Diagnostic Accuracy of the National Library of Medicine (NLM) Screener Application in Screening for Malaria Parasites Among Blood Donors at the Korle-Bu Blood Bank, Ghana: A Cross-Sectional Study

Background and Aims: Malaria is a deadly disease spread through the bite of an infected female Anopheles mosquito and remains the leading cause of mortality globally. Screening donated blood for malaria parasites is essential to prevent its transmission; however, conventional methods have limitations in accuracy and sensitivity. The National Library of Medicine (NLM) mobile application that uses machine learning algorithms to detect malaria parasites in blood smears could reduce some of these limitations. Hence, its performance in different settings needs to be evaluated. This study aimed to evaluate the performance of the NLM screener in screening malaria parasites against microscopy, rapid diagnostic tests (RDTs), and polymerase chain reaction (PCR) among blood donors at the Korle-Bu Teaching Hospital (KBTH) in Accra, Ghana. Methods: We conducted a cross-sectional study of 300 blood donors at the KBTH in Ghana. Each donor sample was tested with PCR (reference standard), microscopy, RDT, and the NLM screener app. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Cohen’s κ were calculated with 95% confidence intervals (CIs). Agreement and paired outcomes were assessed with McNemar’s exact test. Results: PCR identified Plasmodium falciparum in 18/300 donors (6.0%). The NLM screener app showed sensitivity of 38.9% (7/18; 95% CI: 20.3–61.4), specificity of 60.6% (171/282; 95% CI: 54.8–66.2), PPV of 5.9% (7/118; 95% CI: 2.9–11.7), and NPV of 94.0% (171/182; 95% CI: 89.5–96.6), with negligible agreement (κ = −0.001). RDT and microscopy had lower sensitivities (44.4% and 27.8%, respectively) but perfect specificity (100%). Conclusion: The NLM screener app demonstrated low diagnostic performance in this setting. Applied to our donor pool, it would have led to 111 unnecessary discards (false positives) and 11 missed infections (false negatives). While promising, the app requires substantial improvement and validation before consideration for clinical use in transfusion safety programs.