Faculty Bibliography

BACKGROUND: The care of traumatized children would benefit significantly from accurate predictive models for Posttraumatic Stress Disorder (PTSD), using information available around the time of trauma. Machine Learning (ML) computational methods have yielded strong results in recent applications across many diseases and data types, yet they have not been previously applied to childhood PTSD. Since these methods have not been applied to this complex and debilitating disorder, there is a great deal that remains to be learned about their application. The first step is to prove the concept: Can ML methods - as applied in other fields - produce predictive classification models for childhood PTSD? Additionally, we seek to determine if specific variables can be identified - from the aforementioned predictive classification models - with putative causal relations to PTSD. METHODS: ML predictive classification methods - with causal discovery feature selection - were applied to a data set of 163 children hospitalized with an injury and PTSD was determined three months after hospital discharge. At the time of hospitalization, 105 risk factor variables were collected spanning a range of biopsychosocial domains. RESULTS: Seven percent of subjects had a high level of PTSD symptoms. A predictive classification model was discovered with significant predictive accuracy. A predictive model constructed based on subsets of potentially causally relevant features achieves similar predictivity compared to the best predictive model constructed with all variables. Causal Discovery feature selection methods identified 58 variables of which 10 were identified as most stable. CONCLUSIONS: In this first proof-of-concept application of ML methods to predict childhood Posttraumatic Stress we were able to determine both predictive classification models for childhood PTSD and identify several causal variables. This set of techniques has great potential for enhancing the methodological toolkit in the field and future studies should seek to replicate, refine, and extend the results produced in this study.

Improvements in the quality of mental health care in the United States depend on the successful implementation of evidence-based treatments (EBT's) in typical settings of care. Unfortunately, there is little evidence that EBT's are used in ways that would approximate their established fidelity standards in such settings. This article describes an approach to more successful implementation of EBT's via a collaborative process between intervention developers and intervention users (e.g. providers, administrators, consumers) called Lead-user Innovation. Lead-user Innovation democratizes the implementation process by integrating the expertise of lead-users in the delivery, adaptation, innovation and evaluation of EBT's.

The primary aim of this study was to assess the prevalence of posttraumatic stress disorder (PTSD) in young children hospitalized for burns. A secondary aim was to assess partial PTSD in this population. PTSD diagnosis and symptoms were evaluated utilizing both the diagnostic interview for children and adolescents (DICA-P) module and the PTSD semistructured interview and observational report (PTSDSSI). PTSD symptomatology was assessed from parent interviews at 1 month after discharge from a major pediatric burn center. Four of the 42 participants who completed the DICA-P met full criteria for a diagnosis of PTSD (10%).Of the 39 who also completed the PTSDSSI, 1 (3%) met full criteria for PTSD diagnosis on that measure. Twenty-seven percentage of subjects met partial criteria on the DICA-P vs 16% on the PTSDSSI, without impairment. Furthermore, 67% of subjects met DICA-P criteria for the re-experiencing cluster and 54% met the PTSDSSI re-experiencing criteria. Although only a small percentage met full PTSD diagnostic criteria by either measure, a high percentage of young children with burns manifested some posttraumatic symptoms 1 month after discharge. Because PTSDSSI diagnosis is strongly linked to the diagnostic and statistical manual-5 criteria for "PTSD in children 6 years and younger," these results may offer clues to current diagnoses of PTSD in young children. Future research is needed to improve care by determining the risk factors and course of PTSD to further refine the diagnostic criteria for identifying children most in need of intervention, such as those hospitalized for burn injuries.

