Search

[This corrects the article DOI: 10.1371/journal.pone.0265254.].

Artificial intelligence and machine learning (AI/ML) is becoming increasingly more accessible to biomedical researchers with significant potential to transform biomedicine through optimization of highly-accurate predictive models and enabling better understanding of disease biology. Automated machine learning (AutoML) in particular is positioned to democratize artificial intelligence (AI) by reducing the amount of human input and ML expertise needed. However, successful translation of AI/ML in biomedicine requires moving beyond optimizing only for prediction accuracy and towards establishing reproducible clinical and biological inferences. This is especially challenging for clinical studies on rare disorders where the smaller patient cohorts and corresponding sample size is an obstacle for reproducible modeling results. Here, we present a model-agnostic framework to reinforce AutoML using strategies and tools of explainable and reproducible AI, including novel metrics to assess model reproducibility. The framework enables clinicians to interpret AutoML-generated models for clinical and biological verifiability and consequently integrate domain expertise during model development. We applied the framework towards spinal cord injury prognostication to optimize the intraoperative hemodynamic range during injury-related surgery and additionally identified a strong detrimental relationship between intraoperative hypertension and patient outcome. Furthermore, our analysis captured how evolving clinical practices such as faster time-to-surgery and blood pressure management affect clinical model development. Altogether, we illustrate how expert-augmented AutoML improves inferential reproducibility for biomedical discovery and can ultimately build trust in AI processes towards effective clinical integration.

Long-term outcomes related to emergency department revisit, hospital readmission, and all-cause mortality, have not been well characterized across the spectrum of pediatric traumatic brain injury (TBI). We evaluated emergency department visit outcomes up to 1 year after pediatric TBI, in comparison to a referent group of trauma patients without TBI. We performed a longitudinal, retrospective study of all pediatric trauma patients who presented to emergency departments and hospitals in California from 2005 to 2014. We compared emergency department visits, dispositions, revisits, readmissions, and mortality in pediatric trauma patients with a TBI diagnosis to those without TBI (Other Trauma patients). We identified 208,222 pediatric patients with an index diagnosis of TBI and 1,314,064 patients with an index diagnosis of Other Trauma. Population growth adjusted TBI visits increased by 5.6% while those for Other Trauma decreased by 40.7%. The majority of patients were discharged from the emergency department on their first visit (93.2% for traumatic brain injury vs. 96.5% for Other Trauma). A greater proportion of TBI patients revisited the emergency department (33.4% vs. 3.0%) or were readmitted to the hospital (0.9% vs. 0.04%) at least once within a year of discharge. The health burden within a year after a pediatric TBI visit is considerable and is greater than that of non-TBI trauma. These data suggest that outpatient strategies to monitor for short-term and longer-term sequelae after pediatric TBI are needed to improve patient outcomes, lessen the burden on families, and more appropriately allocate resources in the healthcare system.

IntroductionThe high failure rate of clinical trials in traumatic brain injury (TBI) may be attributable, in part, to the use of untested or insensitive measurement instruments. Of more than 1,000 clinical outcome assessment measures (COAs) for TBI, few have been systematically vetted to determine their performance within specific "contexts of use (COU)." As described in guidance issued by the U.S. Food and Drug Administration (FDA), the COU specifies the population of interest and the purpose for which the COA will be employed. COAs are commonly used for screening, diagnostic categorization, outcome prediction, and establishing treatment effectiveness. COA selection typically relies on expert consensus; there is no established methodology to match the appropriateness of a particular COA to a specific COU. We developed and pilot-tested the Evidence-Based Clinical Outcome assessment Platform (EB-COP) to systematically and transparently evaluate the suitability of TBI COAs for specific purposes.Methods and findingsFollowing a review of existing literature and published guidelines on psychometric standards for COAs, we developed a 6-step, semi-automated, evidence-based assessment platform to grade COA performance for six specific purposes: diagnosis, symptom detection, prognosis, natural history, subgroup stratification and treatment effectiveness. Mandatory quality indicators (QIs) were identified for each purpose using a modified Delphi consensus-building process. The EB-COP framework was incorporated into a Qualtrics software platform and pilot-tested on the Glasgow Outcome Scale-Extended (GOSE), the most widely-used COA in TBI clinical studies.ConclusionThe EB-COP provides a systematic methodology for conducting more precise, evidence-based assessment of COAs by evaluating performance within specific COUs. The EB-COP platform was shown to be feasible when applied to a TBI COA frequently used to detect treatment effects and can be modified to address other populations and COUs. Additional testing and validation of the EB-COP are warranted.

