Application of a Novel Endpoint Staging Framework: Proof of Concept in the AMBAR Study.

Background: A theoretical endpoint staging framework was previously developed and published, aligning outcomes (i.e., memory) to the stage of Alzheimer's disease (AD) in which a given outcome is most relevant (i.e., has the greatest risk of degradation). The framework guides the selection of endpoints measuring outcomes relevant within a target AD population. Here, a proof of concept is presented via post-hoc analyses of the Alzheimer Management by Albumin Replacement (AMBAR) Phase 2b clinical trial in patients with AD (NCT01561053, 2012). Objective: To evaluate whether aligning endpoints measuring cognition, function, and quality of life to hypothesized 'target' stages of AD yields magnitudes of treatment efficacy greater than those reported in the AMBAR full analysis set (FAS). Methods: Three endpoints were tested: ADAS-Cog 12, ADCS-ADL, and QoL-AD. The magnitude of treatment efficacy was hypothesized to be maximized in the target stages of mild, mild-to-moderate, and very mild AD, respectively, compared to the full analysis set (FAS) and non-target stages. Results: For ADAS-Cog 12, the magnitude of treatment efficacy was largest in the non-target stage (-4.0, p = 0.0760) compared to target stage and FAS. For ADCS-ADL and QoL-AD, the magnitude of treatment efficacy was largest in the target stage (14.2, p = 0.0003; 2.4, p < 0.0001, respectively) compared to non-target stage and FAS. Conclusions: Findings indicated that evaluating endpoints in the most relevant AD stage can increase the magnitude of the observed treatment efficacy. Evidence provides preliminary proof of concept for the endpoint staging framework.

Age of first exposure to soccer heading: Associations with cognitive, clinical, and imaging outcomes in the Einstein Soccer Study.

Introduction: The objective of this study is to assess the role of age at first exposure (AFE) to soccer heading as a predictor of known adverse associations of recent and longer-term heading with brain microstructure, cognitive, and behavioral features among adult amateur soccer players. Methods: The sample included 276 active amateur soccer players (196 male and 81 female) aged 18-53 years old. AFE to soccer heading was treated as a binary variable, dichotomized at ≤ 10 years vs. >10 years old, based on a recently promulgated US Soccer policy, which bans heading for athletes ages 10 and under. Results: We found that soccer players who began heading at age 10 or younger performed better on tests of working memory (p = 0.03) and verbal learning (p = 0.02), while accounting for duration of heading exposure, education, sex, and verbal intelligence. No difference in brain microstructure or behavioral measures was observed between the two exposure groups. Discussion: The findings indicate that, among adult amateur soccer players, AFE to heading before age 10 compared to later start of heading, is not associated with adverse outcomes, and may be associated with better cognitive performance in young adulthood. Cumulative heading exposure across the lifespan, rather than early life exposure, may drive risk for adverse effects and should be the focus of future longitudinal studies to inform approaches to enhance player safety.

Optimized Acute Treatment of Migraine Is Associated With Greater Productivity in People With Migraine: Results From the Chronic Migraine Epidemiology and Outcomes (CaMEO) Study.

OBJECTIVE: This study aimed to ascertain whether level of optimization of acute treatment of migraine is related to work productivity across the spectrum of migraine. METHODS: Data were from the Chronic Migraine Epidemiology and Outcomes (CaMEO) Study, an internet-based longitudinal survey. Respondents with migraine who reported full-time employment and use of ≥1 acute prescription medication for migraine were included. We determined relationships among lost productive time (LPT; measured with the Migraine Disability Assessment Scale), acute treatment optimization (Migraine Treatment Optimization Questionnaire- ), and monthly headache days (MHDs). RESULTS: There was a direct relationship between LPT and MHD category. Greater acute treatment optimization was associated with lower total LPT, less absenteeism, and less presenteeism within each MHD category. CONCLUSIONS: Optimizing acute treatment for migraine may reduce LPT in people with migraine and reduce indirect costs.

Toward the optimized assessment of clinical outcomes in studies of novel treatments for Alzheimer's disease.

