BACKGROUND: The heterogeneous nature of traumatic brain injury (TBI) makes outcome prediction difficult. Although a considerable evidence base exists in the form of well-validated predictive models, these models are not widely used. We hypothesised that this prognostic gap, between the availability and use of prognostic data, leads to inaccurate perceptions of patient outcome. We investigated whether outcome predictions in TBI made by expert clinicians were consistent and accurate when compared to a well-validated prognostic model (IMPACT). METHODS: Neurosurgeons and neurointensivists were asked to predict probability of death at 6 months for 12 case vignettes describing patients with isolated TBI. Predictions were compared to IMPACT prognosis for each vignette. To interrogate potential sources of bias in clinical predictions, respondents were given one of two sets of vignettes (A or B) identical apart from one critical factor known to make a large difference to outcome. RESULTS: 27 of 33 questionnaires were returned. Clinicians were consistently more pessimistic about outcomes than the IMPACT model, predicting a significantly greater probability of death (mean difference + 16.3%, 95% CI 13.3-19.4, p < 0.001). There was wide variation between clinicians predicting outcomes for any given vignette (mean range 68.3%), and within the predictions made by each individual: 30% of clinicians were both the most pessimistic respondent, and the most optimistic, for at least one vignette. Clinicians modified their predictions appropriately for most of the factors altered between corresponding vignettes. However when the reported blood glucose was changed, clinicians' predictions deviated widely from IMPACT predictions, indicating that clinicians systematically overlooked the prognostic relevance of this information. CONCLUSION: Clinical experts' predictions of outcome in TBI are widely variable and systematically pessimistic compared to IMPACT. Clinicians overlook important factors in formulating these predictions. Use of well-validated outcome models may add value and consistency to prognostication.
BACKGROUND: The heterogeneous nature of traumatic brain injury (TBI) makes outcome prediction difficult. Although a considerable evidence base exists in the form of well-validated predictive models, these models are not widely used. We hypothesised that this prognostic gap, between the availability and use of prognostic data, leads to inaccurate perceptions of patient outcome. We investigated whether outcome predictions in TBI made by expert clinicians were consistent and accurate when compared to a well-validated prognostic model (IMPACT). METHODS: Neurosurgeons and neurointensivists were asked to predict probability of death at 6 months for 12 case vignettes describing patients with isolated TBI. Predictions were compared to IMPACT prognosis for each vignette. To interrogate potential sources of bias in clinical predictions, respondents were given one of two sets of vignettes (A or B) identical apart from one critical factor known to make a large difference to outcome. RESULTS: 27 of 33 questionnaires were returned. Clinicians were consistently more pessimistic about outcomes than the IMPACT model, predicting a significantly greater probability of death (mean difference + 16.3%, 95% CI 13.3-19.4, p < 0.001). There was wide variation between clinicians predicting outcomes for any given vignette (mean range 68.3%), and within the predictions made by each individual: 30% of clinicians were both the most pessimistic respondent, and the most optimistic, for at least one vignette. Clinicians modified their predictions appropriately for most of the factors altered between corresponding vignettes. However when the reported blood glucose was changed, clinicians' predictions deviated widely from IMPACT predictions, indicating that clinicians systematically overlooked the prognostic relevance of this information. CONCLUSION: Clinical experts' predictions of outcome in TBI are widely variable and systematically pessimistic compared to IMPACT. Clinicians overlook important factors in formulating these predictions. Use of well-validated outcome models may add value and consistency to prognostication.
Authors: Adrian M Svingos; Steven A Robicsek; Ronald L Hayes; Kevin K Wang; Claudia S Robertson; Gretchen M Brophy; Linda Papa; Andrea Gabrielli; H Julia Hannay; Russell M Bauer; Shelley C Heaton Journal: Neurocrit Care Date: 2022-03-01 Impact factor: 3.532
Authors: Cristian Minoccheri; Craig A Williamson; Mark Hemmila; Kevin Ward; Erica B Stein; Jonathan Gryak; Kayvan Najarian Journal: BMC Med Inform Decis Mak Date: 2022-08-01 Impact factor: 3.298