PPIcheck.ai
Gamma_Built 3.0
India's First Indigenously-Built AI Tool for Optimizing PPI Therapy
Merging Clinical Expertise with Intelligent Precision
PPIcheck.ai is revolutionizing the way clinicians make PPI deprescribing decisions.
Our application combines established clinical guidelines with cutting-edge machine learning to provide evidence-based risk assessments, enabling healthcare professionals to optimize PPI therapy effectively.
Model Development
PPIcheck – Deprescriber Guide is a clinical decision support tool developed to aid in rational PPI prescribing and deprescribing decisions. This application employs a hybrid model, integrating established clinical guidelines with machine learning algorithms, ensuring a balanced approach between evidence-based medicine and data-driven insights.
It draws upon authoritative references, including Lexicomp drug interaction tools and data from the CONFOR trial, enabling comprehensive and contextual risk assessment.
The results demonstrated high predictive accuracy, with excellent precision and reliability across key clinical indicators, confirming the model’s potential in guiding appropriate PPI use and deprescribing decisions.
Computational Approach
Our application utilizes a data-driven computational algorithm integrating clinical input variables with advanced machine learning classifiers such as Random Forest and Logistic Regression. The system processes risk factors, drug dosage patterns, and clinical indications to compute a composite risk score. This score is then classified using optimized models trained on synthetic and validated datasets. The approach ensures high accuracy, reproducibility, and clinical relevance by combining domain-specific scoring logic with robust algorithmic decision-making.
PPI Risk Model: Performance Assessment
To enhance model performance, the app was first trained using a synthetic dataset of 10,000 cases, designed to reflect the demographic and prescription patterns of our population. Following this, real-world clinical data from 400 patients was collected. These were split into training and testing sets (50:50) using K-Fold cross-validation, ensuring robust evaluation.
A synthetic dataset of 20,000 entries has been acquired from Central Institutes of India for advanced model development and validation.
Our clinical prediction app demonstrates exceptional performance, with Random Forest and Logistic Regression models achieving near-perfect AUCs of 0.9999 and 0.9966, respectively. The Random Forest model shows outstanding accuracy (99.33%), precision (98.97%), recall (99.74%), and F1 score (99.35%), while Logistic Regression also performs robustly across all metrics. These results confirm the app’s high precision, reliability, and potential for effective use in real-world clinical settings.
Licensed Under the MIT License – Open, Free, and Ready to Use
This project is licensed under the MIT License— a permissive open-source license that allows anyone to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software, as long as the original license and copyright notice are included.
