Clinical validation of a machine-learned, point-of-care system to IDENTIFY pulmonary hypertension

Abstract

Background Pulmonary hypertension (PH) is a collection of diverse disorders, defined by mean pulmonary artery pressure (mPAP) ⩾21 mmHg (most recent guidelines) or ⩾25 mmHg (previous guidelines, that underpins the field’s past work) measured by right heart catheterisation (RHC). Considering the difficulties in diagnosing PH and the subsequent treatment delays, there is a need for novel diagnostics to enable prompt detection. Methods An algorithm to assess mPAP elevation was validated using subjects with elevated mPAP from RHC (positive cohort) and subjects with low probability of PH by stringent screening of transthoracic echocardiography (TTE) PH indicators (negative cohort). 25 mmHg and 21 mmHg were pre-specified as the co-primary and secondary sensitivity end-points, respectively, at 0.70. Specificity was the co-primary end-point at 0.60. The algorithm cut-point was pre-defined. The area under the receiver operator characteristic curve (ROC-AUC) was assessed at both mPAP thresholds. Findings 462 subjects were consecutively enrolled across 18 US clinical sites between August 2019 and September 2022. Sensitivity at 25 mmHg and 21 mmHg was 0.82 (95% CI 0.78–0.87) and 0.78 (95% CI 0.73–0.82), respectively, with specificity of 0.92 (95% CI 0.87–0.96), passing the study end-points. The ROC-AUC values at 25 mmHg and 21 mmHg were 0.95 (95% CI 0.93–0.96) and 0.93 (95% CI 0.91–0.95), respectively. Further, performance was similar across PH subgroups (pre-capillary, combined pre-and post-capillary, and isolated post-capillary), as well as between men and women. Interpretation The algorithm’s performance is comparable, or possibly superior to, TTE, given that the tricuspid regurgitant velocity is not measurable in up to 41% of TTE cases. The test is a stress-free, noninvasive front-line test, presenting advantages to patients, physicians and healthcare systems.