INTRODUCTION: Alzheimer's disease (AD) is characterized by a progressive decline in cognition and daily function, leading to a greater need for caregiver support. Clinical disease is segmented into a preclinical stage, mild cognitive impairment, and mild, moderate, and severe stages of Alzheimer's dementia. Although AD trials enroll participants at various stages of illness, treatment efficacy is often assessed using endpoints based on measures of outcomes that are held fixed across disease stages. We hypothesize that matching the primary outcomes measured in the endpoint hierarchy to the stage of disease targeted by the trial will increase the likelihood of detecting true treatment benefits. AREAS COVERED: We discuss current approaches to assessing clinical outcomes in AD trials, followed by a consideration of how effect detection can be improved by linking the stage of AD to the endpoints that most likely reflect stage-specific disease progression. EXPERT OPINION: Failing to account for stage-specific relevance and sensitivity of clinical outcomes may be one factor that contributes to trial failures in AD. Given the history of failure, experts have begun to scrutinize the relevance and sensitivity of outcomes as a potentially modifiable barrier to successful trials. To this end, we present a framework for refining trial endpoint selection and evaluation.

Repetitive soccer heading adversely impacts short-term learning among adult women.

OBJECTIVES: To determine the impact of 12-month heading exposure on short-term learning. DESIGN: A total of 105 active amateur soccer players, 45 women and 60 men, were administered an EMA-based test of working memory, a version of the two-back, once daily for 14 days. METHODS: Heading exposure of the participants was assessed using "HeadCount", a validated structured questionnaire at the baseline visits. The short-term rate of learning of each individual is quantified by first fitting a quadratic model to the daily performance on the two-back test over a two-week period, then taking the instantaneous rate of the quadratic function at the 7th test. A linear regression model was used to test the association of heading exposure with rates of learning, including age, sex, years of education and history of concussion as covariates, as well as variables describing soccer play and heading within the two-week period. Sensitivity analyses were performed using different methods for quantifying the learning effects and different transformations on 12-month heading exposure. RESULTS: Greater 12-month heading was associated with lower rates of learning among women (p = 0.008) but not among men (p = 0.74). CONCLUSIONS: We have identified evidence for an adverse, albeit subclinical, effect of soccer heading on brain function among young adult players, which selectively affects women in our sample.

Combining patient reported outcomes and EHR data to understand population level treatment needs: correcting for selection bias in the migraine signature study.

BACKGROUND: Electronic health records (EHR) data can be used to understand population level quality of care especially when supplemented with patient reported data. However, survey non-response can result in biased population estimates. As a case study, we demonstrate that EHR and survey data can be combined to estimate primary care population prescription treatment status for migraine stratified by migraine disability, without and with adjustment for survey non-response bias. We selected disability as it is associated with survey participation and patterns of prescribing for migraine. METHODS: A stratified random sample of Sutter Health adult primary care (PC) patients completed a digital survey about headache, migraine, and migraine related disability. The survey data from respondents with migraine were combined with their EHR data to estimate the proportion who had prescription orders for acute or preventive migraine treatments. Separate proportions were also estimated for those with mild disability (denoted "mild migraine") versus moderate to severe disability (denoted mod-severe migraine) without and with correction, using the inverse propensity weighting method, for non-response bias. We hypothesized that correction for non-response bias would result in smaller differences in proportions who had a treatment order by migraine disability status. RESULTS: The response rate among 28,268 patients was 8.2%. Among survey respondents, 37.2% had an acute treatment order and 16.8% had a preventive treatment order. The response bias corrected proportions were 26.2% and 11.6%, respectively, and these estimates did not differ from the total source population estimates (i.e., 26.4% for acute treatments, 12.0% for preventive treatments), validating the correction method. Acute treatment orders proportions were 32.3% for mild migraine versus 37.3% for mod-severe migraine and preventive treatment order proportions were 12.0% for mild migraine and 17.7% for mod-severe migraine. The response bias corrected proportions for acute treatments were 24.8% for mild migraine and 26.6% for mod-severe migraine and the proportions for preventive treatment were 8.1% for mild migraine and 12.0% for mod-severe migraine. CONCLUSIONS: In this study, we combined survey data with EHR data to better understand treatment needs among patients diagnosed with migraine. Migraine-related disability is directly related to preventive treatment orders but less so for acute treatments. Estimates of treatment status by self-reported disability status were substantially over-estimated among those with moderate to severe migraine-related disability without correction for non-response bias.

