JMIR Publications

ABSTRACT

Electronic Medical Records (EMR) can promote healthcare delivery, research, and policy. To maximize benefits, EMR must be complete, accurate, and ensure high data quality. The Light Wave Health Information Management System (LHIMS) is an electronic-health platform implemented by Ghana’s Ministry of Health. As part of this initiative in transitioning to paperless records, in September 2021, Korle Bu Teaching Hospital (KBTH) extended LHIMS to its Neonatal Intensive Care Unit (NICU) - a leading NICU in Ghana. Since implementation at KBTH, no validation studies have been done. This study assessed the completeness, validity, concordance and accuracy of LHIMS EMR data at the KBTH NICU, nine months post implementation. This retrospective study compared EMR and paper-based records (PBR) of 222 newborns randomly selected from the 1,025 NICU admissions from July 1 to December 31, 2022. Data analysis was performed using Python (version 3.10) with pandas, numpy, and scipy libraries. A predefined set of 24 clinical variables relevant to NICU stay were established and analyzed across four domains (i.e. delivery, maternal, neonatal, and admission data). Four metrics were evaluated - completeness, validity, concordance and accuracy. EMR data across all 24 variables was less complete compared to PBR, 37.0% versus 81.3% respectively (p-value <0.01), except for one variable: insurance status was 91.4% complete in EMR and 22.5% complete in PBR (p-value <0.01). Maternal data was most vulnerable among the four domains for completeness, 2.5% for EMR versus 69.8% for PBR. A similar trend was observed for validity: 36.2% for EMR and 80.5% for PBR (p value <0.01); although it improved when missing data was excluded - 89.2% for EMR and 98.7% for PBR (p value = 0.02). EMR concordance and accuracy were 42.6% and 31.4%, respectively. When missing values were included in the analysis, EMR data was inaccurate but significantly accurate when missing values were excluded. Again, maternal data was most affected by missing values: 1.9% accuracy with missing values versus 95.8% without missing values. Some critical neonatal variables such as birth weight, admission temperature and admission random blood sugar had very low EMR completeness (9.5%, 1.4% and 3.6%, respectively). Insurance status and sex of the baby, on the other hand, had very high EMR completeness (91.4% and 98.7%, respectively). The KBTH NICU EMR were generally less complete and less accurate compared with PBR. When missing values were excluded, EMR had higher validity and higher concordance with PBR, indicating that EMR accuracy was markedly affected by the high degree of missing entries. The newborn EMR for the study period was of low data quality and unreliable for policy decisions and clinical research. Improvement in EMR data entry protocols and collaboration between the hospital technology team and NICU health workers could enhance data quality.