Background: Entropy has a fundamental relationship with information and the functioning of all computational systems. Entropy is defined as the number of states available to a system. A system with low entropy has access to fewer states than does one with high entropy. A system with low entropy is more ordered and more predicable than a system with high entropy. Since entropy is related to the functioning of computational systems, there is an emerging theoretical and empirical literature about its role in brain function and dysfunction. We present the results of three integrated studies applying resting state fMRI entropy measurement to understand intelligence, personality, and psychopathology. Brain entropy is an index of an individual's access to brain states at a given time and is measured through the predictivity of brain state over time. Thus, we would expect to observe brain entropic differences between conditions known to be associated with high flexibility (e.g. high intelligence, creativity, novelty seeking) vs. conditions associated with high rigidity (e.g. anxiety, depression, Posttraumatic Stress). The three studies are: Brain entropy and intelligence in 926 adults from the Brain Genomic Superstruct Project, 2. Brain entropy and personality in 926 adults from the Brain Genomic Superstruct Project, and 3. Brain entropy and PTSD in 95 veterans from the NYU Cohen Veterans Data Set. Methods: Subjects: Study 1 (Entropy and Intelligence) and Study 2 (Entropy and Personality) were conducted with data from the Brain Genomics Superstruct Project (BGSP). The BGSP includes 1570 healthy adult participants between the ages of 18 and 35. The current study utilized data from the 926 participants who completed intelligence and personality assessments. Study 3 (Entropy and PTSD) was conducted with data from the NYU Cohen Veterans Data Set. This data set includes 95 combat veterans, 46 with PTSD and 49 without PTSD. fMRI Procedures: Brain Genomics Superstruct Project (BGSP). All MRI data were obtained with 3T Trio scanners (Siemens Healthcare, Erlangen, Germany) at Harvard University and Massachusetts General Hospital. MRI scans for each participant included a high resolution structural scan (T1-weighted multi-echo MPRAGE, TR = 2.2 sec, TE = 1.5/3.4/5.2/7.0 msec, slices = 144, resolution = 1.2 x 1.2 x 1.2 mm) and a resting-state functional scan sensitive to blood oxygenation level-dependent (BOLD) contrast (TR = 3.0 sec, TE = 30 msec, slices = 47, resolution = 3.0 x 3.0 x 3.0 mm, 120 measurements). NYU Cohen Veterans Data Set: All MRI data were obtained with a 3T Trio scanner (Siemens AG, Erlangen Germany). Anatomical images were acquired with magnetization prepared rapid gradient echo sequence with TE/TI/TR = 2.98/900/2300 ms, 256 x 240 matrix, 256 mm x 240 mm fieldof-view, flip angle = 9degree, slice thickness = 1 mm and total slice number = 191; resting state fMRI was obtained using an echo-planar imaging sequence (TR/TE = 2000/29 ms, flip angle = 90degree), 64 x 64 matrix, pixel size 3.125 mm x 3.125 mm, total slice number = 32, slice thickness = 3.5 mm (without gaps), total volume number = 200. fMRI Entropy Analysis: Brain entropy was calculated using the Brain Entropy Mapping Toolbox (BENtbx) (Wang et al, 2014) for MATLAB (MATLAB Release R2015b, The MathWorks Inc., Natick, MA, United States). The BENtbx utilizes Sample Entropy (SampEn). For a given time series, SampEn is a single number representing the predictability of the series. The entropy of highly predictable series is small, close to 0, indicating a lack of variation or disorder. The entropy of unpredictable series is large, indicating a high amount of variation or disorder. The Sample Entropy process first breaks a series into smaller sets of size m. For example, for m = 2, and the BOLD time series is broken into pairs of consecutive values. Each pair is then compared with every other pair to find the maximum distance (absolute value difference) between any number in the first pair and any number in the second pair. If the distance is less than the threshold r, the two pairs are considered a 'match.' This process is then repeated for sets of size m + 1. Sample Entropy is then the ratio: SampEn =-log A/B: Where, A = number of matches using sets of size m+1 and B = number of matches using sets of size m. For perfectly predictable series, A and B will be equal, and entropy will be 0. As disorder in a series increases, B will become greater than A, and the equation will yield an increasingly large positive number. Psychometric Measurement: Study 1: Intelligence was measured with the Shipley Estimated IQ, Vocabulary, and Matrix Reasoning scales. Study 2: Personality was measured for Behavioral Inhibition, Harm Avoidance, Risk Taking, and Novelty Seeking. Study 3: PTSD was measured with the Clinician Administered PTSD Scale (CAPS). Results: Study 1: Shipley Estimated IQ, Vocabulary, and Matrix Reasoning were all associated with higher brain entropy. In particular, Vocabulary was related to higher entropy in the L fusiform gyrus, inferior temporal gyrus, parahippocampal gyrus. Matrix Reasoning was associated with higher entropy in the bilateral superior, medial, inferior frontal gyrus, bilateral orbital gyrus, and R middle frontal gyrus. Study 2: Harm avoidance and Behavioral Inhibition were associated with lower entropy and Novelty Seeking and Risk Taking were associated with higher entropy. Study 3: PTSD was associated with lower entropy, particularly in the L hippocampus and parahippocampal gyrus, inferior and middle temporal lobes: and higher entropy in the R precuneus, and R parietal lobe. Conclusions: Brain entropy may provide a novel approach to understand intelligence, personality, and psychopathology such as PTSD