Imputation of missing values for electronic health record laboratory data.

Laboratory data from Electronic Health Records (EHR) are often used in prediction models where estimation bias and model performance from missingness can be mitigated using imputation methods. We demonstrate the utility of imputation in two real-world EHR-derived cohorts of ischemic stroke from Geisinger and of heart failure from Sutter Health to: (1) characterize the patterns of missingness in laboratory variables; (2) simulate two missing mechanisms, arbitrary and monotone; (3) compare cross-sectional and multi-level multivariate missing imputation algorithms applied to laboratory data; (4) assess whether incorporation of latent information, derived from comorbidity data, can improve the performance of the algorithms. The latter was based on a case study of hemoglobin A1c under a univariate missing imputation framework. Overall, the pattern of missingness in EHR laboratory variables was not at random and was highly associated with patients' comorbidity data; and the multi-level imputation algorithm showed smaller imputation error than the cross-sectional method.

Electronic Health Record Technology Designed for the Clinical Encounter: MS NeuroShare.

OBJECTIVE: Advances in medical discoveries have bolstered expectations of precise and complete care, but delivering on such a promise for complex, chronic neurologic care delivery requires solving last-mile challenges. We describe the iterative human-centered design and pilot process for multiple sclerosis (MS) NeuroShare, a digital health solution that brings practical information to the point of care so that clinicians and patients with MS can view, discuss, and make informed decisions together. METHODS: We initiated a comprehensive human-centered process to iteratively design, develop, and implement a digital health solution for managing MS in the routine outpatient setting of the nonprofit Sutter Health system in Northern California. The human-centered codesign process included 3 phases: discovery and design, development, and implementation and pilot. Stakeholders included Sutter Health's Research Development and Dissemination team, academic domain experts, neurologists, patients with MS, and an advisory group. RESULTS: MS NeuroShare went live in November 2018. It included a patient- and clinician-facing web application that launches from the electronic health record, visually displays a patient's data relevant to MS, and prompts the clinician to comprehensively evaluate and treat the patient. Both patients and clinicians valued the ability to jointly view patient-generated and other data. Preliminary results suggest that MS NeuroShare promotes patient-clinician communication and more active patient participation in decision-making. CONCLUSIONS: Lessons learned in the design and implementation of MS NeuroShare are broadly applicable to the design and implementation of digital tools aiming to improve the experience of delivering and receiving high-quality care for complex, neurologic conditions across large health systems.

The migraine signature study: Methods and baseline results.

