Abstract Background: H-scoring is a semi-quantitative immunohistochemical scoring method used to evaluate protein expression intensity. Scoring intensities can then provide prognostic staining thresholds that may help to explain differences in cancer outcomes. However, traditional H-scoring may be biased by inter-observer variability, in particular for tumors with heterogeneous morphology. Digital pathology in conjunction with machine learning applications offer a potential solution for standardizing H-scoring in histologically complex tumors. Here, we describe challenges and propose best practices for using the open-source digital pathology application QuPath to standardize tumor H-scoring in the setting of cancer health disparities research. Methods: Tissue microarrays (TMAs) were generated from 26 paraffin-embedded tissues of advanced stage laryngeal cancer patients (15 Black and 11 White patients). Triplicate cores were sampled. Immunohistochemistry (IHC) of ABCC1 (MRP1 antibody) protein expression was performed using the ImmPACT® DAB Substrate Kit. IHC-stained and corresponding hematoxylin and eosin (H 0.00001). Conclusion: In summary, standardizing tumor H scores in morphologic heterogenous tumors using digital pathology and machine learning tools requires: 1. consulting with a pathologist to accurately annotate tumor and non-tumor areas, 2. developing staining intensity thresholds that account for negative and intense staining variations, 3. validating classifiers and thresholds across multiple users to ensure H-score correlation. We believe implementing these practices will markedly reduce artifacts and bias, enhance the reliability of tumor H-scoring, and will provide objective and accurate detection of protein expression differences across racial groups. MC, CG (equal contribution) Citation Format: Matthew S. Chang, Christina M. Gobin, Jai L. Walker, Kristianna M. Fredenburg. An approach to H-scoring tumors with morphologic heterogeneity using digital pathology and machine learning abstract. In: Proceedings of the 18th AACR Conference on the Science of Cancer Health Disparities; 2025 Sep 18-21; Baltimore, MD. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2025;34(9 Suppl):Abstract nr B008.
Chang et al. (Thu,) studied this question.