Evaluating Multimodal LLMs for Inpatient Diagnosis: Real-World Performance, Safety, and Cost Across Ten Frontier Models
Authors: Bruce A. Bassett, Amy Rouillard, Sitwala Mundia, Michael Cameron Gramanie, Linda Camara, Ziyaad Dangor, Shabir A. Madhi, Kajal Morar, +3 more
Organizations: School of Computer Science and Applied Mathematics, University of the Witwatersrand, Johannesburg, South Africa · Wits MIND Institute, University of the Witwatersrand, Johannesburg, South Africa · Grai Labs, Cape Town, South Africa · South African Medical Research Council Vaccines and Infectious Diseases Analytics Research Unit, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa · Department of Internal Medicine, Charlotte Maxeke Johannesburg Academic Hospital, South Africa · Department of Paediatrics and Child Health, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa
Abstract
Background: Large language models (LLMs) are increasingly proposed for diagnostic support, but few evaluations use real-world multimodal inpatient data, particularly in low and middle-income country (LMIC) public hospitals. Methods: We conducted VALID, a retrospective evaluation of 539 multimodal inpatient cases from a tertiary public hospital in South Africa. Inputs included radiology imaging (CT, MRI, CXR) and reports, laboratory results, clinical notes, and vital signs. Expert panels adjudicated 300 cases (balanced and discordant subsets) to establish ground truth diagnoses, differentials, and reasoning. Ten multimodal LLMs generated zero-shot outputs. A calibrated three-model LLM Jury scored all outputs and routine ward diagnoses across diagnostic accuracy, differential quality, reasoning, and patient safety (>10,000 evaluations). Primary outcomes were composite scores (S3, S4) and win rates. Results: (i) LLM performance was tightly clustered (<15% variation) despite large cost differences; low-cost models performed comparably to top models. (ii) All LLMs significantly outperformed routine ward diagnoses on average diagnostic and safety scores. (iii) Top performance was achieved by GPT-5.1, followed by Gemini models. (vi) Adding radiology reports improved performance by 6%. (v) Diagnostic and reasoning scores were highly correlated (ρ=0.85). (vi) Output rates varied (65-100%) due to input constraints. Results were robust across subsets and evaluation design. Conclusions: Across a real-world LMIC dataset, multimodal LLMs showed similar diagnostic performance despite large cost differences and outperformed routine care on average safety metrics. Affordability, robustness, and deployment constraints may outweigh marginal performance differences in LMIC settings.
Clinical diagnostic evaluation should not only assess whether models can provide correct diagnoses, but also reflect the realities of clinical practice, including progressive disclosure of multimodal information, dynamic updating of diagnostic hypotheses, and continuous refinement of clinical reasoning. However, existing evaluations of multimodal large language models (MLLMs) typically rely on single-turn or isolated tasks, making it difficult to fully capture the complexity of real-world clinical diagnosis. To bridge this gap, we developed ClinMM-Bench, the largest multi-turn multimodal clinical diagnostic evaluation benchmark to date. ClinMM-Bench contains 1,089 challenging real-world clinical cases and 3,760 medical images across eight specialties. We systematically evaluated 15 representative MLLMs using a two-level evaluation framework that assessed both diagnostic accuracy and diagnostic reasoning quality. Results showed that proprietary models achieved the highest overall diagnostic accuracy, but the proportion of completely correct diagnoses remained limited across all models. In terms of diagnostic reasoning quality, current models can identify plausible diagnostic directions but still have considerable limitations in generating reliable diagnostic reasoning. Error analysis further identified five representative failure modes: information synthesis failure, knowledge mapping error, perception error, premature closure, and visual hallucination.
Evaluating medical AI systems using expert clinician panels is costly and slow, motivating the use of large language models (LLMs) as alternative adjudicators. Here, we evaluate an LLM Jury, composed of three frontier AI models, for scoring 3334 diagnoses on 300 real-world low- and middle-income country (LMIC) hospital cases. Both LLM- and clinician-generated diagnoses are scored against expert panel diagnoses across four dimensions: diagnosis, differential diagnosis, clinical reasoning, and negative treatment risk. The LLM Jury scores are compared with expert and independent re-scoring panel scores to assess error metrics, inter-rater agreement, severe-risk errors, and the effect of post hoc calibration using isotonic regression. In our data, we find that: (i) the uncalibrated LLM Jury scores preserve ordinal agreement with the expert clinician panel scores, but are systematically lower; (ii) the probability of severe-risk errors is lower for the LLM Jury than the human expert re-score panels; (iii) the LLM Jury combined with LLM diagnoses can be used to identify diagnoses at high risk of error, enabling targeted expert review and improved panel efficiency; (iv) the calibrated LLM Jury scores and rankings of diagnosing agents show excellent agreement with those of the primary expert panels; (v) LLM Jury models show no self-preference bias, they did not score diagnoses generated by their own underlying model or models from the same vendor more (or less) favourably than those generated by other models. Together, these results provide evidence that a calibrated LLM Jury is a trustworthy and reliable proxy for expert clinician evaluation in medical AI benchmarking. Confirming these findings in other clinical settings is an important direction for future work.
Multimodal large language models (MLLMs) have demonstrated promise on publicly available dermatology benchmarks. However, benchmark performance may not generalize to real-world dermatologic decision-making. To quantify this benchmark-to-bedside gap, we evaluated four open-weight MLLMs (InternVL-Chat v1.5, LLaVA-Med v1.5, SkinGPT4 and MedGemma-4B-Instruct) and one commercial MLLM (GPT-4.1) across three publicly available dermatology datasets and a retrospective multi-site hospital-based dermatology consultation cohort comprising 5,811 cases and 46,405 clinical images. Models were evaluated on two clinically relevant tasks: differential diagnosis generation and severity-based triage. Diagnostic performance was modest on public datasets and declined substantially in the real-world cohort. On public benchmarks, top-3 diagnostic accuracy reached 26.55% for the best open-weight model and 42.25% for GPT-4.1. On real-world consultation cases using images alone, top-3 diagnostic accuracy fell to 1.50%-13.35% among open-weight models and 24.65% for GPT-4.1. Incorporating clinical context improved performance across all models, increasing top-3 diagnostic accuracy up to 28.75% among open-weight models and 38.93% for GPT-4.1. However, model outputs were highly sensitive to incomplete or erroneous consultation context. For severity-based triage, models achieved moderate sensitivity (above 60%), suggesting potential utility for screening but insufficient reliability for clinical deployment. These findings demonstrate that benchmark performance substantially overestimates the real-world clinical capability of current dermatology MLLMs.