OBJECTIVE: To characterize patients who utilize services for migraine in a large integrated health care network, and describe patterns of care and utilization. BACKGROUND: Within health care systems, migraine is a common reason for seeking primary and neurology care, but relatively little is documented about who seeks care and the factors that explain variation in utilization. METHODS: We conducted a retrospective cohort study using electronic health record (EHR) data from Sutter Health primary care (PC) patients who had at least one office visit to a PC clinic between 2013 and 2017. Migraine status was ascertained from diagnosis codes and medication orders. Control status was assigned to those with no evidence of care for any type of headache. We divided the primary care migraine cohort into two groups: those who received all their care for migraine from PC (denoted PC-M) and those who had ≥1 encounter with a neurologist for migraine (denoted N-M). Migraine cases were also designated as having preexisting migraine if they had an encounter with a migraine diagnosis within (±) 6 months of their first study period PC visit and, otherwise, designated as first migraine consult. Two levels of contrasts included: patients with migraine and controls; and within the group of patients with migraine, PC-M and N-M groups. Comorbid conditions were determined from EHR encounter diagnosis codes. RESULTS: We identified 94,149 patients with migraine (including 21,525 N-M and 72,624 PC-M) and 1,248,763 controls. Comorbidities: Proportions of psychiatric [29.8% (n = 28,054) vs. 11.8% (n = 147,043)], autoimmune [(4.4% (n = 4162) vs. 2.6% (n = 31,981)], pain [13.2% (n = 12,439) vs. 5.8% (n = 72,049)], respiratory [24.6% (n = 23,186) vs. 12.3% (n = 153,692)], neurologic [2.9% (n = 2688) vs. 0.9% (n = 11,321)], and cerebrovascular [1.0% (n = 945) vs. 0.6% (n = 7500)] conditions were higher in the migraine group compared to controls, all p < 0.001. Among patients with migraine, the N-M group was similar to the PC-M group in sex, age, ethnicity, and marital status, but were more likely to have preexisting migraine (49.9% (n = 10,734) vs. 36.2% (n = 26,317), p < 0.001). Proportions of comorbid conditions were higher among the N-M group than the PC-M group {psychiatric [38.5% (n = 8291) vs. 27.2% (n = 19,763)], autoimmune [6.3% (n = 1365) vs. 3.9% (n = 2797)], pain [19.6% (n = 4218) vs. 11.3% (n = 8211)], respiratory [30.3% (n = 6516) vs. 23.0% (n = 16,670)], neurologic [6.0% (n = 1288) vs. 1.9% (n = 1400)], cardiovascular [9.7% (n = 2091) vs. 7.0% (n = 5076)], and cerebrovascular [2.3% (n = 500) vs. 0.6% (n = 445)], all p < 0.001}. Medications: During the study period, 82.6% (n = 77,762) of patients with migraine received ≥1 prescription order for an acute migraine medication [89.4% (n = 19,250) of N-M vs. 80.6% (n = 58,512) of PC]. Opioids were prescribed to 52.9% (n = 49,837) of patients with migraine [63.5% (n = 13,669) for N-M and 49.8% (n = 36,168) for PC-M patients). During the study period, 61.4% (n = 57,810) of patients received ≥1 prescription for a migraine preventive medication [81.4% (n = 17,521) of N-M and 55.5% (n = 40,289) of PC-M patients]. The most commonly prescribed classes of preventive medications were antidepressants. CONCLUSIONS: Among patients with migraine in a large health system, those who were also cared for in neurology were more likely to receive both acute and preventive medication migraine orders than those patients who did not see a neurologist, with triptans and antidepressants the most commonly prescribed classes of acute and preventive pharmacotherapies, respectively. Opioids were prescribed to approximately half of the total sample and more common in the N-M group. Adjusting for demographics, patients with migraine had higher rates of nearly every comorbidity we assessed and were more likely to utilize services compared to those without migraine. Overall, patients with migraine also cared for in neurology practices used more of all health care resource types under consideration and had more medical issues, which may be due in some part to a more severe, frequent and disabling disease state compared to those who sought care exclusively from PC practices.

Persistent Cardiometabolic Health Gaps: Can Therapeutic Care Gaps Be Precisely Identified from Electronic Health Records.

The objective of this study was to determine the strengths and limitations of using structured electronic health records (EHR) to identify and manage cardiometabolic (CM) health gaps. We used medication adherence measures derived from dispense data to attribute related therapeutic care gaps (i.e., no action to close health gaps) to patient- (i.e., failure to retrieve medication or low adherence) or clinician-related (i.e., failure to initiate/titrate medication) behavior. We illustrated how such data can be used to manage health and care gaps for blood pressure (BP), low-density lipoprotein cholesterol (LDL-C), and HbA1c for 240,582 Sutter Health primary care patients. Prevalence of health gaps was 44% for patients with hypertension, 33% with hyperlipidemia, and 57% with diabetes. Failure to retrieve medication was common; this patient-related care gap was highly associated with health gaps (odds ratios (OR): 1.23-1.76). Clinician-related therapeutic care gaps were common (16% for hypertension, and 40% and 27% for hyperlipidemia and diabetes, respectively), and strongly related to health gaps for hyperlipidemia (OR = 5.8; 95% CI: 5.6-6.0) and diabetes (OR = 5.7; 95% CI: 5.4-6.0). Additionally, a substantial minority of care gaps (9% to 21%) were uncertain, meaning we lacked evidence to attribute the gap to either patients or clinicians, hindering efforts to close the gaps.

An electronic, unsupervised patient-reported Expanded Disability Status Scale for multiple sclerosis.

