Optimizing Concussion Care Seeking: Using Machine Learning to Predict Delayed Concussion Reporting.

BACKGROUND: Early medical attention after concussion may minimize symptom duration and burden; however, many concussions are undiagnosed or have a delay in diagnosis after injury. Many concussion symptoms (eg, headache, dizziness) are not visible, meaning that early identification is often contingent on individuals reporting their injury to medical staff. A fundamental understanding of the types and levels of factors that explain when concussions are reported can help identify promising directions for intervention. PURPOSE: To identify individual and institutional factors that predict immediate (vs delayed) injury reporting. STUDY DESIGN: Case-control study; Level of evidence, 3. METHODS: This study was a secondary analysis of data from the Concussion Assessment, Research and Education (CARE) Consortium study. The sample included 3213 collegiate athletes and military service academy cadets who were diagnosed with a concussion during the study period. Participants were from 27 civilian institutions and 3 military institutions in the United States. Machine learning techniques were used to build models predicting who would report an injury immediately after a concussive event (measured by an athletic trainer denoting the injury as being reported "immediately" or "at a delay"), including both individual athlete/cadet and institutional characteristics. RESULTS: In the sample as a whole, combining individual factors enabled prediction of reporting immediacy, with mean accuracies between 55.8% and 62.6%, depending on classifier type and sample subset; adding institutional factors improved reporting prediction accuracies by 1 to 6 percentage points. At the individual level, injury-related altered mental status and loss of consciousness were most predictive of immediate reporting, which may be the result of observable signs leading to the injury report being externally mediated. At the institutional level, important attributes included athletic department annual revenue and ratio of athletes to athletic trainers. CONCLUSION: Further study is needed on the pathways through which institutional decisions about resource allocation, including decisions about sports medicine staffing, may contribute to reporting immediacy. More broadly, the relatively low accuracy of the machine learning models tested suggests the importance of continued expansion in how reporting is understood and facilitated.

Optimizing Concussion Care Seeking: Connecting Care-Seeking Behaviors and Neurophysiological States Through Blood Biomarkers.

BACKGROUND: Timely and appropriate medical care after concussion presents a difficult public health problem. Concussion identification and treatment rely heavily on self-report, but more than half of concussions go unreported or are reported after a delay. If incomplete self-report increases exposure to harm, blood biomarkers may objectively indicate this neurobiological dysfunction. PURPOSE/HYPOTHESIS: The purpose of this study was to compare postconcussion biomarker levels between individuals with different previous concussion diagnosis statuses and care-seeking statuses. It was hypothesized that individuals with undiagnosed concussions and poorer care seeking would show altered biomarker profiles. STUDY DESIGN: Cohort study; Level of evidence, 3. METHODS: Blood samples were collected from 287 military academy cadets and collegiate athletes diagnosed with concussion in the Advanced Research Core of the Concussion Assessment, Research and Education Consortium. The authors extracted each participant's self-reported previous concussion diagnosis status (no history, all diagnosed, ≥1 undiagnosed) and whether they had delayed or immediate symptom onset, symptom reporting, and removal from activity after the incident concussion. The authors compared the following blood biomarkers associated with neural injury between previous concussion diagnosis status groups and care-seeking groups: glial fibrillary acidic protein, ubiquitin c-terminal hydrolase-L1 (UCH-L1), neurofilament light chain (NF-L), and tau protein, captured at baseline, 24 to 48 hours, asymptomatic, and 7 days after unrestricted return to activity using tests of parallel profiles. RESULTS: The undiagnosed previous concussion group (n = 21) had higher levels of NF-L at 24- to 48-hour and asymptomatic time points relative to all diagnosed (n = 72) or no previous concussion (n = 194) groups. For those with delayed removal from activity (n = 127), UCH-L1 was lower at 7 days after return to activity than that for athletes immediately removed from activity (n = 131). No other biomarker differences were observed. CONCLUSION: Individuals with previous undiagnosed concussions or delayed removal from activity showed some different biomarker levels after concussion and after clinical recovery, despite a lack of baseline differences. This may indicate that poorer care seeking can create neurobiological differences in the concussed brain.

Optimizing Concussion Care Seeking: A Longitudinal Analysis of Recovery.

