RESUMO
Sepsis is a major public health emergency and one of the leading causes of morbidity and mortality in critically ill patients. For each hour treatment is delayed, shock-related mortality increases, so early diagnosis and intervention is of utmost importance. However, earlier recognition of shock requires active monitoring, which may be delayed due to subclinical manifestations of the disease at the early phase of onset. Machine learning systems can increase timely detection of shock onset by exploiting complex interactions among continuous physiological waveforms. We use a dataset consisting of high-resolution physiological waveforms from intensive care unit (ICU) of a tertiary hospital system. We investigate the use of mean arterial blood pressure (MAP), pulse arrival time (PAT), heart rate variability (HRV), and heart rate (HR) for the early prediction of shock onset. Using only five minutes of the aforementioned vital signals from 239 ICU patients, our developed models can accurately predict septic shock onset 6 to 36 hours prior to clinical recognition with area under the receiver operating characteristic (AUROC) of 0.84 and 0.8 respectively. This work lays foundations for a robust, efficient, accurate and early prediction of septic shock onset which may help clinicians in their decision-making processes. This study introduces machine learning models that provide fast and accurate predictions of septic shock onset times up to 36 hours in advance. BP, PAT and HR dynamics can independently predict septic shock onset with a look-back period of only 5 mins.
RESUMO
Hypovolemic shock is one of the leading causes of death in the military. The current methods of assessing hypovolemia in field settings rely on a clinician assessment of vital signs, which is an unreliable assessment of hypovolemia severity. These methods often detect hypovolemia when interventional methods are ineffective. Therefore, there is a need to develop real-time sensing methods for the early detection of hypovolemia. Previously, our group developed a random-forest model that successfully estimated absolute blood-volume status (ABVS) from noninvasive wearable sensor data for a porcine model (n = 6). However, this model required normalizing ABVS data using individual baseline data, which may not be present in crisis situations where a wearable sensor might be placed on a patient by the attending clinician. We address this barrier by examining seven individual baseline-free normalization techniques. Using a feature-specific global mean from the ABVS and an external dataset for normalization demonstrated similar performance metrics compared to no normalization (normalization: R2 = 0.82 ± 0.025|0.80 ± 0.032, AUC = 0.86 ± 5.5 × 10-3|0.86 ± 0.013, RMSE = 28.30 ± 0.63%|27.68 ± 0.80%; no normalization: R2 = 0.81 ± 0.045, AUC = 0.86 ± 8.9 × 10-3, RMSE = 28.89 ± 0.84%). This demonstrates that normalization may not be required and develops a foundation for individual baseline-free ABVS prediction.