Predicting need for hospital admission in patients with traumatic brain injury or skull fractures identified on CT imaging: a machine learning approach

Abstract

BackgroundPatients with mild traumatic brain injury on CT scan are routinely admitted for inpatient observation. Only a small proportion of patients require clinical intervention. We recently developed a decision rule using traditional statistical techniques that found neurologically intact patients with isolated simple skull fractures or single bleeds <5 mm with no preinjury antiplatelet or anticoagulant use may be safely discharged from the emergency department. The decision rule achieved a sensitivity of 99.5% (95% CI 98.1% to 99.9%) and specificity of 7.4% (95% CI 6.0% to 9.1%) to clinical deterioration. We aimed to transparently report a machine learning approach to assess if predictive accuracy could be improved.MethodsWe used data from the same retrospective cohort of 1699 initial Glasgow Coma Scale (GCS) 13–15 patients with injuries identified by CT who presented to three English Major Trauma Centres between 2010 and 2017 as in our original study. We assessed the ability of machine learning to predict the same composite outcome measure of deterioration (indicating need for hospital admission). Predictive models were built using gradient boosted decision trees which consisted of an ensemble of decision trees to optimise model performance.ResultsThe final algorithm reported a mean positive predictive value of 29%, mean negative predictive value of 94%, mean area under the curve (C-statistic) of 0.75, mean sensitivity of 99% and mean specificity of 7%. As with logistic regression, GCS, severity and number of brain injuries were found to be important predictors of deterioration.ConclusionWe found no clear advantages over the traditional prediction methods, although the models were, effectively, developed using a smaller data set, due to the need to divide it into training, calibration and validation sets. Future research should focus on developing models that provide clear advantages over existing classical techniques in predicting outcomes in this population.

Enhancing hospital course and outcome prediction in patients with traumatic brain injury: A machine learning study

“…Furthermore, there are studies describing models to predict the functional outcome or The Glasgow Outcome Scale-Extended, 5,[25][26][27][28][29] and more recent studies using images as inputs. 8,30,31 Furthermore, there are studies in the literature similar to ours that describe models for predicting in-hospital mortality, 29,[32][33][34][35] early mortality, [36][37][38] discharge position, 39,40 need for hospital admission, 6 emergency neurosurgery, 41 and length of hospital stay. 4 In addition to contributing to the body of knowledge by describing the efficacy of incorporating ML into patient care to predict multiple outcomes simultaneously in TBI patients, this study is unique since it has used blood biomarkers such as GFAP and UCH-L1, and non-contrast CT CDEs as input variables.…”

“…Research has been conducted to generate predictor variables, methods, and models for enhancing the precision of prediction of outcomes following a TBI, which aids in treatment decisions and the management of expectations. [4][5][6][7][8][9][10] The Glasgow Coma Scale (GCS) has been used to promptly classify the severity of TBI for decades and it correlates with patient mortality and morbidity. 11 However, GCS is subject to interobserver variation and poorly correlates with mortality and morbidity at the favorable end of the spectrum.…”

Pre-hospital prediction of adverse outcomes in patients with suspected COVID-19: Development, application and comparison of machine learning and deep learning methods

“…This cut-off was associated with NPV and PPV values of 99% and 17% respectively. In developing the PRIEST tool, Marincowitz et al [40] selected a predicted probability threshold that led to high NPV (this also implies high sensitivity), but with a relatively high PPV (i.e., at least 96.5% NPV and a minimum PPV of 28%). These restrictions were associated with a sensitivity of 99% but the specificity was reduced to 7%.…”

Building artificial intelligence and machine learning models : a primer for emergency physicians

“…In addition, as more open access ML libraries (containing ML algorithms and other related code), such as scikit-learn.org and CRAN-R packages (cran.r-project.org) and even automated ML (AutoML) become freely available, the temptation is to apply ML to a problem when traditional solutions work well or are more suitable. (4)(5)(6) The current clinical status quo should be examined, and any proposed improvement in its performance considered in terms of potential clinically significant patient or system benefit. A priori determination of an ideally patient centred (or shared) minimal acceptable difference in outcome should be made.…”