BACKGROUND: Approximately half of concussions go undisclosed and therefore undiagnosed. Among diagnosed concussions, 51% to 64% receive delayed medical care. Understanding the influence of undiagnosed concussions and delayed medical care would inform medical and education practices. PURPOSE: To compare postconcussion longitudinal clinical outcomes among (1) individuals with no concussion history, all previous concussions diagnosed, and ≥1 previous concussion undiagnosed, as well as (2) those who have delayed versus immediate symptom onset, symptom reporting, and removal from activity after concussion. STUDY DESIGN: Cohort study; Level of evidence, 2. METHODS: Participants included 2758 military academy cadets and intercollegiate athletes diagnosed with concussion in the CARE Consortium. We determined (1) each participant's previous concussion diagnosis status self-reported at baseline (no history, all diagnosed, ≥1 undiagnosed) and (2) whether the participant had delayed or immediate symptom onset, symptom reporting, and removal from activity. We compared symptom severities, cognition, balance, and recovery duration at baseline, 24 to 48 hours, date of asymptomatic status, and date of unrestricted return to activity using tests of parallel profiles. RESULTS: The ≥1 undiagnosed concussion group had higher baseline symptom burdens (P < .001) than the other 2 groups and poorer baseline verbal memory performance (P = .001) than the all diagnosed group; however, they became asymptomatic and returned to activity sooner than those with no history. Cadets/athletes who delayed symptom reporting had higher symptom burdens 24 to 48 hours after injury (mean ± SE; delayed, 28.8 ± 0.8; immediate, 20.6 ± 0.7), took a median difference of 2 days longer to become asymptomatic, and took 3 days longer to return to activity than those who had immediate symptom reporting. For every 30 minutes of continued participation after injury, days to asymptomatic status increased 8.1% (95% CI, 0.3%-16.4%). CONCLUSION: Clinicians should expect that cadets/athletes who delay reporting concussion symptoms will have acutely higher symptom burdens and take 2 days longer to become asymptomatic. Educational messaging should emphasize the clinical benefits of seeking immediate care for concussion-like symptoms.

Optimizing Concussion Care Seeking: The Influence of Previous Concussion Diagnosis Status on Baseline Assessment Outcomes.

BACKGROUND: The prevalence of unreported concussions is high, and undiagnosed concussions can lead to worse postconcussion outcomes. It is not clear how those with a history of undiagnosed concussion perform on subsequent standard concussion baseline assessments. PURPOSE: To determine if previous concussion diagnosis status was associated with outcomes on the standard baseline concussion assessment battery. STUDY DESIGN: Cross-sectional study; Level of evidence, 3. METHODS: Concussion Assessment, Research, and Education (CARE) Consortium participants (N = 29,934) self-reported concussion history with diagnosis status and completed standard baseline concussion assessments, including assessments for symptoms, mental status, balance, and neurocognition. Multiple linear regression models were used to estimate mean differences and 95% CIs among concussion history groups (no concussion history [n = 23,037; 77.0%], all previous concussions diagnosed [n = 5315; 17.8%], ≥1 previous concussions undiagnosed [n = 1582; 5.3%]) at baseline for all outcomes except symptom severity and Brief Symptom Inventory-18 (BSI-18) score, in which negative binomial models were used to calculate incidence rate ratios (IRRs). All models were adjusted for sex, race, ethnicity, sport contact level, and concussion count. Mean differences with 95% CIs excluding 0.00 and at least a small effect size (≥0.20), and those IRRs with 95% CIs excluding 1.00 and at least a small association (IRR, ≥1.10) were considered significant. RESULTS: The ≥1 previous concussions undiagnosed group reported significantly greater symptom severity scores (IRR, ≥1.38) and BSI-18 (IRR, ≥1.31) scores relative to the no concussion history and all previous concussions diagnosed groups. The ≥1 previous concussions undiagnosed group performed significantly worse on 6 neurocognitive assessments while performing better on only 2 compared with the no concussion history and all previous concussions diagnosed groups. There were no between-group differences on mental status or balance assessments. CONCLUSION: An undiagnosed concussion history was associated with worse clinical indicators at future baseline assessments. Individuals reporting ≥1 previous undiagnosed concussions exhibited worse baseline clinical indicators. This may suggest that concussion-related harm may be exacerbated when injuries are not diagnosed.

Optimizing Concussion Care Seeking: Identification of Factors Predicting Previous Concussion Diagnosis Status.

