Sean Hennessy

Proton-Pump Inhibitor Use and the Risk for Community-Acquired Pneumonia

Context Some studies suggest that proton-pump inhibitors (PPIs) may increase risk for community-acquired pneumonia (CAP). Contribution This large, nested casecontrol study involving adults in general practices in the United Kingdom found that current, long-term PPI use was not associated with increased risk for CAP. Recently started PPI therapy was associated with increased risk: Adjusted odds ratios for PPI use started within 2, 7, and 14 days of CAP diagnosis were 6.5, 3.8, and 3.2, respectively. Implication Long-term PPI use is not associated with increased risk for CAP. The association between recently started PPIs and CAP risk is not clear but is not necessarily causal. The Editors Proton-pump inhibitors (PPIs) are potent gastric acid suppressants. With its excellent efficacy for acid-related diseases and increasing availability of both over-the-counter and generic formulations, PPI use continues to escalate (1). Antiulcer drugs, primarily PPIs, ranked second in overall U.S. retail sales at

Cardiac arrest and ventricular arrhythmia in patients taking antipsychotic drugs: cohort study using administrative data

10.8 billion in 2001 (2). Two studies recently suggested that PPI use may increase the risk for community-acquired pneumonia (CAP) (3, 4). A proposed mechanism is increased bacterial colonization in the upper gastrointestinal tract due to gastric acid suppression (3, 4). However, other noncausal mechanisms may also be responsible for the association between PPIs and CAP. Residual confounding may have contributed to the relatively modest increased risk seen in these studies. Furthermore, both studies found that the association was weakest among current recipients who had been taking PPIs for the longest duration, which is contrary to what one would expect if PPIs increase the risk for CAP. Community-acquired pneumonia leads to significant morbidity and mortality and dramatic costs to the health care system. In the United States, more than 1.1 million persons are hospitalized each year because of CAP (5). In the Medicare program alone, more than

The U.S. Food and Drug Administration's Mini‐Sentinel program: status and direction

