One-shot distributed algorithms for addressing heterogeneity in competing risks data across clinical sites.

OBJECTIVE: To characterize the interplay between multiple medical conditions across sites and account for the heterogeneity in patient population characteristics across sites within a distributed research network, we develop a one-shot algorithm that can efficiently utilize summary-level data from various institutions. By applying our proposed algorithm to a large pediatric cohort across four national Children's hospitals, we replicated a recently published prospective cohort, the RISK study, and quantified the impact of the risk factors associated with the penetrating or stricturing behaviors of pediatric Crohn's disease (PCD). METHODS: In this study, we introduce the ODACoRH algorithm, a one-shot distributed algorithm designed for the competing risks model with heterogeneity. Our approach considers the variability in baseline hazard functions of multiple endpoints of interest across different sites. To accomplish this, we build a surrogate likelihood function by combining patient-level data from the local site with aggregated data from other external sites. We validated our method through extensive simulation studies and replication of the RISK study to investigate the impact of risk factors on the PCD for adolescents and children from four children's hospitals within the PEDSnet, A National Pediatric Learning Health System. To evaluate our ODACoRH algorithm, we compared results from the ODACoRH algorithms with those from meta-analysis as well as those derived from the pooled data. RESULTS: The ODACoRH algorithm had the smallest relative bias to the gold standard method (-0.2%), outperforming the meta-analysis method (-11.4%). In the PCD association study, the estimated subdistribution hazard ratios obtained through the ODACoRH algorithms are identical on par with the results derived from pooled data, which demonstrates the high reliability of our federated learning algorithms. From a clinical standpoint, the identified risk factors for PCD align well with the RISK study published in the Lancet in 2017 and other published studies, supporting the validity of our findings. CONCLUSION: With the ODACoRH algorithm, we demonstrate the capability of effectively integrating data from multiple sites in a decentralized data setting while accounting for between-site heterogeneity. Importantly, our study reveals several crucial clinical risk factors for PCD that merit further investigations.

Pragmatic guidance for embedding pragmatic clinical trials in health plans: Large simple trials aren't so simple.

BACKGROUND: There are unique opportunities related to the design and conduct of pragmatic trials embedded in health insurance plans, which have longitudinal data on member/patient demographics, dates of coverage, and reimbursed medical care, including prescription drug dispensings, vaccine administrations, behavioral healthcare encounters, and some laboratory results. Such trials can be large and efficient, using these data to identify trial-eligible patients and to ascertain outcomes. METHODS: We use our experience primarily with the National Institutes of Health Pragmatic Trials Collaboratory Distributed Research Network, which comprises health plans that participate in the US Food & Drug Administration's Sentinel System, to describe lessons learned from the conduct and planning of embedded pragmatic trials. RESULTS: Information is available for research on more than 75 million people with commercial or Medicare Advantage health plans. We describe three studies that have used or plan to use the Network, as well as a single health plan study, from which we glean our lessons learned. CONCLUSIONS: Studies that are conducted in health plans provide much-needed evidence to drive clinically meaningful changes in care. However, there are many unique aspects of these trials that must be considered in the planning, implementation, and analytic phases. The type of trial best suited for studies embedded in health plans will be those that require large sample sizes, simple interventions that could be disseminated through health plans, and where data available to the health plan can be leveraged. These trials hold potential for substantial long-term impact on our ability to generate evidence to improve care and population health.

Performance of a computable phenotype for identification of patients with diabetes within PCORnet: The Patient-Centered Clinical Research Network.