BACKGROUND: In persons with multiple sclerosis (MS), the Expanded Disability Status Scale (EDSS) is the criterion standard for assessing disability, but its in-person nature constrains patient participation in research and clinical assessments. OBJECTIVE: The aim of this study was to develop and validate a scalable, electronic, unsupervised patient-reported EDSS (ePR-EDSS) that would capture MS-related disability across the spectrum of severity. METHODS: We enrolled 136 adult MS patients, split into a preliminary testing Cohort 1 (n = 50), and a validation Cohort 2 (n = 86), which was evenly distributed across EDSS groups. Each patient completed an ePR-EDSS either immediately before or after a MS clinician's Neurostatus EDSS evaluation. RESULTS: In Cohort 2, mean age was 50.6 years (range = 26-80) and median EDSS was 3.5 (interquartile range (IQR) = [1.5, 5.5]). The ePR-EDSS and EDSS agreed within 1-point for 86% of examinations; kappa for agreement within 1-point was 0.85 (p < 0.001). The correlation coefficient between the two measures was 0.91 (<0.001). DISCUSSION: The ePR-EDSS was highly correlated with EDSS, with good agreement even at lower EDSS levels. For clinical care, the ePR-EDSS could enable the longitudinal monitoring of a patient's disability. For research, it provides a valid and rapid measure across the entire spectrum of disability and permits broader participation with fewer in-person assessments.

Framing potential for adverse effects of repetitive subconcussive impacts in soccer in the context of athlete and non-athlete controls.

The benefits of athletic activity may be attenuated by sport-related head impacts, including soccer-related concussion and subconcussive events. The purpose of this study is to characterize the specific effects of soccer heading on white matter microstructure and cognitive function, independent of concussion, relative to non-athlete controls and relative to active athletes who are not involved in collision sports. 246 amateur soccer players, 72 non-contact/non-collision sports athletes and 110 healthy,non-athlete controls were included in the study, and underwent cognitive testing and 3T diffusion tensor imaging (DTI). Voxelwise linear regression, comparing soccer players and non-contact/non-collision sports athletes healthy,non-athlete controls, identified regions of abnormally low and high fractional anisotropy (FA), axial diffusivity (AD), radial diffusivity (RD) and mean diffusivity (MD) in athlete participants. Generalized estimating equations were used to examine the effects of 2 week and 1 year heading exposure quartile on cognitive performance and on the volume of each high and each low DTI parameter. Athletes with no or lower exposure to repetitive heading exhibited greater expression of low RD, greater expression of high FA and better performance on tasks of attention, processing speed, verbal memory, and working memory compared to non-athletes. Soccer players with the highest exposure to repetitive head impacts, however, did not differ significantly from healthy, non-athletes on either micro-structural features or cognitive performance, findings not explained by concussion history or demographic factors. These results are consistent with the notion that beneficial effects of athletic conditioning or training on brain structure and function may be attenuated by exposure to repeated subconcussive head impacts.

Associations of Apolipoprotein E ε4 Genotype and Ball Heading With Verbal Memory in Amateur Soccer Players.

