[Modelling critical information measurement traumatic surgery decisions. "Sequential Information Appraisal Module...

We studied the quality and quantity of information leading to the emergency physician's decision to intubate severely injured patients on scene. Our aim was to assess intuitive aspects of clinical decision making. The experiment involved three different phases, with a fourth phase examining retest reliability. We used trauma register data from 98 patients. Based on various parameters (physiological data, injury assessment on scene, definite injury pattern), three emergency surgeons were requested to decide on the need for endotracheal intubation.We applied multivariate logistic regression to estimate the likelihood of intubation given certain clinical characteristics or combinations of characteristics. We compared the participants' decisions to those made by "true" emergency physicians on scene. Kappa statistics marked inter-observer agreement beyond chance. The Glasgow Coma Scale (GCS) was the only single predictor of intubation in the ideal test setting (area under the receiver operating characteristics curve [AUC] >98%) as well as on scene (AUC 0.85, 95% confidence interval 0.78-0.92). There was no difference between the discriminatory features of the single item GCS and complex multivariate models that included anatomically defined injury scales (best model in phase 2: AUC 0.96, best model in phase 3: AUC 0.98). Overall inter-observer agreement was substantial in phase 1 (kappa=0.74), fair to moderate in phase 2 (kappa=0.49) and slight to fair in phase 3 (kappa=0.23). Retest reliability ranged from 51% to 91%. Doctors give priority to only a small part of the information available in deciding for or against a particular intervention.