PURPOSE: PCORnet, the National Patient-Centered Clinical Research Network, represents an innovative system for the conduct of observational and pragmatic studies. We describe the identification and validation of a retrospective cohort of patients with type 2 diabetes (T2DM) from four PCORnet sites. METHODS: We adapted existing computable phenotypes (CP) for the identification of patients with T2DM and evaluated their performance across four PCORnet sites (2012-2016). Patients entered the cohort on the earliest date they met one of three CP categories: (CP1) coded T2DM diagnosis (ICD-9/ICD-10) and an antidiabetic prescription, (CP2) diagnosis and glycosylated hemoglobin (HbA1c) ≥6.5%, or (CP3) an antidiabetic prescription and HbA1c ≥6.5%. We required evidence of health care utilization in each of the 2 prior years for each patient, as we also developed an incident T2DM CP to identify the subset of patients without documentation of T2DM in the 365 days before t0 . Among a systematic sample of patients, we calculated the positive predictive value (PPV) for the T2DM CP and incident-T2DM CP using electronic health record (EHR) review as reference. RESULTS: The CP identified 50 657 patients with T2DM. The PPV of patients randomly selected for validation was 96.2% (n = 1572; CI:95.1-97.0) and was consistently high across sites. The PPV for the incident-T2DM CP was 5.8% (CI:4.5-7.5). CONCLUSIONS: The T2DM CP accurately and efficiently identified patients with T2DM across multiple sites that participate in PCORnet, although the incident T2DM CP requires further study. PCORnet is a valuable data source for future epidemiological and comparative effectiveness research among patients with T2DM.

Factors affecting willingness to share electronic health data among California consumers.

BACKGROUND: Robust technology infrastructure is needed to enable learning health care systems to improve quality, access, and cost. Such infrastructure relies on the trust and confidence of individuals to share their health data for healthcare and research. Few studies have addressed consumers' views on electronic data sharing and fewer still have explored the dual purposes of healthcare and research together. The objective of the study is to explore factors that affect consumers' willingness to share electronic health information for healthcare and research. METHODS: This study involved a random-digit dial telephone survey of 800 adult Californians conducted in English and Spanish. Logistic regression was performed using backward selection to test for significant (p-value ≤ 0.05) associations of each explanatory variable with the outcome variable. RESULTS: The odds of consent for electronic data sharing for healthcare decreased as Likert scale ratings for EHR impact on privacy worsened, odds ratio (OR) = 0.74, 95% CI [0.60, 0.90]; security, OR = 0.80, 95% CI [0.66, 0.98]; and quality, OR = 0.59, 95% CI [0.46-0.75]. The odds of consent for sharing for research was greater for those who think EHR will improve research quality, OR = 11.26, 95% CI [4.13, 30.73]; those who value research benefit over privacy OR = 2.72, 95% CI [1.55, 4.78]; and those who value control over research benefit OR = 0.49, 95% CI [0.26, 0.94]. CONCLUSIONS: Consumers' choices about electronically sharing health information are affected by their attitudes toward EHRs as well as beliefs about research benefit and individual control. Design of person-centered interventions utilizing electronically collected health information, and policies regarding data sharing should address these values of importance to people. Understanding of these perspectives is critical for leveraging health data to support learning health care systems.

Learning competing risks across multiple hospitals: one-shot distributed algorithms.

OBJECTIVES: To characterize the complex interplay between multiple clinical conditions in a time-to-event analysis framework using data from multiple hospitals, we developed two novel one-shot distributed algorithms for competing risk models (ODACoR). By applying our algorithms to the EHR data from eight national children's hospitals, we quantified the impacts of a wide range of risk factors on the risk of post-acute sequelae of SARS-COV-2 (PASC) among children and adolescents. MATERIALS AND METHODS: Our ODACoR algorithms are effectively executed due to their devised simplicity and communication efficiency. We evaluated our algorithms via extensive simulation studies as applications to quantification of the impacts of risk factors for PASC among children and adolescents using data from eight children's hospitals including the Children's Hospital of Philadelphia, Cincinnati Children's Hospital Medical Center, Children's Hospital of Colorado covering over 6.5 million pediatric patients. The accuracy of the estimation was assessed by comparing the results from our ODACoR algorithms with the estimators derived from the meta-analysis and the pooled data. RESULTS: The meta-analysis estimator showed a high relative bias (∼40%) when the clinical condition is relatively rare (∼0.5%), whereas ODACoR algorithms exhibited a substantially lower relative bias (∼0.2%). The estimated effects from our ODACoR algorithms were identical on par with the estimates from the pooled data, suggesting the high reliability of our federated learning algorithms. In contrast, the meta-analysis estimate failed to identify risk factors such as age, gender, chronic conditions history, and obesity, compared to the pooled data. DISCUSSION: Our proposed ODACoR algorithms are communication-efficient, highly accurate, and suitable to characterize the complex interplay between multiple clinical conditions. CONCLUSION: Our study demonstrates that our ODACoR algorithms are communication-efficient and can be widely applicable for analyzing multiple clinical conditions in a time-to-event analysis framework.

