Probabilistic prediction of increased intracranial pressure in patients with severe traumatic brain injury

Classification of intracranial pressure epochs using a novel machine learning framework

Patients with acute brain injuries are at risk for life threatening elevated intracranial pressure (ICP). External Ventricular Drains (EVDs) are used to measure and treat ICP, which switch between clamped and draining configurations, with accurate ICP data only available during clamped periods. While traditional guidelines focus on mean ICP values, evolving evidence indicates other waveform features may hold prognostic value. However, current machine learning models using ICP waveforms exclude EVD data due to a lack of digital labels indicating the clamped state, markedly limiting their generalizability. We introduce, detail, and validate CICL (Classification of ICP epochs using a machine Learning framework), a semi-supervised approach to classify ICP segments from EVDs as clamped, draining, or noise. This paves the way for multiple applications, including generalizable ICP crisis prediction, potentially benefiting tens of thousands of patients annually and highlights an innovate methodology to label large high frequency physiological time series datasets.

Machine Learning Based Prediction of Imminent ICP Insults During Neurocritical Care of Traumatic Brain Injury

BACKGROUND

In neurointensive care, increased intracranial pressure (ICP) is a feared secondary brain insult in traumatic brain injury (TBI). A system that predicts ICP insults before they emerge may facilitate early optimization of the physiology, which may in turn lead to stopping the predicted ICP insult from occurring. The aim of this study was to evaluate the performance of different artificial intelligence models in predicting the risk of ICP insults.

METHODS

The models were trained to predict risk of ICP insults starting within 30 min, using the Uppsala high frequency TBI dataset. A restricted dataset consisting of only monitoring data were used, and an unrestricted dataset using monitoring data as well as clinical data, demographic data, and radiological evaluations was used. Four different model classes were compared: Gaussian process regression, logistic regression, random forest classifier, and Extreme Gradient Boosted decision trees (XGBoost).

RESULTS

Six hundred and two patients with TBI were included (total monitoring 138,411 h). On the task of predicting upcoming ICP insults, the Gaussian process regression model performed similarly on the Uppsala high frequency TBI dataset (sensitivity 93.2%, specificity 93.9%, area under the receiver operating characteristic curve [AUROC] 98.3%), as in earlier smaller studies. Using a more flexible model (XGBoost) resulted in a comparable performance (sensitivity 93.8%, specificity 94.6%, AUROC 98.7%). Adding more clinical variables and features further improved the performance of the models slightly (XGBoost: sensitivity 94.1%, specificity of 94.6%, AUROC 98.8%).

CONCLUSIONS

Artificial intelligence models have potential to become valuable tools for predicting ICP insults in advance during neurointensive care. The fact that common off-the-shelf models, such as XGBoost, performed well in predicting ICP insults opens new possibilities that can lead to faster advances in the field and earlier clinical implementations.

Prediction Model for Unfavorable Outcome in Primary Decompressive Craniectomy for Isolated Moderate to Severe Traumatic Brain Injury in India: A Prospective Observational Study

OBJECTIVE

Traumatic brain injury (TBI) prediction models have gained significant attention in recent years because of their potential to aid in clinical decision making. Existing models, such as Corticosteroid Randomization after Significant Head Injury and International Mission for Prognosis and Analysis of Clinical Trials, are currently losing external validity and performance, probably because of their diverse inclusion criteria and changes in treatment modalities over the years. There is a lack of models that predict outcomes strictly pertaining to primary decompression after TBI. In this study, we aimed to develop an easy-to-use prediction model for predicting the risk of poor functional outcomes at 3 months after hospital discharge in adult patients who had undergone primary decompressive craniectomy for isolated moderate-to-severe TBI.

METHODS

We conducted a prospective observational study at our tertiary care hospital. We trained and tested multiple prognostic logistic regression models with ten-fold cross validation to choose the model with the lowest Akaike information criterion, high sensitivity, and positive predictive value (PPV). Using the final model, we generated a nomogram to predict the risk of having a Glasgow outcome scale-extended (GOSE) 1-4 at three months after hospital discharge.

