The leap to ordinal: Detailed functional prognosis after traumatic brain injury with a flexible modelling approach

High Priority Traumatic Brain Injury Science: Analysis of the National Trauma Research Action Plan

INTRODUCTION

The National Trauma Research Action Plan convened 11 topic area panels to complete consensus-driven Delphi surveys to identify high priority trauma research questions. The Neurotrauma Panel identified questions relating to interventional and comparative effectiveness trials in severe traumatic brain injury (sTBI) critical care as highest priority. This qualitative secondary analysis aims to translate results across several Delphi panels into potential studies in sTBI critical care.

METHODS

High priority consensus research questions related to sTBI in the critical phase of care (ranked >6.5 on a 1-9 Likert scale) were screened from the Neurotrauma, Critical Care, Geriatric, and Long-Term Outcomes Panels results. Using grounded theory, two reviewers inductively open-coded questions independently and then refined them for consensus. A similar approach was used to recategorize questions into codes. Each code was then characterized into research project(s) with an aim, design, exposure(s), and outcome(s).

RESULTS

Among 376 high-priority questions reaching consensus, 55 related to sTBI critical care. Twelve projects emerged across eight consensus thematic codes: biomarkers (1 project, average priority score/range 6.92), imaging (1, 6.84), prognostication (1, 6.77), novel neuromonitoring (3, 6.61-6.77), intracranial pressure/cerebral perfusion pressure (2, 6.67-6.76), coagulopathy (2, 6.66-6.74), early rehabilitation (1, 6.67), and pharmacologic intervention (1, 6.66).

CONCLUSIONS

This National Trauma Research Action Plan secondary analysis identified several high-priority research projects in sTBI critical care. While some questions are being addressed in ongoing trials, investigators and funding agencies should consider using these consensus-driven Delphi panel results and subsequent analyses to prioritize future research proposals.

TILTomorrow today: dynamic factors predicting changes in intracranial pressure treatment intensity after traumatic brain injury

Practices for controlling intracranial pressure (ICP) in traumatic brain injury (TBI) patients admitted to the intensive care unit (ICU) vary considerably between centres. To help understand the rational basis for such variance in care, this study aims to identify the patient-level predictors of changes in ICP management. We extracted all heterogeneous data (2008 pre-ICU and ICU variables) collected from a prospective cohort (n = 844, 51 ICUs) of ICP-monitored TBI patients in the Collaborative European NeuroTrauma Effectiveness Research in TBI study. We developed the TILTomorrow modelling strategy, which leverages recurrent neural networks to map a token-embedded time series representation of all variables (including missing values) to an ordinal, dynamic prediction of the following day's five-category therapy intensity level (TIL(Basic)) score. With 20 repeats of fivefold cross-validation, we trained TILTomorrow on different variable sets and applied the TimeSHAP (temporal extension of SHapley Additive exPlanations) algorithm to estimate variable contributions towards predictions of next-day changes in TIL(Basic). Based on Somers' Dxy, the full range of variables explained 68% (95% CI 65-72%) of the ordinal variation in next-day changes in TIL(Basic) on day one and up to 51% (95% CI 45-56%) thereafter, when changes in TIL(Basic) became less frequent. Up to 81% (95% CI 78-85%) of this explanation could be derived from non-treatment variables (i.e., markers of pathophysiology and injury severity), but the prior trajectory of ICU management significantly improved prediction of future de-escalations in ICP-targeted treatment. Whilst there was no significant difference in the predictive discriminability (i.e., area under receiver operating characteristic curve) between next-day escalations (0.80 [95% CI 0.77-0.84]) and de-escalations (0.79 [95% CI 0.76-0.82]) in TIL(Basic) after day two, we found specific predictor effects to be more robust with de-escalations. The most important predictors of day-to-day changes in ICP management included preceding treatments, age, space-occupying lesions, ICP, metabolic derangements, and neurological function. Serial protein biomarkers were also important and may serve a useful role in the clinical armamentarium for assessing therapeutic needs. Approximately half of the ordinal variation in day-to-day changes in TIL(Basic) after day two remained unexplained, underscoring the significant contribution of unmeasured factors or clinicians' personal preferences in ICP treatment. At the same time, specific dynamic markers of pathophysiology associated strongly with changes in treatment intensity and, upon mechanistic investigation, may improve the timing and personalised targeting of future care.

