Search

The HMO Research Network (HMORN) is a consortium of 16 health care systems with integrated research centers. Approximately 475 people participated in its 17(th) annual conference, hosted by the Department of Population Medicine, Harvard Pilgrim Health Care Institute and Harvard Medical School. The theme, "Collaborations in Population-Based Health Research," reflected the network's emphasis on collaborative studies both among its members and with external investigators. Plenary talks highlighted the initial phase of the HMORN's work to establish the NIH-HMO Collaboratory, opportunities for public health collaborations, the work of early career investigators, and the state of the network. Platform and poster presentations showcased a broad spectrum of innovative public domain research in areas including disease epidemiology and treatment, health economics, and information technology. Special interest group sessions and ancillary meetings provided venues for informal conversation and structured work among ongoing groups, including networks in cancer, cardiovascular diseases, lung diseases, medical product safety, and mental health.

Background and Aims: The worldwide introduction of influenza A (H1N1) vaccines in fall 2009 has raised widespread public questions about the safety of these new vaccines. The Vaccine Safety Datalink Project, a collaboration among CDC and eight health care systems in the HMORN, began accelerated monitoring of this vaccine’s safety in early fall 2009. Methods: The VSD Project conducts near real-time monitoring of potential adverse events following vaccination for all new vaccines. Each week, each site updates dynamic data files, and the coordinating center analyzes all data cumulatively. Repeated analysis of the same data requires adjustment for multiple testing. To address this issue, VSD uses sequential statistical methods. Evaluation of H1N1 vaccine safety poses special challenges, including confounding by indication, seasonality, and new priority groups including pregnant women. We are using several analytic methods, including: Self-controlled case-series, which compares exposed vs. unexposed time windows within the same individual; Poisson, which compares the outcomes during the current season with expected rates based on historical data for seasonal influenza vaccine; and Difference-in-difference, which compares the relative risk in exposed vs. unexposed time windows for H1N1 vaccine during the current season with the relative risk for seasonal influenza vaccine during cumulative prior seasons. Results: Pre-specified outcomes of interest include: Guillain-Barre syndrome, neuropathies, seizures, encephalitis, Bell’s palsy, ataxia, anaphylaxis, spontaneous abortion, pre-eclampsia, stroke, myocarditis and wheezing. Historical rates of these outcomes were calculated for time windows after seasonal influenza vaccination in prior seasons (2000/01–2008/09). The sequential analysis methods above will be used, and relative and absolute risks will be estimated for each outcome. For outcomes found associated with vaccination in preliminary analyses, additional evaluation will be conducted using temporal scan analyses, logistic regression, and when appropriate, medical record review.

Background/Aims The infectious risk of human tissue allografts is unknown. Healthcare claims data from large populations may be useful for quantifying and monitoring rates of infection after tissue allograft transplantation, but require accurate algorithms to identify allograft implantations and infections. We explored the feasibility of identifying allograft implantations and infections using claims data.

> 85%, while the completion rate for anticonvulsants ranged from 30-75%. The lowest test completion rate was for phenobarbital levels to monitor phenobarbital (30%). For all cardiovascular and anticonvulsant drugs, the proportion of recommended tests not ordered by the clinician ranged from 5% to 60%. The lowest test order prevalence was for phenobarbital level for phenobarbital use (35%), followed by valproic acid level for valproate use(48%), and carbamazepine level for carbamazepine use (60%). Rates of patient noncompletion of ordered tests for all drugs was generally

