Dataset and Code Repository for: Offline-First Mobile OCR and FHIR Interoperability for Bedside Vital Sign Digitisation in Geriatric Care under Power and Network Constraints: A Ghanaian District Hospital Study

## 📌 Repository Overview This repository contains the complete, live-verified computational datasets, statistical workspaces, image archives, and Python execution pipelines supporting the manuscript for **Preprint P2** under the Sovereign Clinical OS (SCOS) research framework.  The study evaluates an offline-first mobile Optical Character Recognition (OCR) pipeline and an automated, local Large Language Model (Llama-3-8B) usability judge for bedside vital sign digitization in low-resource district hospitals (e.g., Tatale District Hospital, Northern Ghana). ## 2. Dataset Descriptions ### `P2_master_analytic_dataset.csv` ($N = 250$)The primary master analytic dataset generated from the Tatale geriatric cohort in `P1_master_analytic_aligned.csv`. Contains patient demographics (Age $\ge 60$, Gender), ground-truth VitalDB values, parameter-specific absolute extraction errors, character error rates (CER), processing latencies, FHIR compliance flags, and LOINC mapping indicators. ### `P2_ocr_performance_metrics.csv` ($N = 250$)Contains on-device execution times (`latency_ms`), raw Character Error Rates (`cer`), and Mean Absolute Errors (`_ae`) per vital parameter evaluated across 250 simulated bedside clinical chart captures. It includes manual keyboard entry speed benchmarks (`manual_entry_sec`) versus mobile OCR entry speed (`ocr_entry_sec`). ### `P2_fhir_conformance_metrics.csv` ($N = 250$)Contains binary validation flags (`is_fhir_compliant`, `loinc_mapped_correct`) demonstrating whether each digitized vital-sign payload successfully passed through a local instance of the HAPI FHIR R4 Validator and was correctly bound to standard LOINC codes. ### `P2_usability_validation_metrics.csv` ($N = 250$)Benchmarks the usability scores graded by a local Llama-3-8B model acting as an automated usability judge (`llama3_usability_score`) against expert human ratings from the gold-standard UXBench corpus (`gold_standard_score`), including defect detection flags (`llama3_defect_detected`). ### `P2_offline_sync_metrics.csv` ($N = 50$)Models 50 disconnected clinical sessions, tracking simulated offline blackout durations (`offline_duration_hr`), queued record counts (`queued_records`), data preservation success rates (`preservation_rate`), and local-to-server sync latencies (`sync_latency_sec`) upon network restoration. ---