The relationship between patient data and pooled clinical management decisions.

A strong relationship between patient data and preoperative clinical decisions could potentially be used to support clinical decisions in preoperative management. The aim of this exploratory study was to determine the relationship between key patient data and pooled clinical opinions on management. In a previous study, panels of anaesthetists compared the quality of computer-assisted patient health assessments with outpatient consultations and made decisions on the need for preoperative tests, no preoperative outpatient assessment, possible postoperative intensive care unit/high dependency unit requirements and aspiration prophylaxis. In the current study, the relationship between patient data and these decisions was examined using binomial logistic regression analysis. Backward stepwise regression was used to identify independent predictors of each decision (at P >0.15), which were then incorporated into a predictive model. The number of factors related to each decision varied: blood picture (four factors), biochemistry (six factors), coagulation studies (three factors), electrocardiography (eight factors), chest X-ray (seven factors), preoperative outpatient assessment (17 factors), intensive care unit requirement (eight factors) and aspiration prophylaxis (one factor). The factor types also varied, but included surgical complexity, age, gender, number of medications or comorbidities, body mass index, hypertension, central nervous system condition, heart disease, sleep apnoea, smoking, persistent pain and stroke. Models based on these relationships usually demonstrated good sensitivity and specificity, with receiver operating characteristics in the following areas under curve: blood picture (0.75), biochemistry (0.86), coagulation studies (0.71), electrocardiography (0.90), chest X-ray (0.85), outpatient assessment (0.85), postoperative intensive care unit requirement (0.88) and aspiration prophylaxis (0.85). These initial results suggest modelling of patient data may have utility supporting clinicians' preoperative decisions.