Equity and feasibility of remote photoplethysmography for hypertension screening in darker-skinned populations in Nigeria: a multi-site field study
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective
To evaluate the feasibility, diagnostic accuracy, and acceptability of a remote photoplethysmography (rPPG) blood pressure screening tool among adults with Fitzpatrick skin types V and VI in Kebbi State, Nigeria, with additional assessment of the influence of facial tribal markings and internet connectivity.
Methods
We enrolled 306 patients and 30 staff across three hospitals. rPPG blood pressure estimates (Lifelight app) were compared with OMRON M2 cuff measurements (reference standard; 140/90 mmHg thresholds). Outcomes included the proportion of participants for whom the tool produced a blood pressure estimate, agreement (Bland-Altman), diagnostic accuracy (sensitivity/specificity), and patient/staff perceptions.
Results
The tool produced blood pressure estimates for 249 of 306 participants (81.4%), forming the analyzable index-test set. Agreement with the reference standard was limited (SBP MAE 15.37 mmHg, RMSE 19.93; DBP MAE 10.91 mmHg, RMSE 13.61). Within this set, sensitivity for hypertension detection was very low (SBP 0.04; DBP 0.10), with complete failure to detect elevated systolic blood pressure in Fitzpatrick type VI participants (sensitivity = 0.00). Specificity was higher (SBP 0.99; DBP 0.89). Performance was poorest in Fitzpatrick type VI. Lower bandwidth correlated with higher tool failure rates (maximum r = −0.69, minimum r = −0.51). Tribal markings were not associated with obtaining a reading, but subgroup accuracy differed (e.g., SBP bias −9.97 mmHg with marks). Despite these limitations, 70% of patients rated accuracy favourably and more than 90% of staff expressed willingness to adopt the tool, indicating a perception-performance gap.
Conclusion
Remote photoplethysmography was acceptable to patients and staff but is not recommended for population screening without further algorithmic and operational improvements. Despite high specificity (systolic 0.99), critically low sensitivity (systolic 0.04 overall, 0.00 for Fitzpatrick VI) means most hypertensive cases would be missed in screening programmes. Algorithmic refinement for darker skin tones and robust offline or low-bandwidth capability are prerequisites for equitable deployment in low-resource settings.