Importance: Emerging evidence suggests that long-term exposure to ball heading in soccer, the most popular sport in the world, confers risk for adverse cognitive outcomes. However, the extent to which the apolipoprotein E ε4 (APOE ε4) allele, a common risk factor for neurodegeneration, and ball heading are associated with cognition in soccer players remains unknown. Objective: To determine whether the APOE ε4 allele and 12-month ball heading exposure are associated with verbal memory in a cohort of adult amateur soccer players. Design, Settings, and Participants: A total of 379 amateur soccer players were enrolled in the longitudinal Einstein Soccer Study from November 11, 2013, through January 23, 2018. Selection criteria included participation in soccer for more than 5 years and for more than 6 months per year. Of the 379 individuals enrolled in the study, 355 were genotyped. Three players were excluded for reporting extreme levels of heading. Generalized estimating equation linear regression models were employed to combine data across visits for a cross-sectional analysis of the data. Exposures: At each study visit every 3 to 6 months, players completed the HeadCount 12-Month Questionnaire, a validated, computer-based questionnaire to estimate 12-month heading exposure that was categorized as low (quartiles 1 and 2), moderate (quartile 3), and high (quartile 4). Main Outcome and Measures: Verbal memory was assessed at each study visit using the International Shopping List Delayed Recall task from CogState. Results: A total of 352 soccer players (256 men and 96 women; median age, 23 years [interquartile range, 21-28 years]) across a total of 1204 visits were analyzed. High levels of heading were associated with worse verbal memory performance (ß = -0.59; 95% CI, -0.93 to -0.25; P = .001). There was no main association of APOE ε4 with verbal memory (ß = 0.09; 95% CI, -0.24 to 0.42; P = .58). However, there was a significant association of APOE ε4 and heading with performance on the ISRL task (χ2 = 7.22; P = .03 for overall interaction). In APOE ε4-positive players, poorer verbal memory associated with high vs low heading exposure was 4.1-fold greater (APOE ε4 negative, ß = -0.36; 95% CI, -0.75 to 0.03; APOE ε4 positive, ß = -1.49; 95% CI, -2.05 to -0.93), and poorer verbal memory associated with high vs moderate heading exposure was 8.5-fold greater (APOE ε4 negative, ß = -0.13; 95% CI, -0.54 to 0.29; APOE ε4 positive, ß = -1.11, 95% CI, -1.70 to -0.53) compared with that in APOE ε4-negative players. Conclusions and Relevance: This study suggests that the APOE ε4 allele is a risk factor for worse memory performance associated with higher heading exposure in the prior year, which highlights that assessing genetic risks may ultimately play a role in promoting safer soccer play.

TASTE: Temporal and Static Tensor Factorization for Phenotyping Electronic Health Records.

Phenotyping electronic health records (EHR) focuses on defining meaningful patient groups (e.g., heart failure group and diabetes group) and identifying the temporal evolution of patients in those groups. Tensor factorization has been an effective tool for phenotyping. Most of the existing works assume either a static patient representation with aggregate data or only model temporal data. However, real EHR data contain both temporal (e.g., longitudinal clinical visits) and static information (e.g., patient demographics), which are difficult to model simultaneously. In this paper, we propose Temporal And Static TEnsor factorization (TASTE) that jointly models both static and temporal information to extract phenotypes. TASTE combines the PARAFAC2 model with non-negative matrix factorization to model a temporal and a static tensor. To fit the proposed model, we transform the original problem into simpler ones which are optimally solved in an alternating fashion. For each of the sub-problems, our proposed mathematical re-formulations lead to efficient sub-problem solvers. Comprehensive experiments on large EHR data from a heart failure (HF) study confirmed that TASTE is up to 14× faster than several baselines and the resulting phenotypes were confirmed to be clinically meaningful by a cardiologist. Using 60 phenotypes extracted by TASTE, a simple logistic regression can achieve the same level of area under the curve (AUC) for HF prediction compared to a deep learning model using recurrent neural networks (RNN) with 345 features.

Personality and Risk Taking in Sports: A Focus on Unintentional and Intentional Head Impacts in Amateur Soccer Players.

OBJECTIVE: In soccer, unintentional and intentional (heading) head impacts are associated with concussive symptoms and cognitive dysfunction. We examined whether personality traits were associated with these behaviors in soccer players. DESIGN: Cross-sectional study. SETTING AND PARTICIPANTS: Participants completed study visits at the Albert Einstein College of Medicine. A total of 307 adult amateur soccer players, recruited from New York City and the surrounding area, completed 737 HeadCount-2w questionnaires. PREDICTOR VARIABLES: Personality traits (intellect/imagination, conscientiousness, extraversion, agreeableness, and neuroticism) were assessed with the Mini-International Personality Item Pool questionnaire at the baseline study visit. MAIN OUTCOME MEASURES: Participants completed an online questionnaire (HeadCount-2w) to ascertain frequency of intentional head impacts and occurrence of unintentional head impacts every 3 to 6 months. Generalized estimating equations repeated-measures regressions determined whether personality predicted unintentional and intentional impacts. RESULTS: Personality traits were not associated with unintentional head impact(s) or frequency of intentional head impacts. CONCLUSIONS: These findings have important clinical implications, suggesting that personality is not driving the association between high levels of unintentional and intentional head impacts and worse neuropsychological functioning and concussive symptoms.

Using the PARAFAC2 tensor factorization on EHR audit data to understand PCP desktop work.