3.5 billion are spent on pneumonia-related admissions annually (5). Given the widespread use of PPIs, clarifying the potential association between PPI therapy and risk for CAP is of great importance to public health. A recent systematic review on management strategies for gastroesophageal reflux disease, commissioned by the U.S. Agency for Healthcare Research and Quality, called for further studies to elucidate the role of acid suppression on the development of CAP (6). Thus, we sought to more definitively examine the effect of current PPI exposure on CAP development in a large, population-based cohort. Methods Data Source We conducted a nested casecontrol study by using the General Practice Research Database (GPRD). The GPRD comprises prospectively collected, computerized medical records from a sample of general practices throughout the United Kingdom that represents the U.K. population in terms of age, sex, and geographic distribution (7). Under the National Health Service, a general practitioner coordinates the health care of 98% of U.K. residents. The GPRD differs fundamentally from claims databases because it comprises the actual medical record, which contains complete and comprehensive clinical data. The information collected in the database includes demographic characteristics, prescription use, clinical diagnoses, subspecialty consultation notes, and hospital discharge diagnoses. Prescriptions for most medications are written for 1 month in U.K. general practices. Details of every prescription issued include date, dosage, quantity dispensed, duration of therapy, and indication. The Read clinical classification and the Oxford Medical Information System codes are used to classify medical diagnoses (8, 9). The Read clinical classification was adopted in the United Kingdom in 1990 as an electronic medical coding system. These codes incorporate information on medical history, physical examination, procedures, symptoms, medication use, and social history. These codes are cross-referenced with other major systems of medical coding, including but not limited to the International Classification of Diseases, Ninth and Tenth Revisions (ICD-9 and ICD-10), and Current Procedural Terminology-4. The Oxford Medical Information System codes are related to the ICD-9 codes, with the first 3 numbers typically corresponding to the first 3 digits of the ICD-9 codes. The GPRD uses a practice-based quality marker known as up-to-standard to indicate when data recording by the practice adhered to specific quality measures according to GPRD Recording Guidelines, with respect to completeness, continuity, and plausibility (7). We used only data that were collected after the up-to-standard date in each practice to maximize the validity of the study. Previous validation studies have shown that the GPRD captures 90% to 95% of the diagnoses from specialty referral visits and greater than 90% of the principal hospital discharge diagnoses (1012). The institutional review board at the University of Pennsylvania and the GPRD Independent Scientific Advisory Committee approved this study. The University of Pennsylvanias institutional review board granted a waiver of informed consent because we used existing data that contained no patient identifiers. Study Cohort Study participants were drawn from a cohort of approximately 9 million patients who started follow-up in the GPRD from May 1987 to April 2002. The Figure shows the cohort selection process and the exclusion criteria. A substantial proportion of patients met several exclusion criteria. We excluded patients who received a diagnosis of CAP within the initial 6 months of GPRD follow-up to avoid misclassification of prevalent CAP cases as incident cases (13). We excluded patients receiving Helicobacter pylori eradication therapy because their antibiotics concurrent with PPI therapy could potentially mask the effect of PPI use on the risk for CAP. We excluded patients who received a diagnosis of aspiration pneumonia because it differs from CAP in terms of pathophysiology and risk factors. Only 1% of case patients were excluded solely on the basis of this criterion. Figure. Study flow diagram. GPRD = General Practice Research Database. * Some patients met multiple exclusion criteria. Case Patients Case patients consisted of all individuals in the eligible study cohort who received a diagnosis with their first episode of CAP within the GPRD follow-up period. The CAP diagnosis was determined on the basis of a list of relevant Read clinical classification or Oxford Medical Information System codes. The date of the first CAP diagnosis was designated as the index date for each case. To improve the specificity of our CAP case patient definition, we did a secondary analysis in which we restricted our case-patient group to those who had a first episode of CAP in the GPRD that resulted in a hospital admission. A case of CAP was considered to have required hospitalization if the recorded diagnosis was linked with a hospitalization outcome code. Control Participants By using incidence density sampling, we randomly selected up to 10 control participants for each case patient from the eligible study cohort. Incidence density sampling of control participants yields odds ratios (ORs) that are unbiased estimates of the incidence rate ratios (14). Control participants were selected 1 case at a time with replacement (that is, a patient can be a control participant for several cases). For each CAP case, we first identified all patients in the study cohort who remained at risk for CAP on the index date of the case. We then selected, among patients at risk, those who were followed at the same general practice site as the case. From those at-risk patients in the same practice site, we then selected patients who started GPRD follow-up on a date that is within 1 month before or after the GPRD start date of the case. Patients who fulfilled all criteria were eligible matched control participants and were assigned an index date that was the same as the case index date. By definition, all eligible control participants were matched to their respective case patients on index date, general practice site, and both calendar period and duration (1 month) of follow-up before the index date. In the event that 10 or fewer eligible matched control participants were available for a particular case, all control participants were included. If more than 10 eligible matched control participants were available for a case, we randomly selected 10 for our analysis. We then repeated the process for all case patients. We were able to identify 10 control participants for greater than 99% of the cases and at least 1 eligible matched control participant for all cases. The reason for choosing 10 control participants was based on considerations for statistical power and efficiency of analysis. Generally, significant statistical power can be gained by going from a 1-to-1 to a 1-to-10 casecontrol ratio. There would be little further gain in statistical power, yet the computational time would increase significantly by increasing the controlcase ratio above 10. Measure of Exposure The primary exposure of interest was current PPI therapy: use of any PPI (esomeprazole, omeprazole, rabeprazole, pantoprazole, or lansoprazole) within 30 days before the index date. Patients were considered to be exposed if they had a PPI prescription that would have lasted beyond 30 days before the index date. Among current PPI recipients, we examined the effect of daily dose (1.5 defined daily dose/d vs. >1.5 defined daily dose/d) (15) and duration of use before index date (that is, <30 days, 30 to 180 days, or >180 days) as a secondary analysis. We also did a sensitivity analysis by varying the current exposure window to 15, 60, 90, and 120 days. Past recipients were those who were exposed to PPI therapy that ended more than 30 days before the index date in the primary analysis. Statistical Analysis We used conditional logistic re

Effectiveness of live-attenuated Japanese encephalitis vaccine (SA14-14-2): a case-control study

