Publications by Year: 2020

Word embeddings are a popular approach to unsupervised learning of word relationships that are widely used in natural language processing. In this article, we present a new set of embeddings for medical concepts learned using an extremely large collection of multimodal medical data. Leaning on recent theoretical insights, we demonstrate how an insurance claims database of 60 million members, a collection of 20 million clinical notes, and 1.7 million full text biomedical journal articles can be combined to embed concepts into a common space, resulting in the largest ever set of embeddings for 108,477 medical concepts. To evaluate our approach, we present a new benchmark methodology based on statistical power specifically designed to test embeddings of medical concepts. Our approach, called cui2vec, attains state-of-the-art performance relative to previous methods in most instances. Finally, we provide a downloadable set of pre-trained embeddings for other researchers to use, as well as an online tool for interactive exploration of the cui2vec embeddings.

IMPORTANCE: Suicide is a leading cause of mortality, with suicide-related deaths increasing in recent years. Automated methods for individualized risk prediction have great potential to address this growing public health threat. To facilitate their adoption, they must first be validated across diverse health care settings.

OBJECTIVE: To evaluate the generalizability and cross-site performance of a risk prediction method using readily available structured data from electronic health records in predicting incident suicide attempts across multiple, independent, US health care systems.

DESIGN, SETTING, AND PARTICIPANTS: For this prognostic study, data were extracted from longitudinal electronic health record data comprising International Classification of Diseases, Ninth Revision diagnoses, laboratory test results, procedures codes, and medications for more than 3.7 million patients from 5 independent health care systems participating in the Accessible Research Commons for Health network. Across sites, 6 to 17 years' worth of data were available, up to 2018. Outcomes were defined by International Classification of Diseases, Ninth Revision codes reflecting incident suicide attempts (with positive predictive value >0.70 according to expert clinician medical record review). Models were trained using naive Bayes classifiers in each of the 5 systems. Models were cross-validated in independent data sets at each site, and performance metrics were calculated. Data analysis was performed from November 2017 to August 2019.

MAIN OUTCOMES AND MEASURES: The primary outcome was suicide attempt as defined by a previously validated case definition using International Classification of Diseases, Ninth Revision codes. The accuracy and timeliness of the prediction were measured at each site.

RESULTS: Across the 5 health care systems, of the 3 714 105 patients (2 130 454 female [57.2%]) included in the analysis, 39 162 cases (1.1%) were identified. Predictive features varied by site but, as expected, the most common predictors reflected mental health conditions (eg, borderline personality disorder, with odds ratios of 8.1-12.9, and bipolar disorder, with odds ratios of 0.9-9.1) and substance use disorders (eg, drug withdrawal syndrome, with odds ratios of 7.0-12.9). Despite variation in geographical location, demographic characteristics, and population health characteristics, model performance was similar across sites, with areas under the curve ranging from 0.71 (95% CI, 0.70-0.72) to 0.76 (95% CI, 0.75-0.77). Across sites, at a specificity of 90%, the models detected a mean of 38% of cases a mean of 2.1 years in advance.

CONCLUSIONS AND RELEVANCE: Across 5 diverse health care systems, a computationally efficient approach leveraging the full spectrum of structured electronic health record data was able to detect the risk of suicidal behavior in unselected patients. This approach could facilitate the development of clinical decision support tools that inform risk reduction interventions.

BACKGROUND: Intracranial hemorrhage (ICH) is a common and often devastating outcome in patients with brain tumors. Despite this, there is little evidence to guide anticoagulation management following an initial ICH event.

OBJECTIVES: To analyze the risk of recurrent hemorrhagic and thrombotic outcomes after an initial ICH event in patients with brain tumors and prior venous thromboembolism (VTE).

PATIENTS AND METHODS: A retrospective cohort study was performed. Radiographic images obtained after initial ICH were reviewed for the primary outcomes of recurrent ICH and VTE.