Time Series AI Model for Acute Kidney Injury Detection Based on a Multicenter Distributed Research Network: Development and Verification Study.

Background: Acute kidney injury (AKI) is a marker of clinical deterioration and renal toxicity. While there are many studies offering prediction models for the early detection of AKI, those predicting AKI occurrence using distributed research network (DRN)-based time series data are rare. Objective: In this study, we aimed to detect the early occurrence of AKI by applying an interpretable long short-term memory (LSTM)-based model to hospital electronic health record (EHR)-based time series data in patients who took nephrotoxic drugs using a DRN. Methods: We conducted a multi-institutional retrospective cohort study of data from 6 hospitals using a DRN. For each institution, a patient-based data set was constructed using 5 drugs for AKI, and an interpretable multivariable LSTM (IMV-LSTM) model was used for training. This study used propensity score matching to mitigate differences in demographics and clinical characteristics. Additionally, the temporal attention values of the AKI prediction model's contribution variables were demonstrated for each institution and drug, with differences in highly important feature distributions between the case and control data confirmed using 1-way ANOVA. Results: This study analyzed 8643 and 31,012 patients with and without AKI, respectively, across 6 hospitals. When analyzing the distribution of AKI onset, vancomycin showed an earlier onset (median 12, IQR 5-25 days), and acyclovir was the slowest compared to the other drugs (median 23, IQR 10-41 days). Our temporal deep learning model for AKI prediction performed well for most drugs. Acyclovir had the highest average area under the receiver operating characteristic curve score per drug (0.94), followed by acetaminophen (0.93), vancomycin (0.92), naproxen (0.90), and celecoxib (0.89). Based on the temporal attention values of the variables in the AKI prediction model, verified lymphocytes and calcvancomycin ium had the highest attention, whereas lymphocytes, albumin, and hemoglobin tended to decrease over time, and urine pH and prothrombin time tended to increase. Conclusions: Early surveillance of AKI outbreaks can be achieved by applying an IMV-LSTM based on time series data through an EHR-based DRN. This approach can help identify risk factors and enable early detection of adverse drug reactions when prescribing drugs that cause renal toxicity before AKI occurs.

Development and Verification of Time-Series Deep Learning for Drug-Induced Liver Injury Detection in Patients Taking Angiotensin II Receptor Blockers: A Multicenter Distributed Research Network Approach.

OBJECTIVES: The objective of this study was to develop and validate a multicenter-based, multi-model, time-series deep learning model for predicting drug-induced liver injury (DILI) in patients taking angiotensin receptor blockers (ARBs). The study leveraged a national-level multicenter approach, utilizing electronic health records (EHRs) from six hospitals in Korea. METHODS: A retrospective cohort analysis was conducted using EHRs from six hospitals in Korea, comprising a total of 10,852 patients whose data were converted to the Common Data Model. The study assessed the incidence rate of DILI among patients taking ARBs and compared it to a control group. Temporal patterns of important variables were analyzed using an interpretable timeseries model. RESULTS: The overall incidence rate of DILI among patients taking ARBs was found to be 1.09%. The incidence rates varied for each specific ARB drug and institution, with valsartan having the highest rate (1.24%) and olmesartan having the lowest rate (0.83%). The DILI prediction models showed varying performance, measured by the average area under the receiver operating characteristic curve, with telmisartan (0.93), losartan (0.92), and irbesartan (0.90) exhibiting higher classification performance. The aggregated attention scores from the models highlighted the importance of variables such as hematocrit, albumin, prothrombin time, and lymphocytes in predicting DILI. CONCLUSIONS: Implementing a multicenter-based timeseries classification model provided evidence that could be valuable to clinicians regarding temporal patterns associated with DILI in ARB users. This information supports informed decisions regarding appropriate drug use and treatment strategies.