PURPOSE: There is limited understanding of factors affecting concussion diagnosis status using large sample sizes. The study objective was to identify factors that can accurately classify previous concussion diagnosis status among collegiate student-athletes and service academy cadets with concussion history. METHODS: This retrospective study used support vector machine, Gaussian Naïve Bayes, and decision tree machine learning techniques to identify individual (e.g., sex) and institutional (e.g., academic caliber) factors that accurately classify previous concussion diagnosis status (all diagnosed vs 1+ undiagnosed) among Concussion Assessment, Research, and Education Consortium participants with concussion histories ( n = 7714). RESULTS: Across all classifiers, the factors examined enable >50% classification between previous diagnosed and undiagnosed concussion histories. However, across 20-fold cross validation, ROC-AUC accuracy averaged between 56% and 65% using all factors. Similar performance is achieved considering individual risk factors alone. By contrast, classifications with institutional risk factors typically did not distinguish between those with all concussions diagnosed versus 1+ undiagnosed; average performances using only institutional risk factors were almost always <58%, including confidence intervals for many groups <50%. Participants with more extensive concussion histories were more commonly classified as having one or more of those previous concussions undiagnosed. CONCLUSIONS: Although the current study provides preliminary evidence about factors to help classify concussion diagnosis status, more work is needed given the tested models' accuracy. Future work should include a broader set of theoretically indicated factors, at levels ranging from individual behavioral determinants to features of the setting in which the individual was injured.

Contextual associations represented both in neural networks and human behavior.

Contextual associations facilitate object recognition in human vision. However, the role of context in artificial vision remains elusive as does the characteristics that humans use to define context. We investigated whether contextually related objects (bicycle-helmet) are represented more similarly in convolutional neural networks (CNNs) used for image understanding than unrelated objects (bicycle-fork). Stimuli were of objects against a white background and consisted of a diverse set of contexts (N = 73). CNN representations of contextually related objects were more similar to one another than to unrelated objects across all CNN layers. Critically, the similarity found in CNNs correlated with human behavior across multiple experiments assessing contextual relatedness, emerging significant only in the later layers. The results demonstrate that context is inherently represented in CNNs as a result of object recognition training, and that the representation in the later layers of the network tap into the contextual regularities that predict human behavior.

Beliefs affecting concussion reporting among military cadets: advanced observations through machine learning.

BACKGROUND: Untreated concussions are an important health concern. The number of concussions sustained each year is difficult to pinpoint due to diverse reporting routes and many people not reporting. A growing body of literature investigates the motivations for concussion under-reporting, proposing ties with knowledge of concussion outcomes and concussion culture. The present work employs machine learning to identify trends in knowledge and willingness to self-report concussions. METHODS: 2,204 cadets completed a survey addressing athletic and pilot status, concussion symptoms and outcome beliefs, ethical beliefs, demographics, and reporting willingness. RESULTS: Clustering and non-negative matrix analysis identified connections to self-report willingness within: knowledge of symptoms, ethical beliefs, reporting requirements, and belief of long-term concussion outcomes. Support vector machine classification of cadet reporting likelihood reveals symptom and outcome knowledge may be inversely related to reporting among those rating ethics considerations as low, while heightened ethics may predict higher reporting likeliness overall. CONCLUSIONS: Machine-learning analysis bolsters prior theories on the importance of concussion culture in reporting and indicate more symptom knowledge may decrease willingness to report. Uniquely, our analysis indicated importance of ethical behavior may be associated with general concussion reporting willingness, inviting further consideration from healthcare practitioners seeking increased reporting.

Predictive Power of Head Impact Intensity Measures for Recognition Memory Performance.

Subconcussive head injuries are connected to both short-term cognitive changes and long-term neurodegeneration. Further study is required to understand what types of subconcussive impacts might prove detrimental to cognition. We studied cadets at the US Air Force Academy engaged in boxing and physical development, measuring head impact motions during exercise with accelerometers. These head impact measures were compared with post-exercise memory performance. Investigators explored multiple techniques for characterizing the magnitude of head impacts. Boxers received more head impacts and achieved lower performance in post-exercise memory than non-boxers. For several measures of impact motion, impact intensity appeared to set an upper bound on post-exercise memory performance - stronger impacts led to lower expected memory performance. This trend was most significant when impact intensity was measured through a novel technique, applying principal component analysis to boxer motion. Principal component analysis measures also captured more distinct impact information than seven traditional impact measures also tested.

A method for real-time visual stimulus selection in the study of cortical object perception.