Abstract Objective: To examine the rates of cardiac arrest and ventricular arrhythmia in patients with treated schizophrenia and in non-schizophrenic controls. Design: Cohort study of outpatients using administrative data. Setting: 3 US Medicaid programmes. Participants: Patients with schizophrenia treated with clozapine, haloperidol, risperidone, or thioridazine; a control group of patients with glaucoma; and a control group of patients with psoriasis. Main outcome measure: Diagnosis of cardiac arrest or ventricular arrhythmia. Results: Patients with treated schizophrenia had higher rates of cardiac arrest and ventricular arrhythmia than controls, with rate ratios ranging from 1.7 to 3.2. Overall, thioridazine was not associated with an increased risk compared with haloperidol (rate ratio 0.9, 95% confidence interval 0.7 to 1.2). However, thioridazine showed an increased risk of events at doses 600 mg (2.6, 1.0 to 6.6; P=0.049) and a linear dose-response relation (P=0.038). Conclusions: The increased risk of cardiac arrest and ventricular arrhythmia in patients with treated schizophrenia could be due to the disease or its treatment. Overall, the risk with thioridazine was no worse than that with haloperidol. Thioridazine may, however, have a higher risk at high doses, although this finding could be due to chance. To reduce cardiac risk, thioridazine should be prescribed at the lowest dose needed to obtain an optimal therapeutic effect. What is already known on this topic Thioridazine seems to prolong the electrocardiographic QT interval more than haloperidol Although QT prolongation is used as a marker of arrhythmogenicity, it is unknown whether thioridazine is any worse than haloperidol with regard to cardiac safety What this study adds Patients taking antipsychotic drugs had higher risks of cardiac events than control patients with glaucoma or psoriasis Overall, the risk of cardiac arrest and ventricular arrhythmia was not higher with thioridazine than haloperidol Thioridazine may carry a greater risk than haloperidol at high doses Patients should be treated with the lowest dose of thioridazine needed to treat their symptoms

Risk of major cardiovascular events in patients with psoriatic arthritis, psoriasis and rheumatoid arthritis: a population-based cohort study

The Mini‐Sentinel is a pilot program that is developing methods, tools, resources, policies, and procedures to facilitate the use of routinely collected electronic healthcare data to perform active surveillance of the safety of marketed medical products, including drugs, biologics, and medical devices. The U.S. Food and Drug Administration (FDA) initiated the program in 2009 as part of its Sentinel Initiative, in response to a Congressional mandate in the FDA Amendments Act of 2007.

Parenteral ketorolac: the risk for acute renal failure.

BACKGROUND Japanese encephalitis is a major cause of death and disability throughout Asia, including the Indian subcontinent. Although an effective vaccine for Japanese encephalitis is available, hundreds of millions of susceptible individuals remain unimmunised because of the vaccines cost. In 1988, an inexpensive live-attenuated vaccine (SA14-14-2) was licensed in China. We have measured the effectiveness of this vaccine. METHODS In a case-control study in rural Sichuan Province, China, the 56 cases consisted of children admitted to hospital with acute Japanese encephalitis, and were confirmed serologically. 1299 village-matched and age-matched controls were identified, and vaccination histories obtained from pre-existing written records. FINDINGS The effectiveness of one dose was 80% (95% Cl 44 to 93%); that of two doses was 97.5% (86 to 99.6%). Controlling for multiple potential confounders did not alter these results. INTERPRETATION We conclude that a regimen of two doses of live-attenuated Japanese encephalitis vaccine, administered 1 year apart, is effective in the prevention of clinically important disease. Subsequent study is needed to assure the safety of this vaccine.

Design of a national distributed health data network.

Objectives We aimed to quantify the risk of major adverse cardiovascular events (MACE) among patients with psoriatic arthritis (PsA), rheumatoid arthritis (RA) and psoriasis without known PsA compared with the general population after adjusting for traditional cardiovascular risk factors. Methods A population-based longitudinal cohort study from 1994 to 2010 was performed in The Health Improvement Network (THIN), a primary care medical record database in the UK. Patients aged 18–89 years of age with PsA, RA or psoriasis were included. Up to 10 unexposed controls matched on practice and index date were selected for each patient with PsA. Outcomes included cardiovascular death, myocardial infarction, cerebrovascular accidents and the composite outcome (MACE). Cox proportional hazards models were used to calculate the HRs for each outcome adjusted for traditional risk factors. A priori, we hypothesised an interaction between disease status and disease-modifying antirheumatic drug (DMARD) use. Results Patients with PsA (N=8706), RA (N=41 752), psoriasis (N=138 424) and unexposed controls (N=81 573) were identified. After adjustment for traditional risk factors, the risk of MACE was higher in patients with PsA not prescribed a DMARD (HR 1.24, 95% CI 1.03 to 1.49), patients with RA (No DMARD: HR 1.39, 95% CI 1.28 to 1.50, DMARD: HR 1.58, 95% CI 1.46 to 1.70), patients with psoriasis not prescribed a DMARD (HR 1.08, 95% CI 1.02 to 1.15) and patients with severe psoriasis (DMARD users: HR 1.42, 95% CI 1.17 to 1.73). Conclusions Cardiovascular risk should be addressed with all patients affected by psoriasis, PsA or RA.