RESULTS AND CONCLUSIONS: A total of 79 patients with brain tumors who developed ICH on anticoagulation for VTE were analyzed. Fifty-four patients (68.4%) restarted anticoagulation following ICH. The cumulative incidence of recurrent ICH at 1 year was 6.1% (95% confidence interval [CI], 1.5-15.3) following reinitiation of anticoagulation. Following a major ICH (defined as an ICH >10 mL in size, causing symptoms, or requiring intervention), the rate of recurrent ICH upon reexposure to anticoagulation was 14.5% (95% CI, 2.1-38.35), whereas the rate of recurrent ICH following smaller ICH was 2.6% (95% CI, 0.2%-12.0%). Mortality following a recurrent ICH on anticoagulation was 67% at 30 days. The cumulative incidence of recurrent VTE was significantly lower in the restart cohort compared to patients who did not restart anticoagulation (8.1% vs 35.3%; P = .003). We conclude that resumption of anticoagulation is lowest among patients with metastatic brain tumors with small initial ICH. Following an initial major ICH, resumption of anticoagulation was associated with a high rate of recurrent ICH.

Several ongoing international efforts are developing methods of localizing single cells within organs or mapping the entire human body at the single cell level, including the Chan Zuckerberg Initiative's Human Cell Atlas (HCA), and the Knut and Allice Wallenberg Foundation's Human Protein Atlas (HPA), and the National Institutes of Health's Human BioMolecular Atlas Program (HuBMAP). Their goals are to understand cell specialization, interactions, spatial organization in their natural context, and ultimately the function of every cell within the body. In the same way that the Human Genome Project had to assemble sequence data from different people to construct a complete sequence, multiple centers around the world are collecting tissue specimens from diverse populations that vary in age, race, sex, and body size. A challenge will be combining these heterogeneous tissue samples into a 3D reference map that will enable multiscale, multidimensional Google Maps-like exploration of the human body. Key to making alignment of tissue samples work is identifying and using a coordinate system called a Common Coordinate Framework (CCF), which defines the positions, or "addresses," in a reference body, from whole organs down to functional tissue units and individual cells. In this perspective, we examine the concept of a CCF based on the vasculature and describe why it would be an attractive choice for mapping the human body.

BACKGROUND: Over the past decade, the emergence of several large federated clinical data networks has enabled researchers to access data on millions of patients at dozens of health care organizations. Typically, queries are broadcast to each of the sites in the network, which then return aggregate counts of the number of matching patients. However, because patients can receive care from multiple sites in the network, simply adding the numbers frequently double counts patients. Various methods such as the use of trusted third parties or secure multiparty computation have been proposed to link patient records across sites. However, they either have large trade-offs in accuracy and privacy or are not scalable to large networks.

OBJECTIVE: This study aims to enable accurate estimates of the number of patients matching a federated query while providing strong guarantees on the amount of protected medical information revealed.

METHODS: We introduce a novel probabilistic approach to running federated network queries. It combines an algorithm called HyperLogLog with obfuscation in the form of hashing, masking, and homomorphic encryption. It is tunable, in that it allows networks to balance accuracy versus privacy, and it is computationally efficient even for large networks. We built a user-friendly free open-source benchmarking platform to simulate federated queries in large hospital networks. Using this platform, we compare the accuracy, k-anonymity privacy risk (with k=10), and computational runtime of our algorithm with several existing techniques.

RESULTS: In simulated queries matching 1 to 100 million patients in a 100-hospital network, our method was significantly more accurate than adding aggregate counts while maintaining k-anonymity. On average, it required a total of 12 kilobytes of data to be sent to the network hub and added only 5 milliseconds to the overall federated query runtime. This was orders of magnitude better than other approaches, which guaranteed the exact answer.

CONCLUSIONS: Using our method, it is possible to run highly accurate federated queries of clinical data repositories that both protect patient privacy and scale to large networks.

We leveraged the largely untapped resource of electronic health record data to address critical clinical and epidemiological questions about Coronavirus Disease 2019 (COVID-19). To do this, we formed an international consortium (4CE) of 96 hospitals across five countries (www.covidclinical.net). Contributors utilized the Informatics for Integrating Biology and the Bedside (i2b2) or Observational Medical Outcomes Partnership (OMOP) platforms to map to a common data model. The group focused on temporal changes in key laboratory test values. Harmonized data were analyzed locally and converted to a shared aggregate form for rapid analysis and visualization of regional differences and global commonalities. Data covered 27,584 COVID-19 cases with 187,802 laboratory tests. Case counts and laboratory trajectories were concordant with existing literature. Laboratory tests at the time of diagnosis showed hospital-level differences equivalent to country-level variation across the consortium partners. Despite the limitations of decentralized data generation, we established a framework to capture the trajectory of COVID-19 disease in patients and their response to interventions.