BACKGROUND: Activity or audit log data are required for EHR privacy and security management but may also be useful for understanding desktop workflow. OBJECTIVE: We determined if the EHR audit log file, a rich source of complex time-stamped data on desktop activities, could be processed to derive primary care provider (PCP) level workflow measures. METHODS: We analyzed audit log data on 876 PCPs across 17,455 ambulatory care encounters that generated 578,394 time-stamped records. Each individual record represents a user interaction (e.g., point and click) that reflects all or part of a specific activity (e.g., order entry access). No dictionary exists to define how to combine clusters of sequential audit log records to represent identifiable PCP tasks. We determined if PARAFAC2 tensor factorization could: (1) learn to identify audit log record clusters that specifically represent defined PCP tasks; and (2) identify variation in how tasks are completed without the need for ground-truth labels. To interpret the result, we used the following PARAFAC2 factors: a matrix representing the task definitions and a matrix containing the frequency measure of each task for each encounter. RESULTS: PARAFAC2 automatically identified 4 clusters of audit log records that represent 4 common clinical encounter tasks: (1) medications' access, (2) notes' access, (3) order entry access, and (4) diagnosis modification. PARAFAC2 also identified the most common variants in how PCPs accomplish these tasks. It discovered variation in how the notes' access task was done, including identification of 9 distinct variants of notes access that explained 77% of the input data variation for notes. The discovered variants mapped to two known workflows for notes' access and to two distinct PCP user groups who accessed notes by either using the Visit Navigator or the Wrap-Up option. CONCLUSIONS: Our results demonstrate that EHR audit log data can be rapidly processed to create higher-level constructed features that represent time-stamped PCP tasks.

Recurrent Neural Networks for Early Detection of Heart Failure From Longitudinal Electronic Health Record Data: Implications for Temporal Modeling With Respect to Time Before Diagnosis, Data Density, Data Quantity, and Data Type.

BACKGROUND: We determined the impact of data volume and diversity and training conditions on recurrent neural network methods compared with traditional machine learning methods. METHODS AND RESULTS: Using longitudinal electronic health record data, we assessed the relative performance of machine learning models trained to detect a future diagnosis of heart failure in primary care patients. Model performance was assessed in relation to data parameters defined by the combination of different data domains (data diversity), the number of patient records in the training data set (data quantity), the number of encounters per patient (data density), the prediction window length, and the observation window length (ie, the time period before the prediction window that is the source of features for prediction). Data on 4370 incident heart failure cases and 30 132 group-matched controls were used. Recurrent neural network model performance was superior under a variety of conditions that included (1) when data were less diverse (eg, a single data domain like medication or vital signs) given the same training size; (2) as data quantity increased; (3) as density increased; (4) as the observation window length increased; and (5) as the prediction window length decreased. When all data domains were used, the performance of recurrent neural network models increased in relation to the quantity of data used (ie, up to 100% of the data). When data are sparse (ie, fewer features or low dimension), model performance is lower, but a much smaller training set size is required to achieve optimal performance compared with conditions where data are more diverse and includes more features. CONCLUSIONS: Recurrent neural networks are effective for predicting a future diagnosis of heart failure given sufficient training set size. Model performance appears to continue to improve in direct relation to training set size.

Erenumab in chronic migraine: Patient-reported outcomes in a randomized double-blind study.