Cardiovascular Events and Death in Children Exposed and Unexposed to ADHD Agents

Ketorolac tromethamine is the first nonsteroidal anti-inflammatory drug (NSAID) to be approved in the United States for parenteral use as an analgesic. Clinical trials done before the drug was marketed showed that its efficacy was similar to that of moderate doses of parenteral opioids in patients having surgery [1]. Although ketorolac therapy has been discontinued less often than have meperidine hydrochloride and morphine therapy, ketorolac has been associated with the same adverse events that are seen with other NSAIDs; these adverse events include gastrointestinal events, rare allergic reactions, and liver dysfunction. Other NSAIDs have been associated with the renal syndromes of acute renal failure, interstitial nephritis, the nephrotic syndrome, hyponatremia, and hyperkalemia [2-14], but these were not reported in the premarketing clinical trials of ketorolac. Since ketorolac has been marketed, it has been widely used in clinical settings other than clinical trials. As have other NSAIDs [14-18], ketorolac has been associated with acute renal failure [19-27]. The appropriate role of ketorolac and all NSAIDs has consequently been questioned, especially for patients who are considered to be at high risk for acute renal failure [13, 14]. We did a large cohort study to evaluate the effects (including nephrotoxicity) of parenteral ketorolac in the postmarketing clinical setting. We previously reported on the risks for gastrointestinal bleeding and surgical-site bleeding associated with ketorolac [28]. We now compare the potential risk for acute renal failure after administration of ketorolac with the risk after administration of opioids among hospitalized patients. Methods Study Sample This retrospective cohort study was done using 35 community-based and tertiary care hospitals in the Philadelphia area. Data collection began on 18 November 1991 and ended on 31 August 1993. All patients who were identified from hospital pharmacy records as having received parenteral ketorolac during the data collection period were potentially eligible for inclusion in the group receiving ketorolac, regardless of whether they had concomitantly received opioids. The comparison group consisted of patients who received parenteral opioids (without ketorolac) and was matched to the ketorolac group by hospital, admitting service (any medical service compared with any surgical service), and date on which therapy was initiated. Use of ketorolac or opioids was verified by examining medication administration records. Patients who were receiving long-term dialysis were excluded. A course of ketorolac or opioids was defined as the time from administration of the first dose through the third day after administration of the final dose. If more than 3 days had elapsed between consecutive doses, a new course was defined as starting after the lapse. We collected data on all courses of ketorolac. Data on repeated courses of opioids were not abstracted because the purpose of the unexposed comparison group was to serve as a control group that had indications similar to those of the group receiving ketorolac and not to identify all adverse events that occurred in patients receiving opioids. If more than one course of opioids was available, we chose the course that had the initiation date closest to that of the matched patients course of ketorolac. Data were abstracted from the hospital charts of 9850 patients who had received 10 219 courses of ketorolac and of 10 145 patients who had received 10 145 courses of opioids. Only 326 of the 9850 patients receiving ketorolac (3.3%) received more than one course. Of these patients, 291 received two courses, 27 received three courses, and 8 received four courses. All analyses are presented by treatment course because each course represented a separate opportunity for an adverse outcome. However, separate analyses of each patients first course alone yielded nearly identical results. Data Collection Trained nurses used a computer-based data entry system to abstract data from hospital charts. The data collected included demographic characteristics, medical history, dose and duration of ketorolac or opioid therapy, occurrence of surgery, use of concomitant medication, laboratory data, and adverse events (regardless of whether the hospital staff or the abstracter thought that these events were caused by the drug). Definitions of Acute Renal Failure The principal definition of acute renal failure was a peak serum creatinine concentration that was 50% greater than the baseline value and 1) an absolute increase of at least 44.2 mol/L if the baseline concentration was less than 132.6 mol/L or 2) an absolute increase of at least 88.4 mol/L if the baseline concentration was 132.6 mol/L or greater. Patients for whom baseline serum creatinine values were not available did not meet the definition of acute renal failure even if their peak serum creatinine concentration was abnormally