Paper-based POC platform enables rapid whole-blood PT/INR and haematocrit monitoring with AI classification
The study reports a hybrid paper-based point-of-care platform designed for simultaneous whole-blood PT/INR and haematocrit analysis in decentralized settings. Systematic optimization of reagent ratios, substrate types, and strip geometry yielded strong analytical performance, with haematocrit estimation achieving an R² of 0.9788 and PT/INR measurement reaching an R² of approximately 0.94. Support vector machine classifiers integrated into a companion application achieved diagnostic accuracies of 98.5% for non-anticoagulated and 96.9% for anticoagulated cohorts. The low-cost, smartphone-compatible system offers a practical solution for accessible anticoagulation monitoring and warfarin dose management outside centralised laboratories.
The original study
Design and development of a smart, intuitive and cost-efficient paper-based platform for whole blood PT/INR diagnostics.
- Authors
- Saha A, Bajpai A, Shukla P, Verma R, Krishna VK, Bhattacharya S
- Journal
- Lab on a chip
- Type
- Journal Article
- PMID
- 42638485
Original abstract
Frequent monitoring of prothrombin time and international normalized ratio (PT/INR) is essential for effective management of anticoagulation therapy, yet conventional laboratory-based testing remains costly, time-consuming, and difficult to access in decentralized settings. Addressing this challenge, a hybrid paper-based point-of-care (POC) platform is designed for rapid and affordable whole-blood PT/INR analysis. The assay is developed through systematic optimization of key parameters, including sample volume, blood-to-reagent ratio, reaction time, substrate type, assay geometry, and strip dimensions for both haematocrit (HCT) and PT/INR modules. The integrated system simultaneously estimates HCT using radial intensity-based detection on Whatman Grade 4 paper (R2 = 0.9788) and measures PT/INR on high-porosity MF1 glass-fibre strips (R2 ≈ 0.94), enabling correction of HCT-induced bias. Image-derived features were analyzed using AI/ML-based models, where support vector machine classifiers achieved diagnostic accuracies of 98.5% and 96.9% for non-anticoagulated and anticoagulated cohorts, respectively. Integrated with a custom desktop/web/smartphone application and an automated warfarin dosage calculator, the platform provides a low-cost, rapid, and user-friendly solution for accessible anticoagulation monitoring.