RESULTS

A total of 215 patients were included in this study. Variables with an absolute standardized difference >0·25 when grouped by GOSE 1-4/5-8 at three months were included in multivariable modeling. The model of choice had an accuracy of 87·91% (95% confidence interval of 82·78%-91·95%), a sensitivity of 84·42%, specificity of 89·86%, PPV of 82·28% (72·06%-89·96%), negative predictive value of 91·18% (85·09%-95·36%), LR+ of 8·32 (5·02-13·80), and LR-of 0·17 (0·10-0·29).

CONCLUSIONS

Our study provides a ready-to-use prognostic nomogram derived from prospective data that can predict the risk of having a GOSE of 1-4 at three months following primary decompressive craniectomy with high sensitivity, PPV, and low LR-.

Evolving concepts in intracranial pressure monitoring - from traditional monitoring to precision medicine

A wide range of acute brain injuries, including both traumatic and non-traumatic causes, can result in elevated intracranial pressure (ICP), which in turn can cause further secondary injury to the brain, initiating a vicious cascade of propagating injury. Elevated ICP is therefore a neurological injury that requires intensive monitoring and time-sensitive interventions. Patients at high risk for developing elevated ICP undergo placement of invasive ICP monitors including external ventricular drains, intraparenchymal ICP monitors, and lumbar drains. These monitors all generate an ICP waveform, but each has its own unique caveats in monitoring and accuracy. Current ICP monitoring and management clinical guidelines focus on the mean ICP derived from the ICP waveform, with standard thresholds of treating ICP greater than 20 ​mmHg or 22 ​mmHg applied broadly to a wide range of patients. However, this one-size fits all approach has been criticized and there is a need to develop personalized, evidence-based and possibly multi-factorial precision-medicine based approaches to the problem. This paper provides historical and physiological context to the problem of elevated ICP, provides an overview of the challenges of the current paradigm of ICP management strategies, and discusses advances in ICP waveform analysis, emerging non-invasive ICP monitoring techniques, and applications of machine learning to create predictive algorithms.

Intracranial pressure monitoring in adult patients with traumatic brain injury: challenges and innovations

Intracranial pressure monitoring enables the detection and treatment of intracranial hypertension, a potentially lethal insult after traumatic brain injury. Despite its widespread use, robust evidence supporting intracranial pressure monitoring and treatment remains sparse. International studies have shown large variations between centres regarding the indications for intracranial pressure monitoring and treatment of intracranial hypertension. Experts have reviewed these two aspects and, by consensus, provided practical approaches for monitoring and treatment. Advances have occurred in methods for non-invasive estimation of intracranial pressure although, for now, a reliable way to non-invasively and continuously measure intracranial pressure remains aspirational. Analysis of the intracranial pressure signal can provide information on brain compliance (ie, the ability of the cranium to tolerate volume changes) and on cerebral autoregulation (ie, the ability of cerebral blood vessels to react to changes in blood pressure). The information derived from the intracranial pressure signal might allow for more individualised patient management. Machine learning and artificial intelligence approaches are being increasingly applied to intracranial pressure monitoring, but many obstacles need to be overcome before their use in clinical practice could be attempted. Robust clinical trials are needed to support indications for intracranial pressure monitoring and treatment. Progress in non-invasive assessment of intracranial pressure and in signal analysis (for targeted treatment) will also be crucial.

Using Artificial Intelligence to Predict Intracranial Hypertension in Patients After Traumatic Brain Injury: A Systematic Review