BACKGROUND:The current classification of traumatic brain injury (TBI) into "mild", "moderate", or "severe" does not adequately account for the patient heterogeneity that still exists within each of these categories. The objective of this study was to identify "sub-groups" of mild TBI (mTBI) patients based on data available at the time of the initial post-TBI patient evaluation and to determine if the sub-grouping correlates with patient outcomes at 90 and 180 days post-TBI. METHODS:Data from patients in the TRACK-TBI Pilot dataset who had a Glasgow Coma Scale (GCS) score of 13 to 15 at arrival to the Emergency Department and a closed head injury were included. Considering 53 clinical variables that are typically available during the initial evaluation of the patient with mild TBI, sparse heirarchial clustering with cluster quality assessment was used to identify the optimal number of patient sub-groups. Patient sub-groups were then compared for ten outcomes measured at 90 or 180 days post-TBI. RESULTS:Amongst the 485 patients with mTBI, optimal clustering was based on the inclusion of 12 clinical variables that divided the patients into 5 mild TBI sub-groups. Clinical variables driving the sub-clustering included: gender, employment status, marital status, TBI due to falling, brain CT scan result, systolic blood pressure, diastolic blood pressure, administration of IV fluids in the Emergency Department, alcohol use, tobacco use, history of neurologic disease, and history of psychiatric disease. These 5 mild TBI sub-groups differed in their 90 day and 180 day outcomes within several domains including global outcomes, persistence of TBI-related symptoms, and neuropsychological impairment. CONCLUSIONS:Sub-groups of patients with mTBI can be identified according to clinical variables that are relatively easy to obtain at the time of initial patient evaluation. A patient's sub-group assignment is associated with multidimensional patient outcomes at 90 and 180 days. These findings support the notion that there are clinically meaningful subgroups of patients amongst those currently classified as having mTBI.

BACKGROUND:Traumatic brain injury (TBI) is a complex disorder that is traditionally stratified based on clinical signs and symptoms. Recent imaging and molecular biomarker innovations provide unprecedented opportunities for improved TBI precision medicine, incorporating patho-anatomical and molecular mechanisms. Complete integration of these diverse data for TBI diagnosis and patient stratification remains an unmet challenge. METHODS AND FINDINGS:The Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) Pilot multicenter study enrolled 586 acute TBI patients and collected diverse common data elements (TBI-CDEs) across the study population, including imaging, genetics, and clinical outcomes. We then applied topology-based data-driven discovery to identify natural subgroups of patients, based on the TBI-CDEs collected. Our hypothesis was two-fold: 1) A machine learning tool known as topological data analysis (TDA) would reveal data-driven patterns in patient outcomes to identify candidate biomarkers of recovery, and 2) TDA-identified biomarkers would significantly predict patient outcome recovery after TBI using more traditional methods of univariate statistical tests. TDA algorithms organized and mapped the data of TBI patients in multidimensional space, identifying a subset of mild TBI patients with a specific multivariate phenotype associated with unfavorable outcome at 3 and 6 months after injury. Further analyses revealed that this patient subset had high rates of post-traumatic stress disorder (PTSD), and enrichment in several distinct genetic polymorphisms associated with cellular responses to stress and DNA damage (PARP1), and in striatal dopamine processing (ANKK1, COMT, DRD2). CONCLUSIONS:TDA identified a unique diagnostic subgroup of patients with unfavorable outcome after mild TBI that were significantly predicted by the presence of specific genetic polymorphisms. Machine learning methods such as TDA may provide a robust method for patient stratification and treatment planning targeting identified biomarkers in future clinical trials in TBI patients. TRIAL REGISTRATION:ClinicalTrials.gov Identifier NCT01565551.