3 Beyond 15mm: towards a new personalised, AI measured definition of hypertrophic cardiomyopathy using an age, sex, and body surface area -adjusted approach

Introduction

Hypertrophic cardiomyopathy (HCM) is characterised by unexplained left ventricular hypertrophy (LVH). Current diagnostic criteria require ≥15mm maximal wall thickness (MWT). However, MWT is influenced by age, sex, ethnicity and anthropometrics, meaning current “one size fits all” approach is likely inaccurate and biased (e.g., underdiagnosing smaller females or some ethnicities). To define normal and therefore redefine pathology using superhuman AI approaches to measure MWT adjusted for age, sex, and body surface area (BSA).

Materials and Methods

CMRs from 4118 healthy UK Biobank participants were analysed (mean age 61.4, 49.6% male, 1.83 BSA) Analysis was automated using previously published AI algorithms (Augusto et al., 2020) that a) exceed human performance for test:retest measurement. Age and BSA were used as independent variables in sex-specific multiple linear regression models. Limits of normality were calculated (mean+2SD) of the residual as a percentage of the predicted MWT. 258 healthy subjects recruited locally and 149 genotype-positive HCM patients were used to validate the model. Sensitivity and specificity were calculated.

Results

The model r² for MWT was 0.39 – ie 40% of all MWT variation is explained by age, sex and body size (rather than individual variation/disease). The upper limit MWT is calculated as males: 1.299*(3.90*BSA+0.03*Age+2.06) and females: 1.249*(3.20*BSA+0.07*Age-0.34). Over a representative population (UK Biobank healthy volunteers), the 15mm threshold was >1mm too high in 58% and >1mm too low in 19% and appropriate in just 23% of cases. For overt HCM patients (mean MWT 21.3+/-5.1) using the new individualised MWT cut-off value, sensitivity is preserved (91%, 136/149). For UK Biobank subjects, 4% (162/4118) are classified as abnormal with 4%(10/258) of a second hold-out healthy volunteer set giving a sensitivity of 91% and specificity of 96%.

Conclusion

The current “one size fits all” 15mm cut point for abnormal LVH is only appropriate in one in four of the population and biased. Using superhuman AI for wall thickness, we propose an age, sex and BSA adjusted MWT to overcome these that overcomes this whilst preserving sensitivity and specificity. Further refinement may include athleticism, comorbidity and ethnicity.

Acknowledgements

Augusto, J. B., Davies, R. H., Bhuva, A. N., Knott, K. D., Seraphim, A., Alfarih, M., Lau, C., Hughes, R. K., Lopes, L. R., Shiwani, H., Treibel, T. A., Gerber, B. L., Hamilton-Craig, C., Ntusi, N. A. B., Pontone, G., Desai, M. Y., Greenwood, J. P., Swoboda, P. P., Captur, G., … Moon, J. C. (2020). Diagnosis and risk stratification in hypertrophic cardiomyopathy using machine learning wall thickness measurement: a comparison with human test-retest performance. The Lancet Digital Health, 0(0). https://doi.org/10.1016/S2589–7500(20)30267–3.