Faculty Bibliography

Background/UNASSIGNED:The opioid epidemic in the United States is averaging over 100 deaths per day due to overdose. The effectiveness of opioids as pain treatments, and the drug-seeking behavior of opioid addicts, leads physicians in the United States to issue over 200 million opioid prescriptions every year. To better understand the biomedical profile of opioid-dependent patients, we analyzed information from electronic health records (EHR) including lab tests, vital signs, medical procedures, prescriptions, and other data from millions of patients to predict opioid substance dependence. Results/UNASSIGNED:We trained a machine learning model to classify patients by likelihood of having a diagnosis of substance dependence using EHR data from patients diagnosed with substance dependence, along with control patients with no history of substance-related conditions, matched by age, gender, and status of HIV, hepatitis C, and sickle cell disease. The top machine learning classifier using all features achieved a mean area under the receiver operating characteristic (AUROC) curve of ~ 92%, and analysis of the model uncovered associations between basic clinical factors and substance dependence. Additionally, diagnoses, prescriptions, and procedures prior to the diagnoses of substance dependence were analyzed to elucidate the clinical profile of substance-dependent patients, relative to controls. Conclusions/UNASSIGNED:The predictive model may hold utility for identifying patients at risk of developing dependence, risk of overdose, and opioid-seeking patients that report other symptoms in their visits to the emergency room.

Patient-generated health data (PGHD), collected from mobile apps and devices, represents an opportunity for remote patient monitoring and timely interventions to prevent acute exacerbations of chronic illness-if data are seen and shared by care teams. This case report describes the technical aspects of integrating data from a popular smartphone platform to a commonly used EHR vendor and explores the challenges and potential of this approach for disease management. Consented subjects using the Asthma Health app (built on Apple's ResearchKit platform) were able to share data on inhaler usage and peak expiratory flow rate (PEFR) with a local pulmonologist who ordered this data on Epic's EHR. For users who had installed and activated Epic's patient portal (MyChart) on their iPhone and enabled sharing of health data between apps via HealthKit, the pulmonologist could review PGHD and, if necessary, make recommendations. Four patients agreed to share data with their pulmonologist, though only two patients submitted more than one data point across the 4.5-month trial period. One of these patients submitted 101 PEFR readings across 65 days; another submitted 24 PEFR and inhaler usage readings across 66 days. PEFR for both patients fell within predefined physiologic parameters, except once where a low threshold notification was sent to the pulmonologist, who responded with a telephone discussion and new e-prescription to address symptoms. This research describes the technical considerations and implementation challenges of using commonly available frameworks for sharing PGHD, for the purpose of remote monitoring to support timely care interventions.

BACKGROUND:The misuse of prescription opioids (MUPO) is a leading public health concern. Social media are playing an expanded role in public health research, but there are few methods for estimating established epidemiological metrics from social media. The purpose of this study was to demonstrate that the geographic variation of social media posts mentioning prescription opioid misuse strongly correlates with government estimates of MUPO in the last month. METHODS:We wrote software to acquire publicly available tweets from Twitter from 2012 to 2014 that contained at least one keyword related to prescription opioid use (n = 3,611,528). A medical toxicologist and emergency physician curated the list of keywords. We used the semantic distance (SemD) to automatically quantify the similarity of meaning between tweets and identify tweets that mentioned MUPO. We defined the SemD between two words as the shortest distance between the two corresponding word-centroids. Each word-centroid represented all recognized meanings of a word. We validated this automatic identification with manual curation. We used Twitter metadata to estimate the location of each tweet. We compared our estimated geographic distribution with the 2013-2015 National Surveys on Drug Usage and Health (NSDUH). RESULTS:Tweets that mentioned MUPO formed a distinct cluster far away from semantically unrelated tweets. The state-by-state correlation between Twitter and NSDUH was highly significant across all NSDUH survey years. The correlation was strongest between Twitter and NSDUH data from those aged 18-25 (r = 0.94, p < 0.01 for 2012; r = 0.94, p < 0.01 for 2013; r = 0.71, p = 0.02 for 2014). The correlation was driven by discussions of opioid use, even after controlling for geographic variation in Twitter usage. CONCLUSIONS:Mentions of MUPO on Twitter correlate strongly with state-by-state NSDUH estimates of MUPO. We have also demonstrated that a natural language processing can be used to analyze social media to provide insights for syndromic toxicosurveillance.

With the proliferation of smartphones over the past several years, apps now play a prominent role in many social and work contexts, including medicine. This is enough of a phenomenon to have inspired the abbreviation “mHealth,” short for mobile health. The number of app-driven clinical calculators, checklists, and risk scores in common use in the emergency department (ED) has significantly increased and shows no sign of slowing. Thanks to this digital development and innovation, clinical decision support is now just a finger-tap away.

