Understanding vital-sign abnormalities in critical care patients

Introduction

We have previously shown how a data fusion index, based on the integration of vital signs continuously monitored in a step-down unit (SDU), can identify the cardiorespiratory instability that often precedes adverse events [1]. We show how the correlations between vital signs, captured in a data fusion index based on a probabilistic model of normality [2], change during times of cardiorespiratory instability.

Methods

An observational study was carried out in a 24-bed SDU, in which vital signs from 326 patients were continuously recorded [1]. The existing standard of care used single-channel Medical Emergency Team (MET) activation criteria to determine when individual vital signs became abnormal. Retrospective evaluation of the continuous vital sign data identified that MET criteria were exceeded on 238 nonartefactual occasions. One hundred and eleven of these were indicative of sufficient cardiorespiratory instability that they should have required an MET call (critical events). Vital-sign dynamics during these events were compared with data from stable periods, using order statistics and covariance analysis.

Results

We found the probability distributions of the most extreme and the median data for each parameter over 1-minute intervals. The bivariate Gaussian distributions of best fit for data from critical events (abnormal respiratory rate (RR) or heart rate (HR)) show significant differences in covariance when compared with those calculated for data intervals from stable patients (see Figure 1, in which covariance is indicated by ellipse orientation). Under normal conditions (black ellipses), RR is correlated with HR. During critical events, tachycardia occurs with little or no variation in RR (red ellipses); tachypnoea occurs with little or no variation in HR (blue ellipses).

Figure 1

Conclusions

Dynamics and correlations that exist in vital signs during periods of stability change significantly during periods of abnormality. A statistical approach that integrates vital signs and captures correlations between them will be sensitive to cardiorespiratory deterioration.