Background: CDC is building a public health information grid to connect public health stakeholders nationwide. The grid will enable efficient and secure sharing of data and applications between public health authorities at the national, state, and local levels as well as researchers and other interested parties (JAMIA 2008;15:705.). The grid may also support federated analysis of distributed data sources to detect clusters that might be invisible to smaller, isolated systems. Grid protocols allow data owners to control data access. Broad sharing of data, however, requires a model to customize the level of patient identifiers seen by users depending upon their role. We describe here a work-in-progress implementation of an ambulatory syndromic surveillance module on the public health grid, including a model for sharing reports with different levels of granularity depending upon user privileges. Methods: We are installing a grid node at Atrius Health, a large multisite, multispecialty ambulatory medical practice serving over 600,000 patients predominantly in eastern Massachusetts. Standard CDC-defined ICD-9-and temperature-based syndrome algorithms developed for the National Bioterrorism Demonstration Project (MMWR 2004;53 Suppl:43.) will be applied to ambulatory encounter data gathered nightly by the Electronic medical record Support for Public Health system (ESP, esphealth.org). ESP consists of a dedicated server populated with nightly extracts of comprehensive encounter data from the electronic medical record system of Atrius Health (JAMIA 2009:16:18). ESP currently analyzes this data to submit fully-identifiable case reports on patients with notifiable conditions to the state health department (MMWR 2008; 57:373.). The syndromic surveillance addition to ESP will modify ESPs reporting module to vary the level of report granularity according to users’ access privileges. ESP reports will include fully de-identified aggregate counts stratified by time period and zip code for basic grid users; semi-identified unit records with age, gender, and geo-code for intermediate users; and fully identifiable unit records with patient names and addresses for highly privileged users such as the state health department. Results: Installation is currently underway. Conclusions: Controlled distribution of syndromic surveillance data through the CDC’s nascent grid will help develop the emerging paradigm for sharing public health information and applications.

Researchers, policy makers and others commonly use electronic data routinely collected during the delivery of, or payment for, medical care to study the “real world” effectiveness, comparative effectiveness, safety, and costs of medical interventions. However, even very large individual healthcare data resources are often not big enough to adequately conduct post-marketing evidence studies. To improve the ability to use multiple distributed data resources, like the HMORN’s Virtual Data Warehouse, the Developing Evidence to Inform Decisions about Effectiveness (DEcIDE) centers at the HMORN Center for Education and Research on Therapeutics (CERT) and the University of Pennsylvania are developing a design for a scalable distributed research network. Funded by AHRQ’s Effective Health Care Program, the project intends to develop the framework for a distributed research network that will help close the knowledge gap regarding the “real world” effectiveness, comparative effectiveness, and safety of medical technologies. The network will have these important attributes: distributed architecture, strong local control of data uses, and federated querying. Constructing the network is a challenge, even among sites with fully operational local virtual data warehouse (VDW) installations, because of differences in computing environments and information systems, the need for responsible stewardship of clinical records, concerns related to data privacy, software development obstacles, the use of proprietary data, and the need to develop effective governance mechanisms. To investigate some of these issues, the project includes implementation of a proof of concept prototype involving these HMORN sites: Kaiser Permanente Northern California, Kaiser Permanente Colorado, Harvard Pilgrim Health Care, Group Health Cooperative, and Geisinger Health System. The goals of the proof of concept work are to distribute a query to separately stored data resources, execute the query remotely, and present aggregated results to an authorized system user. The proof of concept will be designed with a single network portal that manages the network software, including security, authorization, authentication, messaging, and query aggregation. We will present our findings from the proof of concept, including information on specific architecture and design used, the success of the proof of concept, and lessons learned from implementation.

Background/Aims: Our knowledge of safe medication use during pregnancy is limited, due primarily to the lack of rigorous studies evaluating birth outcomes associated with in utero exposure to medications. A collaborative research program, the Medication Use in Pregnancy and Birth Outcomes Program, between the Food and Drug Administration (FDA) and researchers at the HMO Research Network Center for Education and Research in Therapeutics (CERT), Kaiser Permanente Northern and Southern California, and Vanderbilt University, aims to develop common, necessary data linkages to enable the conduct of multiple studies of medication use and outcomes in pregnancy across participating sites. Methods: Specific components of the program include: maternal-infant record linkage of all women delivering an infant between 2001 and 2007, linkage to birth certificate data for infants delivered between 2001 and 2007, validation of data elements within the datasets through medical chart abstractions, and annual update of the datasets. Results: This collaboration will provide a large ethnically and geographically diverse population, with approximately 90,000 births per year. Data files on maternal and child enrollment, outpatient drug dispensing, and inpatient and outpatient diagnoses and procedures, in addition to data elements obtained from birth certificate data, will be developed and maintained at each of the program sites. The proposed poster will describe the program and the available standard file definitions and data elements. Conclusions: The program will be an important resource for the evaluation of birth outcomes (birth defects and other perinatal events) associated with medication use during pregnancy, patterns and trends in medication use during pregnancy, and studies of the risks and benefits of medication use for the pregnant women themselves.

