Non-contact fingertip video to differentiate dengue haemorrhagic fever in plasma leakage from dengue fever: A prospective proof-of-concept cohort study

Most patients with dengue recover without intervention, but a minority develop plasma leakage that threatens the circulation, and differentiating them is the central problem in dengue care. Platelet count is the marker in routine serial use and is itself one of the four defining criteria of dengue haemorrhagic fever (DHF), so thrombocytopenia in dengue fever (DF) can overlap with DHF on the very marker used to distinguish them. We asked whether a non-contact optical signal carries discriminative information beyond platelet count. We conducted a prospective observational cohort study at a national dengue referral centre. All DHF recordings were obtained at or after plasma-leakage onset, so the analysis compares established clinical states rather than pre-leakage risk. Serial non-contact fingertip videos were recorded with contemporaneous clinical and haematological measurements. In a held-out set of 50 recordings (20 DHF, 30 DF), video-derived probability separated the groups with an area under the receiver operating characteristic curve (AUC) of 0.963 (95% CI 0.918–1.000; 0.957 averaged across eight architectures). In the 49 recordings with a platelet count, the video AUC was 0.961 against 0.888 for platelet count (difference 0.074, p = 0.19). In exploratory analyses the video added information to platelet count (likelihood-ratio p < 0.001), showed no significant within-group correlation with platelet count, and retained an AUC of 0.970 against thrombocytopenic DF, where platelet count fell to 0.759. Bedside pulse rate, body temperature and haematocrit did not separate the groups and were unrelated to the video score. A ~30-second fingertip video separated DHF in plasma leakage from DF in this test set, and exploratory analyses suggest it carries information beyond platelet count. These findings are proof of concept from single-centre internal validation, not evidence of pre-leakage prediction; prospective pre-leakage studies and external validation are needed to establish whether the signal can support earlier risk stratification.