Intracranial hypertension (IH) is a key driver of secondary brain injury in patients with traumatic brain injury. Lowering intracranial pressure (ICP) as soon as IH occurs is important, but a preemptive approach would be more beneficial. We systematically reviewed the artificial intelligence (AI) models, variables, performances, risks of bias, and clinical machine learning (ML) readiness levels of IH prediction models using AI. We conducted a systematic search until 12-03-2023 in three databases. Only studies predicting IH or ICP in patients with traumatic brain injury with a validation of the AI model were included. We extracted type of AI model, prediction variables, model performance, validation type, and prediction window length. Risk of bias was assessed with the Prediction Model Risk of Bias Assessment Tool, and we determined the clinical ML readiness level. Eleven out of 399 nonduplicate publications were included. A gaussian processes model using ICP and mean arterial pressure was most common. The maximum reported area under the receiver operating characteristic curve was 0.94. Four studies conducted external validation, and one study a prospective clinical validation. The prediction window length preceding IH varied between 30 and 60 min. Most studies (73%) had high risk of bias. The highest clinical ML readiness level was 6 of 9, indicating "real-time model testing" stage in one study. Several IH prediction models using AI performed well, were externally validated, and appeared ready to be tested in the clinical workflow (clinical ML readiness level 5 of 9). A Gaussian processes model was most used, and ICP and mean arterial pressure were frequently used variables. However, most studies showed a high risk of bias. Our findings may help position AI for IH prediction on the path to ultimate clinical integration and thereby guide researchers plan and design future studies.

Random forest-based prediction of intracranial hypertension in patients with traumatic brain injury

BACKGROUND

Treatment and prevention of intracranial hypertension (IH) to minimize secondary brain injury are central to the neurocritical care management of traumatic brain injury (TBI). Predicting the onset of IH in advance allows for a more aggressive prophylactic treatment. This study aimed to develop random forest (RF) models for predicting IH events in TBI patients.

METHODS

We analyzed prospectively collected data from patients admitted to the intensive care unit with invasive intracranial pressure (ICP) monitoring. Patients with persistent ICP > 22 mmHg in the early postoperative period (first 6 h) were excluded to focus on IH events that had not yet occurred. ICP-related data from the initial 6 h were used to extract linear (ICP, cerebral perfusion pressure, pressure reactivity index, and cerebrospinal fluid compensatory reserve index) and nonlinear features (complexity of ICP and cerebral perfusion pressure). IH was defined as ICP > 22 mmHg for > 5 min, and severe IH (SIH) as ICP > 22 mmHg for > 1 h during the subsequent ICP monitoring period. RF models were then developed using baseline characteristics (age, sex, and initial Glasgow Coma Scale score) along with linear and nonlinear features. Fivefold cross-validation was performed to avoid overfitting.

RESULTS

The study included 69 patients. Forty-three patients (62.3%) experienced an IH event, of whom 30 (43%) progressed to SIH. The median time to IH events was 9.83 h, and to SIH events, it was 11.22 h. The RF model showed acceptable performance in predicting IH with an area under the curve (AUC) of 0.76 and excellent performance in predicting SIH (AUC = 0.84). Cross-validation analysis confirmed the stability of the results.

CONCLUSIONS

The presented RF model can forecast subsequent IH events, particularly severe ones, in TBI patients using ICP data from the early postoperative period. It provides researchers and clinicians with a potentially predictive pathway and framework that could help triage patients requiring more intensive neurological treatment at an early stage.

A deep learning approach for generating intracranial pressure waveforms from extracranial signals routinely measured in the intensive care unit

Intracranial pressure (ICP) is commonly monitored to guide treatment in patients with serious brain disorders such as traumatic brain injury and stroke. Established methods to assess ICP are resource intensive and highly invasive. We hypothesized that ICP waveforms can be computed noninvasively from three extracranial physiological waveforms routinely acquired in the Intensive Care Unit (ICU): arterial blood pressure (ABP), photoplethysmography (PPG), and electrocardiography (ECG). We evaluated over 600 h of high-frequency (125 Hz) simultaneously acquired ICP, ABP, ECG, and PPG waveform data in 10 patients admitted to the ICU with critical brain disorders. The data were segmented in non-overlapping 10-s windows, and ABP, ECG, and PPG waveforms were used to train deep learning (DL) models to re-create concurrent ICP. The predictive performance of six different DL models was evaluated in single- and multi-patient iterations. The mean average error (MAE) ± SD of the best-performing models was 1.34 ± 0.59 mmHg in the single-patient and 5.10 ± 0.11 mmHg in the multi-patient analysis. Ablation analysis was conducted to compare contributions from single physiologic sources and demonstrated statistically indistinguishable performances across the top DL models for each waveform (MAE±SD 6.33 ± 0.73, 6.65 ± 0.96, and 7.30 ± 1.28 mmHg, respectively, for ECG, PPG, and ABP; p = 0.42). Results support the preliminary feasibility and accuracy of DL-enabled continuous noninvasive ICP waveform computation using extracranial physiological waveforms. With refinement and further validation, this method could represent a safer and more accessible alternative to invasive ICP, enabling assessment and treatment in low-resource settings.