Duration of medication treatment for opioid-use disorder and risk of overdose among Medicaid enrollees in 11 states: a retrospective cohort study.

BACKGROUND AND AIMS: Medication for opioid use disorder (MOUD) reduces harms associated with opioid use disorder (OUD), including risk of overdose. Understanding how variation in MOUD duration influences overdose risk is important as health-care payers increasingly remove barriers to treatment continuation (e.g. prior authorization). This study measured the association between MOUD continuation, relative to discontinuation, and opioid-related overdose among Medicaid beneficiaries. DESIGN: Retrospective cohort study using landmark survival analysis. We estimated the association between treatment continuation and overdose risk at 5 points after the index, or first, MOUD claim. Censoring events included death and disenrollment. SETTING AND PARTICIPANTS: Medicaid programs in 11 US states: Delaware, Kentucky, Maryland, Maine, Michigan, North Carolina, Ohio, Pennsylvania, Virginia, West Virginia and Wisconsin. A total of 293 180 Medicaid beneficiaries aged 18-64 years with a diagnosis of OUD and had a first MOUD claim between 2016 and 2017. MEASUREMENTS: MOUD formulations included methadone, buprenorphine and naltrexone. We measured medically treated opioid-related overdose within claims within 12 months of the index MOUD claim. FINDINGS: Results were consistent across states. In pooled results, 5.1% of beneficiaries had an overdose, and 67% discontinued MOUD before an overdose or censoring event within 12 months. Beneficiaries who continued MOUD beyond 60 days had a lower relative overdose hazard ratio (HR) compared with those who discontinued by day 60 [HR = 0.39; 95% confidence interval (CI) = 0.36-0.42; P < 0.0001]. MOUD continuation was associated with lower overdose risk at 120 days (HR = 0.34; 95% CI = 0.31-0.37; P < 0.0001), 180 days (HR = 0.31; 95% CI = 0.29-0.34; P < 0.0001), 240 days (HR = 0.29; 95% CI = 0.26-0.31; P < 0.0001) and 300 days (HR = 0.28; 95% CI = 0.24-0.32; P < 0.0001). The hazard of overdose was 10% lower with each additional 60 days of MOUD (95% CI = 0.88-0.92; P < 0.0001). CONCLUSIONS: Continuation of medication for opioid use disorder (MOUD) in US Medicaid beneficiaries was associated with a substantial reduction in overdose risk up to 12 months after the first claim for MOUD.

Evaluation of the Privacy Risks of Personal Health Identifiers and Quasi-Identifiers in a Distributed Research Network: Development and Validation Study.

