Confusion matrices to refine a novel scoring system for cardiovascular instability in intensive care

Cardiovascular instability is common in intensive care. It can be as a result of the primary disease process or secondary to therapeutic interventions e.g. sedation. It is often described in the literature yet is poorly defined and difficult to quantify. We have devised novel qualitative and quantitative scores for cardiovascular instability. Charting of clinical information at Glasgow Royal Infirmary is now done electronically. This allows the analysis of significant quantities of physiological data at frequent intervals on a scale which would previously have been impractical. We have devised a numerical rule base which subsequently categorises patients from A (stable) to E (highly unstable) according to the degree of derangement of a range of physiological parameters taking into account the amount of sedation or inotropic support.1 The parameters include heart rate, mean arterial pressure, central venous pressure, inspired oxygen concentration and oxygen saturation. A computer programme has been designed to interrogate a patient record of parameters and predict a level of stability based on the rule system. Time points are hourly throughout the patient stay in intensive care. Clinicians (without any knowledge of the numerical rules) have been asked to score the same patient record from A to E at the same time points. The results have been incorporated into a confusion matrix which compares the scores given by the clinician with the scores predicted by the computer programme. A diagonal line from top left to bottom right in an example of a matrix (see figure 1) represents complete agreement between the predicted score and the expert clinical opinion. Time points where discrepancy exists are subsequently analysed and if they occur frequently then a refinement is made to the rule base and the data set interrogated again the aim being to produce a rule base which can as accurately as possible predict the expert opinion. When validated this will be able to be used at the bedside to give an indication as to the cardiovascular stability of a patient in intensive care.