Background: This presentation details the application of algorithms related to Complex Systems Science/Network Science, Causal Discovery, and Machine Learning Predictive Analytics and Intervention Modeling to understand the emergence and sustenance of Posttraumatic Stress Disorder (PTSD) in acutely traumatized children. There is a great need to develop new computational approaches to understand risk for PTSD given its complex etiology. Our application of these approaches is dedicated to identify children who are at highest risk for PTSD and to identify promising prevention and treatment targets. The research that will be presented examines risk and intervention targets with a longitudinal data set on acutely injured children. Methods: The data set is comprised of information on 163 children aged 7-18 collected as part of a National Institute of Mental Health funded study (R01 MH063247) on risk factors for PTSD in children hospitalized with injuries. The basic design follows: injured children were assessed within hours or days after their hospitalization and reassessed 3 months and 1 year following discharge. The data set includes variables measured during the hospitalization period and at each follow-up and includes domains such as early childhood development, demographics, school and social function, family stress, parent symptoms and functioning, psychosocial stress, qualities and magnitude of injury, candidate genes, neuroendocrine response, psychophysiologic response, and child symptoms and functioning. PTSD was measured with the UCLA PTSD Reaction Index. We apply a unique computational approach called the Complex Systems-Causal Network (CS-CN) method designed to discover sets of variables related to psychiatric disorders that together possess well-known properties of Complex Adaptive Systems (e.g. efficiency of information transfer, modularity, power-law scaling, robustness) and, if such properties are demonstrated, the variables that disproportionally contribute to the systems robust qualities are determined. We then apply Machine Learning Predictive Analytics with Causal Discovery Feature Selection and Intervention Modeling (Pearl's 'Do Calculus') to determine if PTSD can be predicted from variables measured around the time of the trauma, if any of the predictive variables have causal influence on the development of PTSD, and the effect on PTSD if intervention is modeled on any of the discovered causal variables. Results: The CS-CN method revealed a network of 110 variables and 166 bivariate relations that had strong adaptive properties compared with 1000 permutations of a random network. The variables that most contributed to its adaptive properties were CRHR1 gene, FKBP5 gene, age, socioeconomic status, and acute anxiety. Machine Learning analyses revealed an accurate and reliable predictive model for PTSD from variables measured at the time of trauma (AUC =.78) and modeling the influence of change (i.e. Judea Pearl's 'Do Calculus') in several 'remediable' causal variables (e.g. acute pain, pulse rate, anxiety, parent's symptoms of acute stress) led to reduction in PTSD symptoms. Conclusions: New computational methods can lead to reliable and accurate predictive models for PTSD and identify promising prevention and treatment targets