We have shown that CRI derived from continuous, non-invasive algorithmic analysis of the pulse waveform in children and adults with severe dengue can predict the onset of shock. The CRI was able to predict an episode of re-shock within 12 h from measurement and predicted a narrowed pulse pressure (PP ≤ 20 mmHg)—a key diagnostic criterion of DSS—within 6 h (Table 2). We identified a CRI cutoff value of 0.4 as providing the best sensitivity and specificity, when distinguishing patients who developed re-shock from those that did not—a finding that held true up to 2 h prior to the event (Table 3, Additional file 1: Fig. S3 and Fig. S4). In our analyses, we show that the CRI had negative correlation with heart rate and systolic and diastolic blood pressure for the period of monitoring. As fluid leakage occurs in an individual, the CRI decrease is accompanied by a rise in heart rate, but also a rise in diastolic blood pressure due to compensatory vasoconstriction in the period prior to circulatory collapse. These results suggest that CRI could be a potentially helpful monitoring tool for the management of DSS in healthcare settings around the world, where obtaining contemporaneous vital signs, the assessment of intravascular volume of individual patients, and trending their responses to fluid resuscitation remain as huge challenges due to workload and limited healthcare resources.
The exaggerated inflammatory response in severe dengue results in a profound vascular leakage syndrome and shock [18]. The ability to provide judicious fluid therapy is therefore crucial and needs to be balanced with the risk of excessive fluid replacement, leading to complications of volume overload and respiratory distress [19]. Changes in conventional vital signs occur late in the setting of shock and lack specificity [8], and the use of cuff-based or invasive blood pressure monitoring techniques in severe dengue can result in bleeding as a result of profound thrombocytopenia. These modalities also require a level of healthcare expertise and infrastructure often unavailable in resource-limited settings—frequent, repeated vital signs measurements for patients in ward settings for example is often challenging. The CRI operates through photoplethysmography-based pulse waveform analysis—similar to pulse oximetry [8]. CRI has been clinically validated in severe trauma [20], and there is evidence that an individualized assessment of volume status is provided [21]. In our study, we adopted re-shock as our endpoint because it is a robust clinical endpoint that is widely used and clinically meaningful—the onset of shock prompts additional fluid replacement and interventions. Pulse pressure (PP) was used as a secondary endpoint, because a narrowed PP is a key diagnostic criterion for DSS, according to the WHO classification [7].
Traditional vital sign monitoring such as heart rate is an indicator of hemodynamic status, but can be affected by many factors, including temperature, stress, arrhythmias, and pain [7, 22], and the relative bradycardia observed in dengue patients could mask deterioration [23], so more robust ways of monitoring hemodynamic status are required. Other studies show that CRI is a more sensitive and specific indicator of decreased central blood volume status when compared to heart rate, blood pressure, SpO2, lactate, perfusion index, tissue oxygenation, etc. [8,9,10, 20, 24,25,26,27,28,29]. In dengue endemic settings such as Vietnam, it is essential that any technological intervention be robust, accepted by patients, cost-effective and adaptable to existing workflows—monitoring through PPG fulfills these criteria due to its non-invasive nature, relatively low costs, and staff familiarity with the use of pulse oximetry [30]. The use of CRI and other PPG-based monitoring devices could enable rapid risk-stratification and triage of patients most at risk, prior to overt clinical deterioration. In our study, patients were connected to a finger probe with data transferred wirelessly to a laptop. A standalone CRI monitor is FDA-cleared and commercially available; however, integration into bedside multi-parameter monitors would be more practical. The ability of the CRI to provide a continuous real-time assessment of cardiovascular status additionally could be promising for guiding individualized fluid management and allows for the development of novel algorithms which are dynamic and responsive to patient clinical state. We observed changes in CRI values in response to fluid therapy (Fig. 1), suggesting a benefit of CRI monitoring not only to detect shock but also to evaluate the response to fluid during resuscitation and guide further interventions. Work in coupling intrinsic connectivity [31] and automated alerts to the end-user could greatly improve caseload management in austere healthcare settings and in the context of epidemics [32]. This work currently forms an active area of research within our group [33].
The strengths of the study include its prospective design, the use of clinically relevant endpoints, and large patient sample size with the use of prolonged continuous CRI data to construct prediction models. Limitations include exclusion of enrolled patients because of inadequate or incomplete CRI data—this reflects the nature of PPG, which can be influenced by motion artifact and poor skin perfusion, which is relevant when monitoring patients in shock. A comparison with other measurements of fluid responsiveness, such as echocardiography, was not feasible at the time of this study but warrants investigation. We were also not able to directly compare the predictive performance of the CRI with that of other vital signs (such as heart rate and blood pressure) given the study design, in which the endpoint of shock used is itself defined by the pulse pressure. Ideally, a randomized head-to-head study allocating patients into different monitoring groups would allow for more robust analyses. Finally, as an observational study, we could not assess the role of CRI in terms of its actual clinical utility—we speculate that the automated and continuous nature of patient monitoring is scalable in healthcare settings and can provide useful information to guide clinicians in triage and treatment. The characterization of utility metrics in the healthcare system beyond its predictive value and establishing its role in patient management compared with standard practice is now needed.