Background: While medication use to manage cardiovascular disease during pregnancy is widespread, data on its safety for the developing infant is scarce. We used population-based data from 5 HMOs to study risks for perinatal complications and congenital defects among infants exposed in-utero to beta -blockers and calcium-channel blockers. Methods: We studied women older than 15 years who delivered an infant between 1/1/96 to 12/31/00, and who had been continuously enrolled with prescription drug coverage for one year prior to delivery. We further limited our study to mother-infant pairs with 30 and 365 days follow-up for the evaluation of perinatal outcomes and congenital anomalies, respectively. Information on prescription drug dispensing, and inpatient and outpatient diagnoses and procedures was obtained from automated databases at each health plan. Results: There were a total of 87,407 mother-infant pairs with at least 30 days follow- up in the five health plans. Of these, a total of 405 full- term infants were exposed during pregnancy to beta-blockers and 721 full- term infants were exposed to calcium-channel blockers. Infants exposed to beta-blockers in the third trimester of pregnancy were at over three-fold increased risk for hypoglycemia (RR 3.1; 95% CI 2.2, 4.2), for feeding problems (RR 1.8; 95% CI 1.3, 2.5) and for prolonged hospitalization (RR 2.0; 95% CI 1.3, 3.1). Infants exposed to calcium-channel blockers in the third trimester had an increased risk for neonatal seizures (RR 3.6; 95% CI 1.3, 10.4), and for hematological disorders (RR 2.6; 95% 1.4, 5.1). A chart review confirmed the seizures and hypoglycemia cases but found no serious hematologic conditions common to the calcium-channel exposed infants. There were no increased risks for congenital anomalies among either group of infants. Conclusions: Infants whose mothers receive beta-blockers are at increased risk for neonatal hypoglycemia. Beta-blockers can cross the placenta, increasing insulin and decreasing glucagon in the newborn, leading to hypoglycemia. Infants whose mothers take calcium-channel blockers are at increased risk for neonatal seizures, although the mechanism leading to this increased risk is less clear. These data should be disseminated to the obstetric and neonatal community so that appropriate newborn management strategies may be developed.

Background/Aims: Improving access to multiple sources of electronic health information is crucial for population-based health care research. But it also raises serious issues relating to data confidentiality and security. We propose to develop methods and tools (e.g., SAS macros) to enable multivariate analyses to be conducted using multiple datasets that are stored separately without exposing any patient level information. Phase I of this project focuses on linear regression. Methods: A two-step process is carried out by SAS macros. The macros will be installed on both the central and site computers that hold the analytic datasets. After the central computer initiates the process, each site computer executes the SAS program locally and submits intermediate summarized statistical results to the central computer. The central computer then combines the intermediate results, and computes the parameter estimates. A second round of program execution delivers the variance/covariance estimates of the parameter estimates. Results and Conclusions: Simulated separated data sets were used to test the macros. The results were identical to the outputs from the SAS PROC REG procedure when the data sets were combined. The macro enables remote execution of multivariate linear regression analysis within the context of a multi- institutional distributed research network. The successful implementation of the secure statistical analysis within a distributed database model will overcome one of the largest hurdles in fully developing such a model, namely, the current need to combine data for multivariate analyses. Obviating this need will save the time and resources currently spent developing data-sharing agreements and will greatly minimize the patient confidentiality and proprietary concerns