Does LLM Assistance Improve Healthcare Delivery? An Evaluation Using On-site Physicians and Laboratory Tests∗

We deployed large language model (LLM) decision support for health workers at two outpatient clinics in Nigeria. For each patient, health workers drafted care plans that were optionally revised after LLM feedback. We compared unassisted and assisted plans using blinded randomized assessments by on-site physicians who evaluated and treated the same patients and using results from laboratory tests for common conditions. Academic physicians performed blinded retrospective reviews of a subset of notes. In response to LLM feedback, health workers changed their prescribing for more than half of patients. Health workers reported high satisfaction with LLM feedback and retrospective academic reviewers rated LLM-assisted plans more favorably. However, on-site physicians observed little to no improvement in diagnostic alignment or treatment decisions. Laboratory testing showed mixed effects of LLM-assistance, which removed negative tests for malaria but added them for urinary tract infection and anemia, with no significant increase in the detection rates for the tested conditions. This highlights a gap between chart-based reviews and real-world clinical relevance that may be especially important in evaluating the effectiveness of LLM-based interventions.