Non-invasive near-infrared measurement of blood concentration in chest tube output: a proof-of-concept study

Background: Chest tube management requires assessing drainage volume and blood content, but only volume can be measured objectively at the bedside; visual estimation of blood concentration is unreliable. We developed and evaluated a low-cost near-infrared spectroscopy (NIRS) device for non-invasive measurement of blood concentration in chest tube output. Methods: An Arduino-based prototype with six light-emitting diode wavelengths (660, 740, 765, 810, 850, 940 nm), an OPT101 photodiode, and a 3D-printed enclosure was constructed. Bovine blood was diluted with normal saline into ten concentrations (0–90%, 10% increments), with five randomized measurements per dilution (n=50). Single-wavelength exponential, two-wavelength composite, and six-wavelength multivariable models were fitted. Each measurement was assigned to the calibrated level for which the fitted signal lay closest to the observed value. Performance was assessed as apparent and cross-validated coefficient of determination, classification accuracy within one dilution level, and Bland-Altman agreement, with cluster bootstrap confidence intervals. Results: Within-tube coefficients of variation had a median of 15% (interquartile range 8% to 26%). The 660 nm wavelength showed the strongest relationship with blood proportion (signal-direction R²=0.89). In the prediction direction, the 660 nm model achieved R²=0.796 apparent and 0.741 under leave-one-tube-out cross-validation, whereas six-wavelength regression achieved the highest apparent value (0.804) but fell to 0.697 when cross-validated. Classification accuracy within one dilution level was 74% (95% CI 60%;84%) under all three evaluation schemes, and 68% for the composite. Bland-Altman analysis showed a bias of 2.0% (95% CI −2.2;4.6), with 95% limits of agreement of −23.4% to 27.4% relative to ground-truth. Conclusions: This study demonstrates the feasibility of a low-cost NIRS device for non-invasive estimation of blood concentration in chest tube output. The wide limits of agreement indicate that the device supports discrimination between low and high blood content rather than point estimation, pending validation in clinical drainage samples.