ABSTRACT
Introduction: Activity monitoring is important in the ICU where delirium, sedation, and critical illness are associated with both inactivity and agitation. Staff monitoring of motion and sleep is intermittent and resource intense. Wearable actigraphic devices are poorly tolerated and limited to limb motion. Here we demonstrate continuous AI video monitoring in the ICU to provide alwayson, unobtrusive patient activity monitoring. Methods: We conducted a pilot study of AI video monitoring in the Duke University Hospital Medical Intensive Care Unit. Video carts continuously recorded data on encrypted hard drives. Second-by-second AI analysis generated binary motion “counts” that were summed to generate our patient motion metric: counts per minute (CPM). Scene intelligence from AI object and people detectors provided room environment information. These data streams along with de-identified (blurred) video data were used to generate prototype graphical and visual summaries of patient activity patterns and the hospital room environment. Results: We enrolled 22 patients and collected 2155 hours (116 days) of video. Representative time-series data streams are shown in the Figure (top left). These data were acquired from a 76-year-old with liver failure and an escalating nasal cannula oxygen requirement who was endotracheally intubated on the subsequent day. Note 1) the declining patient activity as the patient deteriorates and 2) the significant bedside activity (high acuity) throughout the day. We developed a prototype “overnight report” that summarizes patient activity and room environment. The Figure (bottom left) shows the overnight report for a 54-year-old post-COVID-19 patient admitted to the MICU for respiratory failure with hypoactive delirium that resolved per CAM-ICU on day 5 of data collection. Notably, our report demonstrates significant overnight movement, possibly consistent with a mixed or hyperactive delirium. To visually summarize patient motion, we generated activity “heat maps” over 10-minute intervals. As a control, we showed that the intubated and sedated liver failure patient generated a still heat map (Figure upper right). Further, we visualized daytime hypoactivity/sleep in the delirious post-COVID patient (Figure lower right), suggesting disrupted circadian rhythm, giving additional context to the negative CAM assessment. Conclusions: We demonstrated the feasibility of AI to monitor patient activity in a quaternary-care MICU. Our method has advantages compared to wearable actigraphic methods for monitoring patient activity, including being unobtrusive and being able to visualize and summarize wholebody motion. The data presented here suggest that such monitoring may be able to provide clinically actionable insights in delirium care and sedation weaning.
ABSTRACT
Rationale: Inhaled nitric oxide (iNO) has been used for several years as an adjunctive therapy for improving oxygenation in adults with acute respiratory distress syndrome (ARDS). Recently, several authors have suggested iNO as a useful therapy in the setting of ARDS secondary to COVID-19. Nevertheless, there remains unclear evidence regarding the utility of iNO in adults with both COVIDand non-COVID-associated ARDS, and still less evidence regarding who might benefit from this costly treatment. We sought to investigate the effect of iNO on oxygenation in adults with ARDS secondary to both COVID and non-COVID etiologies, evaluate the difference in outcomes between patients with and without COVID receiving iNO, and explore the cost associated with the administration of iNO in an intensive care setting. Methods: We conducted a retrospective cohort study in the medical and surgical intensive care units at a tertiary-care academic medical center. All patients with ARDS who received iNO over a two-year period were considered for inclusion. Exclusion criteria included prior pulmonary hypertension already on a pulmonary vasodilator, initiation of iNO prior to arrival at our institution, or lack of an arterial blood gas immediately before and after the initiation of iNO. Outcomes measured included change in PaO2/FiO2, 30-day mortality, and the cost of iNO administration. Results: 177 consecutive patients were evaluated, of whom 108 met criteria for inclusion. Change in PaO2/FiO2 ratio following iNO administration was significantly smaller in patients with COVID than in patients with non-COVID ARDS (22.9% vs 60.4%, p = 0.002). Among COVID patients there was no significant improvement in PaO2/FiO2 following the administration of iNO (95% CI [-64%, 108%]). A response in PaO2/FiO2 (defined as >10% increase) was not associated with 30-day mortality (p = 0.29). The average cost of iNO administration among all patients was $66597.79, and there was a trend toward greater cost in patients deemed P/F responders ($76433 vs $53195, p = 0.07). There was no difference in these outcomes in patients receiving iNO for refractory hypoxia versus patients receiving iNO for RV dysfunction. Conclusions: In this study, iNO administration incurred an average cost of $66597.79 per patient and showed no association with improved PaO2/FiO2 ratio in patients with COVIDARDS. PaO2/FiO2 changes in COVID patients were significantly smaller than in non-COVID patients. An increase in PaO2/FiO2 > 10% was a poor predictor of 30-day mortality but did show a trend toward increased cost burden.