diff --git a/README.md b/README.md index 7d9d86a..ecd87f4 100644 --- a/README.md +++ b/README.md @@ -19,11 +19,11 @@ The code on this repository is commented throughout to provide a description of ### Background Existing methods to characterise the evolving condition of traumatic brain injury (TBI) patients in the intensive care unit (ICU) do not capture the context necessary for individualising treatment. We aimed to develop a modelling strategy which integrates all heterogenous data stored in medical records to produce an interpretable disease course for each TBI patient’s ICU stay. ### Methods -From a prospective cohort (_n_=1,550, 65 centres, 19 countries) of European TBI patients, we extracted all 1,166 variables collected before or during ICU stay as well as six-month functional outcome on the Glasgow Outcome Scale – Extended (GOSE). We trained recurrent neural network models to map a token-embedded time series representation of all variables to an ordinal GOSE prognosis every two hours. With 20 repeats of five-fold cross-validation, we evaluated calibration and the explanation of ordinal variance in GOSE with Somers’ _Dxy_. Furthermore, we implemented the TimeSHAP algorithm to calculate the contribution of variables and prior timepoints towards significant transitions in patient trajectories. +From a prospective cohort (n=1,550, 65 centres, 19 countries) of European TBI patients, we extracted all 1,166 variables collected before or during ICU stay as well as six-month functional outcome on the Glasgow Outcome Scale – Extended (GOSE). We trained recurrent neural network models to map a token-embedded time series representation of all variables to an ordinal GOSE prognosis every two hours. With repeated cross-validation, we evaluated calibration and the explanation of ordinal variance in GOSE with Somers’ Dxy. Furthermore, we implemented the TimeSHAP algorithm to calculate the contribution of variables and prior timepoints towards transitions in patient trajectories. ### Findings -Our modelling strategy achieved calibration at eight hours post-admission, and the full range of variables explained up to 52·2% (95% CI: 50·2%–54·3%) of the variance in ordinal, six-month functional outcome. Most of this explanation was derived from pre-ICU information. Information collected during ICU stay increased explanation, though not enough to counter the difficulty of characterising longer-stay patients. Static variables with the highest contributions were physician-based prognoses and certain demographic and CT features. Among dynamic variables, markers of intracranial hypertension and neurological function contributed the most. +Our modelling strategy achieved calibration at eight hours post-admission, and the full range of variables explained up to 52·2% (95% CI: 50·2%–54·3%) of the variance in ordinal functional outcome. Most of this explanation was derived from pre-ICU and admission information. Information collected in the ICU increased explanation (by up to 5·2% [95% CI: 4·2%–6·2%]), though not enough to counter poorer overall performance in longer-stay (>5·75 days) patients. Static variables with the highest contributions were physician-based prognoses and certain demographic and CT features. Among dynamic variables, markers of intracranial hypertension and neurological function contributed the most. ### Interpretation -We show the feasibility of a data-driven approach for individualised TBI characterisation without the need for variable curation, cleaning, or imputation. Our results also highlight potential investigative avenues to help explain the remaining half of variance in functional outcome. +We show the feasibility of a data-driven approach for individualised TBI characterisation with minimal pre-processing and integration of missing data. Our results also highlight investigative avenues to help explain the remaining half of variance in functional outcome. ### Funding NIHR Brain Injury MedTech Co-operative, EU 7th Framework, Hannelore Kohl, OneMind, Integra Neurosciences, Gates Cambridge.