Deriving Automated Device Metadata From Intracranial Pressure Waveforms: A Transforming Research and Clinical Knowledge in Traumatic Brain Injury ICU Physiology Cohort Analysis

IMPORTANCE

Treatment for intracranial pressure (ICP) has been increasingly informed by machine learning (ML)-derived ICP waveform characteristics. There are gaps, however, in understanding how ICP monitor type may bias waveform characteristics used for these predictive tools since differences between external ventricular drain (EVD) and intraparenchymal monitor (IPM)-derived waveforms have not been well accounted for.

OBJECTIVES

We sought to develop a proof-of-concept ML model differentiating ICP waveforms originating from an EVD or IPM.

DESIGN, SETTING, AND PARTICIPANTS

We examined raw ICP waveform data from the ICU physiology cohort within the prospective Transforming Research and Clinical Knowledge in Traumatic Brain Injury multicenter study.

MAIN OUTCOMES AND MEASURES

Nested patient-wise five-fold cross-validation and group analysis with bagged decision trees (BDT) and linear discriminant analysis were used for feature selection and fair evaluation. Nine patients were kept as unseen hold-outs for further evaluation.

RESULTS

ICP waveform data totaling 14,110 hours were included from 82 patients (EVD, 47; IPM, 26; both, 9). Mean age, Glasgow Coma Scale (GCS) total, and GCS motor score upon admission, as well as the presence and amount of midline shift, were similar between groups. The model mean area under the receiver operating characteristic curve (AU-ROC) exceeded 0.874 across all folds. In additional rigorous cluster-based subgroup analysis, targeted at testing the resilience of models to cross-validation with smaller subsets constructed to develop models in one confounder set and test them in another subset, AU-ROC exceeded 0.811. In a similar analysis using propensity score-based rather than cluster-based subgroup analysis, the mean AU-ROC exceeded 0.827. Of 842 extracted ICP features, 62 were invariant within every analysis, representing the most accurate and robust differences between ICP monitor types. For the nine patient hold-outs, an AU-ROC of 0.826 was obtained using BDT.

CONCLUSIONS AND RELEVANCE

The developed proof-of-concept ML model identified differences in EVD- and IPM-derived ICP signals, which can provide missing contextual data for large-scale retrospective datasets, prevent bias in computational models that ingest ICP data indiscriminately, and control for confounding using our model's output as a propensity score by to adjust for the monitoring method that was clinically indicated. Furthermore, the invariant features may be leveraged as ICP features for anomaly detection.

Time-Series Modeling and Forecasting of Cerebral Pressure-Flow Physiology: A Scoping Systematic Review of the Human and Animal Literature

The modeling and forecasting of cerebral pressure-flow dynamics in the time-frequency domain have promising implications for veterinary and human life sciences research, enhancing clinical care by predicting cerebral blood flow (CBF)/perfusion, nutrient delivery, and intracranial pressure (ICP)/compliance behavior in advance. Despite its potential, the literature lacks coherence regarding the optimal model type, structure, data streams, and performance. This systematic scoping review comprehensively examines the current landscape of cerebral physiological time-series modeling and forecasting. It focuses on temporally resolved cerebral pressure-flow and oxygen delivery data streams obtained from invasive/non-invasive cerebral sensors. A thorough search of databases identified 88 studies for evaluation, covering diverse cerebral physiologic signals from healthy volunteers, patients with various conditions, and animal subjects. Methodologies range from traditional statistical time-series analysis to innovative machine learning algorithms. A total of 30 studies in healthy cohorts and 23 studies in patient cohorts with traumatic brain injury (TBI) concentrated on modeling CBFv and predicting ICP, respectively. Animal studies exclusively analyzed CBF/CBFv. Of the 88 studies, 65 predominantly used traditional statistical time-series analysis, with transfer function analysis (TFA), wavelet analysis, and autoregressive (AR) models being prominent. Among machine learning algorithms, support vector machine (SVM) was widely utilized, and decision trees showed promise, especially in ICP prediction. Nonlinear models and multi-input models were prevalent, emphasizing the significance of multivariate modeling and forecasting. This review clarifies knowledge gaps and sets the stage for future research to advance cerebral physiologic signal analysis, benefiting neurocritical care applications.