elevated. Our secondary definition required, in addition to laboratory evidence, a notation in the hospital chart that acute renal failure had occurred during the course of therapy with the analgesic drug. Unless otherwise stated, the results presented are those obtained using our principal definition. Statistical Analysis Data on demographic characteristics and medical history were compared between the two groups using the independent sample t-test for continuous variables and the chi-square statistic [29] for discrete variables. The proportion of patients in each group for whom data on serum creatinine concentration were included in the medical record was described. Both matched and unmatched analyses were done. Because point estimates and 95% CIs did not greatly differ between the two types of analysis, we report the results of the unmatched analyses [30]. We did a survival analysis using Cox proportional-hazards regression to explore the association of ketorolac administration with the rate of acute renal failure [30, 31]. Survival was defined as the interval from the initiation of analgesic drug therapy until either acute renal failure or the end of the course (3 days after the end of drug therapy), whichever occurred first. Unadjusted rate ratios comparing the rate of acute renal failure in patients receiving ketorolac with the rate in patients receiving only opioids were calculated using standard proportional hazards methods and are reported with 95% CIs [30, 31]. Multivariate proportional hazards models were fit, with simultaneous adjustment for the influence of potential confounding variables that were defined a priori. These variables included age; type of pain (acute or chronic) that served as the indication for analgesic administration; medical admission; admission to an intensive care or trauma unit; concomitant use of NSAIDs other than ketorolac, aminoglycoside antibiotics and other antibiotics, or angiotensin-converting enzyme inhibitors and other antihypertensive drugs; and a history of cancer, congestive heart failure, kidney disease, diabetes mellitus, hypertension, NSAID use, drug abuse, or cirrhosis. A time-dependent covariate that indicated the duration of analgesic therapy was incorporated into all models. We did a sensitivity analysis to assess the potential effect of the fact that a smaller proportion of patients in the ketorolac group had serum creatinine values measured during their treatment course. In this analysis, we recalculated the unadjusted relative risk for acute renal failure under the assumption that the risk for acute renal failure among study patients without measures of serum creatinine was the same as the risk among patients in the same group who did have measures recorded. This assumption is extreme because it assigns the risk for acute renal failure that was seen among patients who had laboratory data to patients who did not have laboratory data and thus probably had low morbidity. Because we were interested in the possible nephrotoxicity of ketorolac in patients who had a high risk for acute renal failure, we evaluated the interactions between the administration of ketorolac and coexisting conditions that may have predisposed patients to acute renal failure. These coexisting conditions include congestive heart failure; cirrhosis; a history of renal disease, diabetes mellitus, or hypertension; age older than 65 years; and heart failure, cirrhosis, or a history of renal disease [13]. We also explored potential interactions between ketorolac and the concomitant use of either aminoglycoside antibiotics or angiotensin-converting enzyme inhibitors. Finally, we explored the potential interaction of ketorolac with the presence of any condition known to predispose patients to acute renal failure or the concomitant administration of aminoglycoside antibiotics or angiotensin-converting enzyme inhibitors. Interactions were assessed on the basis of the statistical significance of the relevant product term in the multivariate models. Duration of analgesic therapy was defined as the number of days during which the analgesic drug was administered (in patients who did not have renal failure and in those whose event occurred after the last day of therapy) or the number of days from the onset of therapy until renal failure. The interaction between duration of therapy and choice of analgesic agent was analyzed in two ways. First, a set of proportional hazards models was fit; each model included patients who had received analgesic therapy for no longer than a specified duration (that is, those receiving therapy for as many as 2 days, as many as 3 days, and so forth). In any given model, patients who received an analgesic drug for longer than the specified duration for that model were included, but their follow-up was censored at that specified duration. For example, a patient who received analgesic therapy for 5 days was included in the model of as many as 4 days of analgesic therapy but was censored in that analysis a

