03Clinical AI rule

Assistive, not autonomous

Clinical reviewability, role-specific abstraction, and value-level provenance are design requirements. Platform-level usefulness remains a planned clinician-reader and pilot evaluation.

ElementReported detailSupported interpretation

Data scalePatients: 25,644 admission records; medication: 60,442 rows; analyses: 1,197,540 rows; prescriptions: 8,914 rows.Large linked hospital tables require preprocessing before modelling or reporting.

Mortality moduleHeld-out accuracy 92.9%, AUC 0.843, precision 0.87, recall 0.64, F1 0.738.Retrospective classifier performance exceeded the raw imbalanced baseline.

Infection moduleDBSCAN: silhouette 0.91, mapped-label AUC 0.465.Internal clusters did not align with the clinical toxin-presence target.

Report generatorStructured values are bound to source rows; GPT-4 was used as a comparator.Traceable templates are treated as authoritative; generative rewriting is optional only.

Experimental Pathway

From hospital rows to reviewable clinical reports

Knowledge Atlas

MITOF concept atlas

Explore how linked clinical data, AI modules, user roles, and validation requirements are connected inside the framework.

Evidence Modules

Interactive module evidence

Select a MITOF component to inspect its data source, validation setup, and evidence status. Values are shown only where the study reports them explicitly.

Article Quiz

Test your MITOF knowledge

Answer ten questions covering the framework, datasets, validation results, reporting design, and clinical boundaries described in the article.

0/10 questions Score: 0