Digitising HIV Testing Registers to Support National Scale-Up of Three-Test HIV Diagnostic Algorithm: A Case Study of an AI-Powered Monitoring and Evaluation System in Malawi

BackgroundBy November 2022, Malawi became one of the first countries to implement the 2019 World Health Organization (WHO) recommended three-test HIV algorithm nationally. To support the scale-up, a new monitoring and evaluation (M&E) system, ScanForm, was introduced to digitise individual-level records from paper-based HIV testing registers using artificial intelligence (AI). We describe the national scale-up of the ScanForm M&E system alongside the three-test algorithm and assess the performance of the M&E system in monitoring quality assurance and generating routine program reporting. MethodsWe conducted a descriptive study using routinely collected HIV testing data captured through ScanForm between November 2022 and September 2025. HIV testing providers photographed completed handwritten register pages, which were automatically transcribed using AI, validated and summarised into reports. The system performance was assessed by evaluating changes over time in numbers of recording errors and HIV testing services (HTS) protocol deviations detected in the first image submitted per register page, data completeness, and reporting timeliness. Trends in recording errors and protocol deviations were analysed among facilities with at least 24 months of follow-up, using each facilitys activation date as the baseline. ResultsBy January 2023, 260 (30%) of 867 HTS facilities had adopted the M&E system and the three-test algorithm. By September 2024, 98% coverage nationwide was achieved (853/867 facilities). During the study period, 9,082,891 HTS encounters were captured through ScanForm and HIV positivity was 1.8%. A total of 1,412,133 errors and deviations were identified, representing an overall error rate of 0.6%. Of these, 1,370,002 (97%) were recording errors and 42,131 (3%) were deviations in testing protocol. Only two types of deviations were specific to the implementation of the three-test HIV algorithm: 15,051 (36%) were deviations from the HTS testing algorithm, and 4,683 (11%) were misclassifications of HIV test results. Errors and deviations declined by 32% in the first three months and by 58% over 24 months, while the error rate dropped from 1.6% to 0.4% over 24 months. After resolving the validation checks, 99.8% of all mandatory data elements were complete. Approximately 94.6% of all HTS records were submitted on time. ConclusionThe national rollout of the AI-powered M&E system was rapid, achieving national coverage within two years, and the system provided timely and complete program data. Errors and deviations declined during implementation, indicating improved data quality and adherence to the HTS protocol. Integrating digital data systems into routine service delivery has the potential to strengthen guideline implementation and enhance the quality of HIV services at scale.