BACKGROUND: Privacy should be protected in medical data that include patient information. A distributed research network (DRN) is one of the challenges in privacy protection and in the encouragement of multi-institutional clinical research. A DRN standardizes multi-institutional data into a common structure and terminology called a common data model (CDM), and it only shares analysis results. It is necessary to measure how a DRN protects patient information privacy even without sharing data in practice. OBJECTIVE: This study aimed to quantify the privacy risk of a DRN by comparing different deidentification levels focusing on personal health identifiers (PHIs) and quasi-identifiers (QIs). METHODS: We detected PHIs and QIs in an Observational Medical Outcomes Partnership (OMOP) CDM as threatening privacy, based on 18 Health Insurance Portability and Accountability Act of 1996 (HIPPA) identifiers and previous studies. To compare the privacy risk according to the different privacy policies, we generated limited and safe harbor data sets based on 16 PHIs and 12 QIs as threatening privacy from the Synthetic Public Use File 5 Percent (SynPUF5PCT) data set, which is a public data set of the OMOP CDM. With minimum cell size and equivalence class methods, we measured the privacy risk reduction with a trust differential gap obtained by comparing the two data sets. We also measured the gap in randomly sampled records from the two data sets to adjust the number of PHI or QI records. RESULTS: The gaps averaged 31.448% and 73.798% for PHIs and QIs, respectively, with a minimum cell size of one, which represents a unique record in a data set. Among PHIs, the national provider identifier had the highest gap of 71.236% (71.244% and 0.007% in the limited and safe harbor data sets, respectively). The maximum size of the equivalence class, which has the largest size of an indistinguishable set of records, averaged 771. In 1000 random samples of PHIs, Device_exposure_start_date had the highest gap of 33.730% (87.705% and 53.975% in the data sets). Among QIs, Death had the highest gap of 99.212% (99.997% and 0.784% in the data sets). In 1000, 10,000, and 100,000 random samples of QIs, Device_treatment had the highest gaps of 12.980% (99.980% and 87.000% in the data sets), 60.118% (99.831% and 39.713%), and 93.597% (98.805% and 5.207%), respectively, and in 1 million random samples, Death had the highest gap of 99.063% (99.998% and 0.934% in the data sets). CONCLUSIONS: In this study, we verified and quantified the privacy risk of PHIs and QIs in the DRN. Although this study used limited PHIs and QIs for verification, the privacy limitations found in this study could be used as a quality measurement index for deidentification of multi-institutional collaboration research, thereby increasing DRN safety.

Diabetes medication regimens and patient clinical characteristics in the national patient-centered clinical research network, PCORnet.

We used electronic medical record (EMR) data in the National Patient-Centered Clinical Research Network (PCORnet) to characterize "real-world" prescription patterns of Type 2 diabetes (T2D) medications. We identified a retrospective cohort of 613,203 adult patients with T2D from 33 datamarts (median patient number: 12,711) from 2012 through 2017 using a validated computable phenotype. We characterized outpatient T2D prescriptions for each patient in the 90 days before and after cohort entry, as well as demographics, comorbidities, non-T2D prescriptions, and clinical and laboratory variables in the 730 days prior to cohort entry. Approximately half of the individuals in the cohort were females and 20% Black. Hypertension (60.3%) and hyperlipidemia (50.5%) were highly prevalent. Most patients were prescribed either a single T2D drug class (42.2%) or had no evidence of a T2D prescription in the EMR (42.4%). A smaller percentage was prescribed multiple T2D drug types (15.4%). Among patients prescribed a single T2D drug type, metformin was the most common (42.6%), followed by insulin (18.2%) and sulfonylureas (13.9%). Newer classes represented approximately 13% of single T2D drug type prescriptions (dipeptidyl peptidase-4 inhibitors [6.6%], glucagon-like peptide-1 receptor agonists [2.5%], thiazolidinediones [2.0%], and sodium-glucose cotransporter-2 inhibitors [1.6%]). Among patients prescribed multiple T2D drug types, the most common combination was metformin and sulfonylureas (63.5%). Metformin-based regimens were highly prevalent in PCORnet's T2D population, whereas newer agents were prescribed less frequently. PCORnet is a novel source for the potential conduct of observational studies among patients with T2D.

Comparing Prescribing and Dispensing Data of the PCORnet Common Data Model Within PCORnet Antibiotics and Childhood Growth Study.

