Innovative Paper-to-Digital Solutions in Kenyan Hospitals: A Pathway to Enhanced Clinical Data Collection
In the realm of healthcare, particularly in environments with limited electronic health record (EHR) adoption, the collection and reporting of patient data remain challenging. A recent study has explored a promising hybrid paper-to-digital approach that could significantly alleviate these challenges, specifically within neonatal units of Kenyan hospitals. This novel method not only reduces the time and effort involved but also enriches the volume and variety of patient-level clinical data available in health information systems.
Methods and Analysis: Designing a Hybrid Solution
The project employed human-centered design principles, engaging nurses, physicians, and records officers from neonatal units across eight geographically diverse Kenyan hospitals. The aim was to design and implement two key components: (1) machine-readable versions of patient transfer, admission, and discharge forms, and (2) a minimally viable paper-to-digital pipeline for clinical data extraction. The design phase spanned from February to October 2024, with testing conducted between November 2024 and May 2025.
During the testing phase, all newborn admissions at the participating hospitals used the new machine-readable forms. The pipeline utilized artificial intelligence (AI) models to automatically extract clinical data from these forms, with continuous evaluation and updates to the models to ensure accuracy and efficiency.
Results: Enhanced Efficiency and Accuracy
The results were promising. Of the 7,118 patients admitted to the neonatal wards, 91.1% had at least one designed form in their medical records. A diverse group of 1,615 physicians utilized the forms, demonstrating widespread acceptance and usability. The clinical data pipeline exhibited high accuracy in extracting checkbox fields, with an AUCROC of 99.25%, an F1 score of 98.91%, a positive predictive value of 98.85%, and sensitivity of 99.02%. For free text fields, the accuracy was 95.66% with a character error rate of 1.66% and a word error rate of 4.34%.
Overall, the hybrid approach reduced the time spent on manual data extraction by 50% and decreased the time on forms requiring data correction by 60%, highlighting the significant efficiency gains achieved.
Conclusion: A Scalable Solution for Sub-Saharan Africa
This co-developed paper-to-digital pipeline represents a scalable solution for generating patient-level data in routine clinical settings, particularly in regions like sub-Saharan Africa where EHRs are often inaccessible or challenging to implement. The study underscores the potential of AI-based extraction technologies in enhancing healthcare data systems, offering a model that could be replicated in similar contexts worldwide.
For further details, access the full study Here.
“`