The properties utilized by visual object perception in the mid- and high-level ventral visual pathway are poorly understood. To better establish and explore possible models of these properties, we adopt a data-driven approach in which we repeatedly interrogate neural units using functional Magnetic Resonance Imaging (fMRI) to establish each unit's image selectivity. This approach to imaging necessitates a search through a broad space of stimulus properties using a limited number of samples. To more quickly identify the complex visual features underlying human cortical object perception, we implemented a new functional magnetic resonance imaging protocol in which visual stimuli are selected in real-time based on BOLD responses to recently shown images. Two variations of this protocol were developed, one relying on natural object stimuli and a second based on synthetic object stimuli, both embedded in feature spaces based on the complex visual properties of the objects. During fMRI scanning, we continuously controlled stimulus selection in the context of a real-time search through these image spaces in order to maximize neural responses across pre-determined 1cm(3) rain regions. Elsewhere we have reported the patterns of cortical selectivity revealed by this approach (Leeds et al., 2014). In contrast, here our objective is to present more detailed methods and explore the technical and biological factors influencing the behavior of our real-time stimulus search. We observe that: 1) Searches converged more reliably when exploring a more precisely parameterized space of synthetic objects; 2) real-time estimation of cortical responses to stimuli is reasonably consistent; 3) search behavior was acceptably robust to delays in stimulus displays and subject motion effects. Overall, our results indicate that real-time fMRI methods may provide a valuable platform for continuing study of localized neural selectivity, both for visual object representation and beyond.

Exploration of complex visual feature spaces for object perception.

The mid- and high-level visual properties supporting object perception in the ventral visual pathway are poorly understood. In the absence of well-specified theory, many groups have adopted a data-driven approach in which they progressively interrogate neural units to establish each unit's selectivity. Such methods are challenging in that they require search through a wide space of feature models and stimuli using a limited number of samples. To more rapidly identify higher-level features underlying human cortical object perception, we implemented a novel functional magnetic resonance imaging method in which visual stimuli are selected in real-time based on BOLD responses to recently shown stimuli. This work was inspired by earlier primate physiology work, in which neural selectivity for mid-level features in IT was characterized using a simple parametric approach (Hung et al., 2012). To extend such work to human neuroimaging, we used natural and synthetic object stimuli embedded in feature spaces constructed on the basis of the complex visual properties of the objects themselves. During fMRI scanning, we employed a real-time search method to control continuous stimulus selection within each image space. This search was designed to maximize neural responses across a pre-determined 1 cm(3) brain region within ventral cortex. To assess the value of this method for understanding object encoding, we examined both the behavior of the method itself and the complex visual properties the method identified as reliably activating selected brain regions. We observed: (1) Regions selective for both holistic and component object features and for a variety of surface properties; (2) Object stimulus pairs near one another in feature space that produce responses at the opposite extremes of the measured activity range. Together, these results suggest that real-time fMRI methods may yield more widely informative measures of selectivity within the broad classes of visual features associated with cortical object representation.

Comparing visual representations across human fMRI and computational vision.

Feedforward visual object perception recruits a cortical network that is assumed to be hierarchical, progressing from basic visual features to complete object representations. However, the nature of the intermediate features related to this transformation remains poorly understood. Here, we explore how well different computer vision recognition models account for neural object encoding across the human cortical visual pathway as measured using fMRI. These neural data, collected during the viewing of 60 images of real-world objects, were analyzed with a searchlight procedure as in Kriegeskorte, Goebel, and Bandettini (2006): Within each searchlight sphere, the obtained patterns of neural activity for all 60 objects were compared to model responses for each computer recognition algorithm using representational dissimilarity analysis (Kriegeskorte et al., 2008). Although each of the computer vision methods significantly accounted for some of the neural data, among the different models, the scale invariant feature transform (Lowe, 2004), encoding local visual properties gathered from "interest points," was best able to accurately and consistently account for stimulus representations within the ventral pathway. More generally, when present, significance was observed in regions of the ventral-temporal cortex associated with intermediate-level object perception. Differences in model effectiveness and the neural location of significant matches may be attributable to the fact that each model implements a different featural basis for representing objects (e.g., more holistic or more parts-based). Overall, we conclude that well-known computer vision recognition systems may serve as viable proxies for theories of intermediate visual object representation.

A framework for the analysis of acoustical cardiac signals.

Skilled cardiologists perform cardiac auscultation, acquiring and interpreting heart sounds, by implicitly carrying out a sequence of steps. These include discarding clinically irrelevant beats, selectively tuning in to particular frequencies and aggregating information across time to make a diagnosis. In this paper, we formalize a series of analytical stages for processing heart sounds, propose algorithms to enable computers to approximate these steps, and investigate the effectiveness of each step in extracting relevant information from actual patient data. Through such reasoning, we provide insight into the relative difficulty of the various tasks involved in the accurate interpretation of heart sounds. We also evaluate the contribution of each analytical stage in the overall assessment of patients. We expect our framework and associated software to be useful to educators wanting to teach cardiac auscultation, and to primary care physicians, who can benefit from presentation tools for computer-assisted diagnosis of cardiac disorders. Researchers may also employ the comprehensive processing provided by our framework to develop more powerful, fully automated auscultation applications.