The feasibility of using mobile health applications to conduct observational clinical studies requires rigorous validation. Here, we report initial findings from the Asthma Mobile Health Study, a research study, including recruitment, consent, and enrollment, conducted entirely remotely by smartphone. We achieved secure bidirectional data flow between investigators and 7,593 participants from across the United States, including many with severe asthma. Our platform enabled prospective collection of longitudinal, multidimensional data (e.g., surveys, devices, geolocation, and air quality) in a subset of users over the 6-month study period. Consistent trending and correlation of interrelated variables support the quality of data obtained via this method. We detected increased reporting of asthma symptoms in regions affected by heat, pollen, and wildfires. Potential challenges with this technology include selection bias, low retention rates, reporting bias, and data security. These issues require attention to realize the full potential of mobile platforms in research and patient care.

In our recent Asthma Mobile Health Study (AMHS), thousands of asthma patients across the country contributed medical data through the iPhone Asthma Health App on a daily basis for an extended period of time. The collected data included daily self-reported asthma symptoms, symptom triggers, and real time geographic location information. The AMHS is just one of many studies occurring in the context of now many thousands of mobile health apps aimed at improving wellness and better managing chronic disease conditions, leveraging the passive and active collection of data from mobile, handheld smart devices. The ability to identify patient groups or patterns of symptoms that might predict adverse outcomes such as asthma exacerbations or hospitalizations from these types of large, prospectively collected data sets, would be of significant general interest. However, conventional clustering methods cannot be applied to these types of longitudinally collected data, especially survey data actively collected from app users, given heterogeneous patterns of missing values due to: 1) varying survey response rates among different users, 2) varying survey response rates over time of each user, and 3) non-overlapping periods of enrollment among different users. To handle such complicated missing data structure, we proposed a probability imputation model to infer missing data. We also employed a consensus clustering strategy in tandem with the multiple imputation procedure. Through simulation studies under a range of scenarios reflecting real data conditions, we identified favorable performance of the proposed method over other strategies that impute the missing value through low-rank matrix completion. When applying the proposed new method to study asthma triggers and symptoms collected as part of the AMHS, we identified several patient groups with distinct phenotype patterns. Further validation of the methods described in this paper might be used to identify clinically important patterns in large data sets with complicated missing data structure, improving the ability to use such data sets to identify at-risk populations for potential intervention.

BACKGROUND:Older adults are at risk for inadequate emergency department (ED) pain care. Unrelieved acute pain is associated with poor outcomes. Clinical decision support systems (CDSS) hold promise to improve patient care, but CDSS quality varies widely, particularly when usability evaluation is not employed. OBJECTIVE:To conduct an iterative usability and redesign process of a novel geriatric abdominal pain care CDSS. We hypothesized this process would result in the creation of more usable and favorable pain care interventions. METHODS:Thirteen emergency physicians familiar with the Electronic Health Record (EHR) in use at the study site were recruited. Over a 10-week period, 17 1-hour usability test sessions were conducted across 3 rounds of testing. Participants were given 3 patient scenarios and provided simulated clinical care using the EHR, while interacting with the CDSS interventions. Quantitative System Usability Scores (SUS), favorability scores and qualitative narrative feedback were collected for each session. Using a multi-step review process by an interdisciplinary team, positive and negative usability issues in effectiveness, efficiency, and satisfaction were considered, prioritized and incorporated in the iterative redesign process of the CDSS. Video analysis was used to determine the appropriateness of the CDS appearances during simulated clinical care. RESULTS:Over the 3 rounds of usability evaluations and subsequent redesign processes, mean SUS progressively improved from 74.8 to 81.2 to 88.9; mean favorability scores improved from 3.23 to 4.29 (1 worst, 5 best). Video analysis revealed that, in the course of the iterative redesign processes, rates of physicians' acknowledgment of CDS interventions increased, however most rates of desired actions by physicians (such as more frequent pain score updates) decreased. CONCLUSION/CONCLUSIONS:The iterative usability redesign process was instrumental in improving the usability of the CDSS; if implemented in practice, it could improve geriatric pain care. The usability evaluation process led to improved acknowledgement and favorability. Incorporating usability testing when designing CDSS interventions for studies may be effective to enhance clinician use.

When physicians search the web for personal information about their patients, others have argued that this undermines patients' trust, and the physician-patient relationship in general. We add that this practice also places other relationships at risk, and could jeopardize a physician's career. Yet there are also reports of web searches that have unambiguously helped in the care of patients, suggesting circumstances in which a routine search of the web could be beneficial. We advance the notion that, just as nonverbal cues and unsolicited information can be useful in clinical decision making, so too can online information from patients. As electronic records grow more voluminous and span more types of data, searching these resources will become a clinical skill, to be used judiciously and with care--just as evaluating the literature is, today. But to proscribe web searches of patients' information altogether is as nonsensical as disregarding findings from physical exams-instead, what's needed are guidelines for when to look and how to evaluate what's uncovered, online.