Researchers often use prescribing data from electronic health records (EHR) or dispensing data from medication or medical claims to determine medication utilization. However, neither source has complete information on medication use. We compared antibiotic prescribing and dispensing records for 200,395 patients in the National Patient-Centered Clinical Research Network (PCORnet) Antibiotics and Childhood Growth Study. We stratified analyses by delivery system type [closed integrated (cIDS) and non-cIDS]; 90.5 percent and 39.4 percent of prescribing records had matching dispensing records, and 92.7 percent and 64.0 percent of dispensing records had matching prescribing records at cIDS and non-cIDS, respectively. Most of the dispensings without a matching prescription did not have same-day encounters in the EHR, suggesting they were medications given outside the institution providing data, such as those from urgent care or retail clinics. The sensitivity of prescriptions in the EHR, using dispensings as a gold standard, was 99.1 percent and 89.9 percent for cIDS and non-cIDS, respectively. Only 0.7 percent and 6.1 percent of patients at cIDS and non-cIDS, respectively, were classified as false-negative, i.e. entirely unexposed to antibiotics when they in fact had dispensings. These patients were more likely to have a complex chronic condition or asthma. Overall, prescription records worked well to identify exposure to antibiotics. EHR data, such as the data available in PCORnet, is a unique and vital resource for clinical research. Closing data gaps by understanding why prescriptions may not be captured can improve this type of data, making it more robust for observational research.

Privacy, Security, and Patient Engagement: The Changing Health Data Governance Landscape.

CONTEXT: The rapid emergence of new technologies support collection and use of a wide variety of data from clinical, genomic, social and behavioral, environmental, and financial sources, and have a great impact on the governance of personal health information. PAPERS IN THE SPECIAL ISSUE: The papers in this special issue on governance touch on the topic from a variety of focuses, including leadership perspectives, local and federal case studies, and the future importance of patient engagement. THEMES: This special issue focuses on three major themes-that data governance is growing in importance and presenting new challenges that must be addressed, that health care organizations must prioritize governance design, implementation, and functions as a priority, and that governance seems to be naturally converging on an archetype as described by this set of papers. FUTURE STATE OF GOVERNANCE: In order to deal with issues such as data de- and re-identification, data governance must be studied as its own field.

Comparison of consumers' views on electronic data sharing for healthcare and research.

UNLABELLED: New models of healthcare delivery such as accountable care organizations and patient-centered medical homes seek to improve quality, access, and cost. They rely on a robust, secure technology infrastructure provided by health information exchanges (HIEs) and distributed research networks and the willingness of patients to share their data. There are few large, in-depth studies of US consumers' views on privacy, security, and consent in electronic data sharing for healthcare and research together. OBJECTIVE: This paper addresses this gap, reporting on a survey which asks about California consumers' views of data sharing for healthcare and research together. MATERIALS AND METHODS: The survey conducted was a representative, random-digit dial telephone survey of 800 Californians, performed in Spanish and English. RESULTS: There is a great deal of concern that HIEs will worsen privacy (40.3%) and security (42.5%). Consumers are in favor of electronic data sharing but elements of transparency are important: individual control, who has access, and the purpose for use of data. Respondents were more likely to agree to share deidentified information for research than to share identified information for healthcare (76.2% vs 57.3%, p < .001). DISCUSSION: While consumers show willingness to share health information electronically, they value individual control and privacy. Responsiveness to these needs, rather than mere reliance on Health Insurance Portability and Accountability Act (HIPAA), may improve support of data networks. CONCLUSION: Responsiveness to the public's concerns regarding their health information is a pre-requisite for patient-centeredness. This is one of the first in-depth studies of attitudes about electronic data sharing that compares attitudes of the same individual towards healthcare and research.

Data governance requirements for distributed clinical research networks: triangulating perspectives of diverse stakeholders.

There is currently limited information on best practices for the development of governance requirements for distributed research networks (DRNs), an emerging model that promotes clinical data reuse and improves timeliness of comparative effectiveness research. Much of the existing information is based on a single type of stakeholder such as researchers or administrators. This paper reports on a triangulated approach to developing DRN data governance requirements based on a combination of policy analysis with experts, interviews with institutional leaders, and patient focus groups. This approach is illustrated with an example from the Scalable National Network for Effectiveness Research, which resulted in 91 requirements. These requirements were analyzed against the Fair Information Practice Principles (FIPPs) and Health Insurance Portability and Accountability Act (HIPAA) protected versus non-protected health information. The requirements addressed all FIPPs, showing how a DRN's technical infrastructure is able to fulfill HIPAA regulations, protect privacy, and provide a trustworthy platform for research.