A Model Predicting the 6-Month Disability of Patients With Traumatic Brain Injury to Assess the Quality of Care in Intensive Care Units: Results from the CREACTIVE Study

Assessing quality of care is essential for improving the management of patients experiencing traumatic brain injury (TBI). This study aimed at devising a rigorous framework to evaluate the quality of TBI care provided by intensive care units (ICUs) and applying it to the Collaborative Research on Acute Traumatic Brain Injury in Intensive Care Medicine in Europe (CREACTIVE) consortium, which involved 83 ICUs from seven countries. The performance of the centers was assessed in terms of patients' outcomes, as measured by the 6-month Glasgow Outcome Scale-Extended (GOS-E). To account for the between-center differences in the characteristics of the admitted patients, we developed a multinomial logistic regression model estimating the probability of a four-level categorization of the GOS-E: good recovery (GR), moderate disability (MD), severe disability (SD), and death or vegetative state (D/VS). A total of 5928 patients admitted to the participating ICUs between March 2014 and March 2019 were analyzed. The model included 11 predictors and demonstrated good discrimination (area under the receiver operating characteristic [ROC] curve in the validation set for GR: 0.836, MD: 0.802, SD: 0.706, D/VS: 0.890) and calibration, both overall (Hosmer-Lemeshow test p value: 0.87) and in several subgroups, defined by prognostically relevant variables. The model was used as a benchmark for assessing quality of care by comparing the observed number of patients experiencing GR, MD, SD, and D/VS to the corresponding numbers expected in each category by the model, computing observed/expected (O/E) ratios. The four center-specific ratios were assembled with polar representations and used to provide a multidimensional assessment of the ICUs, overcoming the loss of information consequent to the traditional dichotomizations of the outcome in TBI research. The proposed framework can help in identifying strengths and weaknesses of current TBI care, triggering the changes that are necessary to improve patient outcomes.

Development of a Multimodal Machine Learning-Based Prognostication Model for Traumatic Brain Injury Using Clinical Data and Computed Tomography Scans: A CENTER-TBI and CINTER-TBI Study

Computed tomography (CT) is an important imaging modality for guiding prognostication in patients with traumatic brain injury (TBI). However, because of the specialized expertise necessary, timely and dependable TBI prognostication based on CT imaging remains challenging. This study aimed to enhance the efficiency and reliability of TBI prognostication by employing machine learning (ML) techniques on CT images. A retrospective analysis was conducted on the Collaborative European NeuroTrauma Effectiveness Research in TBI (CENTER-TBI) data set (n = 1016). An ML-driven binary classifier was developed to predict favorable or unfavorable outcomes at 6 months post-injury. The prognostic performance was assessed using the area under the curve (AUC) over fivefold cross-validation and compared with conventional models that depend on clinical variables and CT scoring systems. An external validation was performed using the Comparative Indian Neurotrauma Effectiveness Research in Traumatic Brain Injury (CINTER-TBI) data set (n = 348). The developed model achieved superior performance without the necessity for manual CT assessments (AUC = 0.846 [95% CI: 0.843-0.849]) compared with the model based on the clinical and laboratory variables (AUC = 0.817 [95% CI: 0.814-0.820]) and established CT scoring systems requiring manual interpretations (AUC = 0.829 [95% CI: 0.826-0.832] for Marshall and 0.838 [95% CI: 0.835-0.841] for International Mission for Prognosis and Analysis of Clinical Trials in TBI [IMPACT]). The external validation demonstrated the prognostic capacity of the developed model to be significantly better (AUC = 0.859 [95% CI: 0.857-0.862]) than the model using clinical variables (AUC = 0.809 [95% CI: 0.798-0.820]). This study established an ML-based model that provides efficient and reliable TBI prognosis based on CT scans, with potential implications for earlier intervention and improved patient outcomes.

Refining outcome prediction after traumatic brain injury with machine learning algorithms

Outcome after traumatic brain injury (TBI) is typically assessed using the Glasgow outcome scale extended (GOSE) with levels from 1 (death) to 8 (upper good recovery). Outcome prediction has classically been dichotomized into either dead/alive or favorable/unfavorable outcome. Binary outcome prediction models limit the possibility of detecting subtle yet significant improvements. We set out to explore different machine learning methods with the purpose of mapping their predictions to the full 8 grade scale GOSE following TBI. The models were set up using the variables: age, GCS-motor score, pupillary reaction, and Marshall CT score. For model setup and internal validation, a total of 866 patients could be included. For external validation, a cohort of 369 patients were included from Leuven, Belgium, and a cohort of 573 patients from the US multi-center ProTECT III study. Our findings indicate that proportional odds logistic regression (POLR), random forest regression, and a neural network model achieved accuracy values of 0.3-0.35 when applied to internal data, compared to the random baseline which is 0.125 for eight categories. The models demonstrated satisfactory performance during external validation in the data from Leuven, however, their performance were not satisfactory when applied to the ProTECT III dataset.

