Skin tone and diagnostic equity in contactless blood pressure screening: a prospective observational field evaluation of remote photoplethysmography in Nigeria.

摘要

OBJECTIVES: To evaluate diagnostic equity, feasibility and acceptability of a remote photoplethysmography-based blood pressure screening application among adults with darker skin tones in Nigeria. DESIGN: Prospective observational multisite field evaluation. SETTING: Three hospitals in Kebbi State, Nigeria. PARTICIPANTS: Adults with Fitzpatrick skin types V-VI. OUTCOME MEASURES: Feasibility, agreement, diagnostic accuracy, acceptability, and equity relevant factors including facial tribal markings and internet bandwidth, using automated cuff measurements as the reference standard and a 140 over 90 mm Hg hypertension threshold. RESULTS: Among 306 enrolled participants, 249 (81.4%) produced usable readings. Agreement was poor (systolic mean absolute error (MAE) 15.4 mm Hg, root mean square error (RMSE) 19.9; diastolic MAE 10.9 mm Hg, RMSE 13.6). Sensitivity for threshold-based systolic and diastolic blood pressure classification was very low (systolic 0.04; diastolic 0.10), with systolic sensitivity 0.00 in Fitzpatrick type VI. Specificity was high (systolic 0.99; diastolic 0.89). Lower internet bandwidth correlated with reading failure (r = -0.69 to -0.51). While 70% of patients and over 90% of staff rated the tool favourably, technical limitations created a clear perception-performance gap. In an exploratory interaction analysis, Fitzpatrick type VI was associated with higher odds of measurement failure (OR 5.08, 95% CI 2.41 to 10.72), but there was no clear evidence that facial tribal markings modified this association (interaction OR 0.66, 95% CI 0.16 to 2.73; p=0.564). CONCLUSIONS: Remote photoplethysmography (rPPG)-based blood pressure screening was feasible but showed inadequate performance in this darker-skinned field cohort, with critically low sensitivity. Without algorithmic recalibration for skin tone diversity and improved offline functionality, cloud-dependent rPPG systems deployed without spectrum-balanced validation may risk exacerbating diagnostic inequities in similar settings.