OBJECTIVE: To determine the effect of erenumab, a human monoclonal antibody targeting the calcitonin gene-related peptide receptor, on health-related quality of life (HRQoL), headache impact, and disability in patients with chronic migraine (CM). METHODS: In this double-blind, placebo-controlled study, 667 adults with CM were randomized (3:2:2) to placebo or erenumab (70 or 140 mg monthly). Exploratory endpoints included migraine-specific HRQoL (Migraine-Specific Quality-of-Life Questionnaire [MSQ]), headache impact (Headache Impact Test-6 [HIT-6]), migraine-related disability (Migraine Disability Assessment [MIDAS] test), and pain interference (Patient-Reported Outcomes Measurement Information System [PROMIS] Pain Interference Scale short form 6b). RESULTS: Improvements were observed for all endpoints in both erenumab groups at month 3, with greater changes relative to placebo observed at month 1 for many outcomes. All 3 MSQ domains were improved from baseline with treatment differences for both doses exceeding minimally important differences established for MSQ-role function-restrictive (≥3.2) and MSQ-emotional functioning (≥7.5) and for MSQ-role function-preventive (≥4.5) for erenumab 140 mg. Changes from baseline in HIT-6 scores at month 3 were -5.6 for both doses vs -3.1 for placebo. MIDAS scores at month 3 improved by -19.4 days for 70 mg and -19.8 days for 140 mg vs -7.5 days for placebo. Individual-level minimally important difference was achieved by larger proportions of erenumab-treated participants than placebo for all MSQ domains and HIT-6. Lower proportions of erenumab-treated participants had MIDAS scores of severe (≥21) or very severe (≥41) or PROMIS scores ≥60 at month 3. CONCLUSIONS: Erenumab-treated patients with CM experienced clinically relevant improvements across a broad range of patient-reported outcomes. CLINICALTRIALSGOV IDENTIFIER: NCT02066415. CLASSIFICATION OF EVIDENCE: This study provides Class II evidence that for patients with CM, erenumab treatment improves HRQoL, headache impact, and disability.

Predicting need for advanced illness or palliative care in a primary care population using electronic health record data.

Timely outreach to individuals in an advanced stage of illness offers opportunities to exercise decision control over health care. Predictive models built using Electronic health record (EHR) data are being explored as a way to anticipate such need with enough lead time for patient engagement. Prior studies have focused on hospitalized patients, who typically have more data available for predicting care needs. It is unclear if prediction driven outreach is feasible in the primary care setting. In this study, we apply predictive modeling to the primary care population of a large, regional health system and systematically examine the impact of technical choices, such as requiring a minimum number of health care encounters (data density requirements) and aggregating diagnosis codes using Clinical Classifications Software (CCS) groupings to reduce dimensionality, on model performance in terms of discrimination and positive predictive value. We assembled a cohort of 349,667 primary care patients between 65 and 90 years of age who sought care from Sutter Health between July 1, 2011 and June 30, 2014, of whom 2.1% died during the study period. EHR data comprising demographics, encounters, orders, and diagnoses for each patient from a 12 month observation window prior to the point when a prediction is made were extracted. L1 regularized logistic regression and gradient boosted tree models were fit to training data and tuned by cross validation. Model performance in predicting one year mortality was assessed using held-out test patients. Our experiments systematically varied three factors: model type, diagnosis coding, and data density requirements. We found substantial, consistent benefit from using gradient boosting vs logistic regression (mean AUROC over all other technical choices of 84.8% vs 80.7% respectively). There was no benefit from aggregation of ICD codes into CCS code groups (mean AUROC over all other technical choices of 82.9% vs 82.6% respectively). Likewise increasing data density requirements did not affect discrimination (mean AUROC over other technical choices ranged from 82.5% to 83%). We also examine model performance as a function of lead time, which is the interval between death and when a prediction was made. In subgroup analysis by lead time, mean AUROC over all other choices ranged from 87.9% for patients who died within 0 to 3 months to 83.6% for those who died 9 to 12 months after prediction time.

Moderating Effects of Patient Characteristics on the Impact of Financial Incentives.

While financial incentives to providers or patients are increasingly common as a quality improvement strategy, their impact on patient subgroups and health care disparities is unclear. To examine these patterns, we analyzed data from a randomized clinical trial of financial incentives to lower low-density lipoprotein (LDL) cholesterol levels in patients at risk for cardiovascular disease. Patients with higher baseline LDL experienced greater cholesterol reductions in the shared incentive arm (0.23 mg/dL per unit change in baseline LDL, 95% CI [-0.46, -0.00]) but were also less likely to have medication potency increases in the physician incentive arm ( OR = 0.98, 95% CI [0.97, 0.996]). Uninsured patients and those of race other than Black or White were less likely to have potency increases in the shared incentive arm ( OR = 0.15, 95% CI [0.03, 0.70] and OR = 0.09, 95% CI [0.01, 0.93], respectively). These findings suggest some differential response to incentives, particularly in the form of targeted medication changes.