Therapy Intensity Level Scale for Traumatic Brain Injury: Clinimetric Assessment on Neuro-Monitored Patients Across 52 European Intensive Care Units

Intracranial pressure (ICP) data from traumatic brain injury (TBI) patients in the intensive care unit (ICU) cannot be interpreted appropriately without accounting for the effect of administered therapy intensity level (TIL) on ICP. A 15-point scale was originally proposed in 1987 to quantify the hourly intensity of ICP-targeted treatment. This scale was subsequently modified-through expert consensus-during the development of TBI Common Data Elements to address statistical limitations and improve usability. The latest 38-point scale (hereafter referred to as TIL) permits integrated scoring for a 24-h period and has a five-category, condensed version (TIL(Basic)) based on qualitative assessment. Here, we perform a total- and component-score analysis of TIL and TIL(Basic) to: 1) validate the scales across the wide variation in contemporary ICP management; 2) compare their performance against that of predecessors; and 3) derive guidelines for proper scale use. From the observational Collaborative European NeuroTrauma Effectiveness Research in TBI (CENTER-TBI) study, we extract clinical data from a prospective cohort of ICP-monitored TBI patients (n = 873) from 52 ICUs across 19 countries. We calculate daily TIL and TIL(Basic) scores (TIL24 and TIL(Basic)24, respectively) from each patient's first week of ICU stay. We also calculate summary TIL and TIL(Basic) scores by taking the first-week maximum (TILmax and TIL(Basic)max) and first-week median (TILmedian and TIL(Basic)median) of TIL24 and TIL(Basic)24 scores for each patient. We find that, across all measures of construct and criterion validity, the latest TIL scale performs significantly greater than or similarly to all alternative scales (including TIL(Basic)) and integrates the widest range of modern ICP treatments. TILmedian outperforms both TILmax and summarized ICP values in detecting refractory intracranial hypertension (RICH) during ICU stay. The RICH detection thresholds which maximize the sum of sensitivity and specificity are TILmedian ≥ 7.5 and TILmax ≥ 14. The TIL24 threshold which maximizes the sum of sensitivity and specificity in the detection of surgical ICP control is TIL24 ≥ 9. The median scores of each TIL component therapy over increasing TIL24 reflect a credible staircase approach to treatment intensity escalation, from head positioning to surgical ICP control, as well as considerable variability in the use of cerebrospinal fluid drainage and decompressive craniectomy. Since TIL(Basic)max suffers from a strong statistical ceiling effect and only covers 17% (95% confidence interval [CI]: 16-18%) of the information in TILmax, TIL(Basic) should not be used instead of TIL for rating maximum treatment intensity. TIL(Basic)24 and TIL(Basic)median can be suitable replacements for TIL24 and TILmedian, respectively (with up to 33% [95% CI: 31-35%] information coverage) when full TIL assessment is infeasible. Accordingly, we derive numerical ranges for categorising TIL24 scores into TIL(Basic)24 scores. In conclusion, our results validate TIL across a spectrum of ICP management and monitoring approaches. TIL is a more sensitive surrogate for pathophysiology than ICP and thus can be considered an intermediate outcome after TBI.

Improvement in Functional Outcome from 6 to 12 Months After Moderate and Severe Traumatic Brain Injury Is Frequent, But May Not Be Detected With the Glasgow Outcome Scale Extended

The aims of this study were (1) to report outcome and change in outcome in patients with moderate and severe traumatic brain injury (mo/sTBI) between 6 and 12 months post-injury as measured by the Glasgow Outcome Scale Extended (GOSE), (2) to explore if demographic/injury-related variables can predict improvement in GOSE score, and (3) to investigate rate of improvement in Disability Rating Scale (DRS) score, in patients with a stable GOSE. All surviving patients ≥16 years of age who were admitted with mo/sTBI (Glasgow Coma Scale [GCS] score ≤13) to the regional trauma center in Central Norway between 2004 and 2019 were prospectively included (n = 439 out of 503 eligible). GOSE and DRS were used to assess outcome. Twelve-months post-injury, 13% with moTBI had severe disability (GOSE 2-4) versus 27% in sTBI, 26% had moderate disability (GOSE 5-6) versus 41% in sTBI and 62% had good recovery (GOSE 7-8) versus 31% in sTBI. From 6 to 12 months post-injury, 27% with moTBI and 32% with sTBI had an improvement, whereas 6% with moTBI and 6% with sTBI had a deterioration in GOSE score. Younger age and higher GCS score were associated with improved GOSE score. Improvement was least frequent for patients with a GOSE score of 3 at 6 months. In patients with a stable GOSE score of 3, an improvement in DRS score was observed in 22 (46%) patients. In conclusion, two thirds and one third of patients with mo/sTBI, respectively, had a good recovery. Importantly, change, mostly improvement, in GOSE score between 6 and 12 months was frequent and argues against the use of 6 months outcome as a time end-point in research. The GOSE does, however, not seem to be sensitive to actual change in function in the lower categories and a combination of outcome measures may be needed to describe the consequences after TBI.