IntraCranial pressure prediction AlgoRithm using machinE learning (I-CARE): Training and Validation Study

OBJECTIVES

Elevated intracranial pressure (ICP) is a potentially devastating complication of neurologic injury. Developing an ICP prediction algorithm to help the clinician adjust treatments and potentially prevent elevated ICP episodes.

DESIGN

Retrospective study.

SETTING

Three hundred thirty-five ICUs at 208 hospitals in the United States.

SUBJECTS

Adults patients from the electronic ICU (eICU) Collaborative Research Database was used to train an ensemble machine learning model to predict the ICP 30 minutes in the future. Predictive performance was evaluated using a left-out test dataset and externally evaluated on the Medical Information Mart for Intensive Care-III (MIMIC-III) Matched Waveform Database.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Predictors included age, assigned sex, laboratories, medications and infusions, input/output, Glasgow Coma Scale (GCS) components, and time-series vitals (heart rate, ICP, mean arterial pressure, respiratory rate, and temperature). Each patient ICU stay was divided into successive 95-minute timeblocks. For each timeblock, the model was trained on nontime-varying covariates as well as on 12 observations of time-varying covariates at 5-minute intervals and asked to predict the 5-minute median ICP 30 minutes after the last observed ICP value. Data from 931 patients with ICP monitoring in the eICU dataset were extracted (46,207 timeblocks). The root mean squared error was 4.51 mm Hg in the eICU test set and 3.56 mm Hg in the MIMIC-III dataset. The most important variables driving ICP prediction were previous ICP history, patients' temperature, weight, serum creatinine, age, GCS, and hemodynamic parameters.

CONCLUSIONS

IntraCranial pressure prediction AlgoRithm using machinE learning, an ensemble machine learning model, trained to predict the ICP of a patient 30 minutes in the future based on baseline characteristics and vitals data from the past hour showed promising predictive performance including in an external validation dataset.

Mortality prediction using medical time series on TBI patients

BACKGROUND AND OBJECTIVE

Traumatic Brain Injury (TBI) is one of the leading causes of injury-related mortality in the world, with severe cases reaching mortality rates of 30-40%. It is highly heterogeneous both in causes and consequences making more complex the medical interpretation and prognosis. Gathering clinical, demographic, and laboratory data to perform a prognosis requires time and skill in several clinical specialties. Artificial intelligence (AI) methods can take advantage of existing data by performing helpful predictions and guiding physicians toward a better prognosis and, consequently, better healthcare. The objective of this work was to develop learning models and evaluate their capability of predicting the mortality of TBI. The predictive model would allow the early assessment of the more serious cases and scarce medical resources can be pointed toward the patients who need them most.

METHODS

Long Short Term Memory (LSTM) and Transformer architectures were tested and compared in performance, coupled with data imbalance, missing data, and feature selection strategies. From the Medical Information Mart for Intensive Care III (MIMIC-III) dataset, a cohort of TBI patients was selected and an analysis of the first 48 hours of multiple time series sequential variables was done to predict hospital mortality.

RESULTS

The best performance was obtained with the Transformer architecture, achieving an AUC of 0.907 with the larger group of features and trained with class proportion class weights and binary cross entropy loss.

CONCLUSIONS

Using the time series sequential data, LSTM and Transformers proved to be both viable options for predicting TBI hospital mortality in 48 hours after admission. Overall, using sequential deep learning models with time series data to predict TBI mortality is viable and can be used as a helpful indicator of the well-being of patients.