Recently, the investigators achieved prediction of episodes of intracranial hypertension
(ICH) and brain tissue hypoxia (BTH) 30 minutes to several hours in advance of overt
deterioration using intuitive and simple models whose inputs include recent intracranial
pressure (ICP) or partial brain tissue oxygen tension (PbtO2) values, the time since the
last crisis, and the fraction of time spent with high/low ICP or PbtO2, respectively. A
retrospective study was conducted based on prospectively collected physiologic data, and
machine-learning-based algorithms were developed to recognize physiologic patterns in
severe traumatic brain injury (TBI) patients that occur in advance of ICP and PbtO2
crises. These events were defined as ICP ≥ 20 mmHg lasting at least 15 minutes and PbtO2
values < 10 mmHg for at least 10 minutes, respectively. The physiologic data preceding
each crisis event were used to identify precursors associated with crisis onset. Further
investigation was conducted to assess how model performance changed as the prediction
time increased, providing an estimate of the clinical timescale of crisis precursors.
Multivariate classification models were applied to recorded data in 30-minute epochs of
time to predict crises between 15 and 360 minutes in the future. The cohort consisted of
817 TBI patients admitted to the neurocritical care unit at Ben Taub Hospital (Houston,
Texas). The algorithm predicted the onset of ICP crises with 30 minutes advance warning
with an area under the curve (AUC) of 0.86, in independent data, using only ICP
measurements and time since last crisis. An analogous algorithm predicted the onset of
PbtO2 crises with 30 minutes advance warning with an AUC of 0.91.
This phase seeks to study prospectively the performance of algorithms and the alerting
system for the prediction of intracranial hypertension and brain hypoxia crises. The
virtual monitor (VM) will be applied to all consecutive severe traumatic brain injury
(TBI) patients. Alerts will be generated by the algorithms and sent to study team
members. Detailed annotation of clinical interventions and events will be collected;
attending physicians and all clinical providers will be blinded to the presence of the
monitor and to the alerts. This will allow for the comparison of prospectively collected
crisis alerts with actual events of intracranial hypertension and brain tissue hypoxia
that follow them (or not). The goal is to establish the sensitivity of the prediction
algorithms as defined by the rate of true positive (TP) predictions (TP predictions are
the ones accompanied by actual crises of ICP and/or PbtO2 within 30 minutes of the
generated alert). Instances of actual crisis episodes that have not been preceded by an
alert will also be collected and analyzed. This phase of the study will be run in 4
cycles, each consisting of 2 months of data collection, followed by 1 month of system
improvements. Technical improvements to the system (such as adjusting the alerting
threshold, user interface features, etc.) will be made at the end of each cycle, based on
feedback provided by the study team from the previous 2 months of data collection.
This phase involves clinical management per local neurocritical care unit standards and
as per Brain Trauma Foundation (BTF) guidelines (see below baseline standard assessments,
and Table 1 for tiered approaches to violations of ICP and PbtO2 thresholds). Alerts will
only be available to study personnel (study physician members not directly involved in
patient care, and study coordinators) for the purposes of evaluating real-time applied
accuracy of the algorithm and alerting system.
Each year in the United States, about 2 million people sustain traumatic brain injury
(TBI), and of these, 500,000 require hospital care. Severe traumatic brain injury (TBI)
is associated with 20-30% mortality and significant disability among most survivors. The
Centers for Disease Control and Prevention (CDC) estimate that 2% of the U.S. population
lives with disabilities directly attributable to TBI, with annual costs exceeding $76.5
billion.
Largely, current treatments are ineffective due to being instituted at a time when
irreversible damage has already occurred. By the time it is recognized that the
traumatized brain is being further injured due to high pressure or lack of oxygen, it is
too late to reverse or repair the insult. A computerized method is proposed to recognize
further brain injury before it occurs. The ability to predict these events ahead of time
will allow clinicians to prepare and prevent these episodes before they cause permanent
brain damage.
The investigators have already developed a method that can provide accurate prediction of
these injurious events. This proposal aims to test predictions in real time. The patients
to benefit from this project include individuals who suffer all kinds of severe trauma to
the brain. This research promises to improve the outcome of these patients by providing
predictions of further brain injury and, in the future, a way to act upon them to prevent
permanent brain damage.
This proposal is divided into two yearly phases. The first year focuses on establishing
the capability of extracting and analyzing data from bedside monitors in the intensive
care unit (ICU). The second year will test in real time the accuracy of the predictions.
The first part of the project (Year 1) does not involve any patient or subject
interaction. This year is solely for informatics infrastructure setup.
In the second part of the project (Year 2), subjects (via their surrogate
decision-makers) will be asked to be enrolled in an observational study where data on ICP
and PbtO2 are collected, and the prediction algorithm is tested for accuracy and
performance. During this phase, clinical management will be per standard of care, and no
additional interventions will be performed. This phase will be conducted under this
protocol and a dedicated consent form (consent for Phase 2; attached at the end).
Efforts will be made to contact the subject or their caretakers at 6 months for follow-up
on recovery.
Approximately 120 individuals will take part in this study at the University of Chicago
and Ben Taub General Hospital in Houston.
This project utilizes a computerized method to predict ongoing and future brain injury.
The data to be collected includes two sources: the first is the electronic medical record
(EMR), which includes demographic information, characteristics of the injury, laboratory
values, and data from imaging. The second source is directly from the bedside ICU
monitor, including vital signs such as intracranial pressure (ICP), brain tissue oxygen
(PbtO2), and mean arterial pressure (MAP). All data will be securely stored in a research
computer database.
Expanded Acronyms/Abbreviations:
Intracranial Hypertension (ICH)
Brain Tissue Hypoxia (BTH)
Intracranial Pressure (ICP)
Partial Brain Tissue Oxygen Tension (PbtO2)
Traumatic Brain Injury (TBI)
Area Under the Curve (AUC)
True Positive (TP)
Virtual Monitor (VM)
Brain Trauma Foundation (BTF)
Centers for Disease Control and Prevention (CDC)
Intensive Care Unit (ICU)
Electronic Medical Record (EMR)
Mean Arterial Pressure (MAP)