Data-driven distillation and precision prognosis in traumatic brain injury with interpretable machine learning

Traumatic brain injury (TBI) affects how the brain functions in the short and long term. Resulting patient outcomes across physical, cognitive, and psychological domains are complex and often difficult to predict. Major challenges to developing personalized treatment for TBI include distilling large quantities of complex data and increasing the precision with which patient outcome prediction (prognoses) can be rendered. We developed and applied interpretable machine learning methods to TBI patient data. We show that complex data describing TBI patients' intake characteristics and outcome phenotypes can be distilled to smaller sets of clinically interpretable latent factors. We demonstrate that 19 clusters of TBI outcomes can be predicted from intake data, a ~ 6× improvement in precision over clinical standards. Finally, we show that 36% of the outcome variance across patients can be predicted. These results demonstrate the importance of interpretable machine learning applied to deeply characterized patients for data-driven distillation and precision prognosis.

Mining the contribution of intensive care clinical course to outcome after traumatic brain injury

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. Here, we integrate all heterogenous data stored in medical records (1166 pre-ICU and ICU variables) to model the individualised contribution of clinical course to 6-month functional outcome on the Glasgow Outcome Scale -Extended (GOSE). On a prospective cohort (n = 1550, 65 centres) of TBI patients, we train recurrent neural network models to map a token-embedded time series representation of all variables (including missing values) to an ordinal GOSE prognosis every 2 h. The full range of variables explains up to 52% (95% CI: 50-54%) of the ordinal variance in functional outcome. Up to 91% (95% CI: 90-91%) of this explanation is derived from pre-ICU and admission information (i.e., static variables). Information collected in the ICU (i.e., dynamic variables) increases explanation (by up to 5% [95% CI: 4-6%]), though not enough to counter poorer overall performance in longer-stay (>5.75 days) patients. Highest-contributing variables include physician-based prognoses, CT features, and markers of neurological function. Whilst static information currently accounts for the majority of functional outcome explanation after TBI, data-driven analysis highlights investigative avenues to improve the dynamic characterisation of longer-stay patients. Moreover, our modelling strategy proves useful for converting large patient records into interpretable time series with missing data integration and minimal processing.

Prediction performance of the machine learning model in predicting mortality risk in patients with traumatic brain injuries: a systematic review and meta-analysis

PURPOSE

With the in-depth application of machine learning(ML) in clinical practice, it has been used to predict the mortality risk in patients with traumatic brain injuries(TBI). However, there are disputes over its predictive accuracy. Therefore, we implemented this systematic review and meta-analysis, to explore the predictive value of ML for TBI.

METHODOLOGY

We systematically retrieved literature published in PubMed, Embase.com, Cochrane, and Web of Science as of November 27, 2022. The prediction model risk of bias(ROB) assessment tool (PROBAST) was used to assess the ROB of models and the applicability of reviewed questions. The random-effects model was adopted for the meta-analysis of the C-index and accuracy of ML models, and a bivariate mixed-effects model for the meta-analysis of the sensitivity and specificity.

RESULT

A total of 47 papers were eligible, including 156 model, with 122 newly developed ML models and 34 clinically recommended mature tools. There were 98 ML models predicting the in-hospital mortality in patients with TBI; the pooled C-index, sensitivity, and specificity were 0.86 (95% CI: 0.84, 0.87), 0.79 (95% CI: 0.75, 0.82), and 0.89 (95% CI: 0.86, 0.92), respectively. There were 24 ML models predicting the out-of-hospital mortality; the pooled C-index, sensitivity, and specificity were 0.83 (95% CI: 0.81, 0.85), 0.74 (95% CI: 0.67, 0.81), and 0.75 (95% CI: 0.66, 0.82), respectively. According to multivariate analysis, GCS score, age, CT classification, pupil size/light reflex, glucose, and systolic blood pressure (SBP) exerted the greatest impact on the model performance.

CONCLUSION

According to the systematic review and meta-analysis, ML models are relatively accurate in predicting the mortality of TBI. A single model often outperforms traditional scoring tools, but the pooled accuracy of models is close to that of traditional scoring tools. The key factors related to model performance include the accepted clinical variables of TBI and the use of CT imaging.