Comparative Risk for Angioedema Associated With the Use of Drugs That Target the Renin-Angiotensin-Aldosterone System

Key Summary Points: Attributes of a National Distributed Health Data Network Supports both observational and intervention studies. Local data holder control over access and uses of data. Mitigates need to share or exchange protected health information. Singular, multipurpose, multi-institutional infrastructure. A distributed health data network is a system that allows secure remote analysis of separate data sets, each derived from a different medical organizations or health plans records. Such networks allow data holders to retain physical control over use of their data, thereby avoiding many obstacles related to confidentiality, regulation, and proprietary interests. They can be used for observational studies, particularly public health surveillance, and can also provide baseline and follow-up data to support clinical trials, including those that use cluster randomization. In addition, a network can monitor use, adoption, and diffusion of new technologies and clinical evidence. Such networks are critical elements of the learning health care system recommended by the Institute of Medicine (1), which supports the use of routinely collected health care data to improve our understanding of the comparative benefits and harms of medical technologies. The United States will soon be able to analyze data from millions of individuals. Congress has mandated that the U.S. Food and Drug Administration develop a postmarket risk identification and analysis system that covers 100 million persons (2). In addition, the expansion of comparative effectiveness research envisioned by Congress requires access to health care information for large, diverse populations in real-world settings (3). Large, centralized data repositories could support these functions, but we and others (4, 5) believe that a distributed health data network has many practical advantages. First, a distributed network allows data holders to retain physical and logical control of their data. Second, it mitigates many security, proprietary, legal, and privacy concerns, including those regulated by the Privacy and Security Rules of the Health Insurance Portability and Accountability Act (6). Third, it eliminates the need to create, maintain, and secure access to central data repositories. Fourth, it minimizes the need to disclose protected health information outside the data-owning entity. Finally, a distributed network allows data holders to assess, track, and authorize requests for all data uses. Several public agencies have supported the development of single-purpose distributed data networks, either directly or in principle (711). These networks are limited in scope and do not support the broad range of public and private needs filled by the network we describe. We favor a single distributed network with multiple usesfor example, one that could be used to study comparative clinical effectiveness and the diffusion of medical technologiesover multiple independent and single-purpose networks. A multipurpose network would reduce the burden on data holders of participating in multiple networks, as well as that on network developers of creating and maintaining redundant infrastructure. The framework that we describe suggests how we could develop a national network with broad capabilities. How Would a National Distributed Health Data Network Work? In the simplest national distributed health data network, each data holder creates a copy of their data (a network datamart) that adheres to a common data model, thus ensuring identical file structures, data fields, and coding systems. Several common data models already exist (10, 1217). The Figure illustrates the basic flow of network operations. Authorized users submit queries by means of a secure Web site. Data holders set authorization policies for each user and query type and can require approvals from privacy boards and institutional review boards. The network interface allows nontechnical users to ask simple questions without assistance (for example, a report on the uptake of a given treatment by age, sex, and geographic region). It also allows sophisticated users to perform complex analyses (for example, comparing the rates of serious cardiovascular outcomes among patients who receive different second-line antihypertensive treatments). For many questions, transferring protected health information will not be necessary. However, it may be necessary to aggregate relatively small amounts of data for analysis. Using the network, data holders may provide limited access to full-text medical records for validation and additional details. It is usually necessary to review only a small proportion of records to confirm diagnoses or to obtain risk factor data that are not coded (such as smoking status). Figure. System operations in a distributed health network. An authorized user accesses the secure network Web site to submit queries (computer programs) to run against data in the network datamarts. The boxes at the far right depict areas under control of the data holder (data holders A through D are shown). Authorization to execute a query is under control of the data holder and can be limited to specific users and uses. Data holders retrieve queries for execution, which eliminates the need for data holders to monitor incoming requests. Query results are encrypted and returned to the central Web site, where they are processed and presented to the requester. Details of each step are recorded for auditing. Example of the Use of a Distributed Network Some research programs already use a distributed network model (10, 14, 18), which provides a relevant starting point to implement a national network. The HMO Research Network Center for Education and Research on Therapeutics has conducted many multisite studies by distributing computer programs that each site applied to a local copy of their data. The outputs are then combined to provide aggregate results. Examples of studies performed in this way include the evaluation of laboratory monitoring practices for medications (1825), the use of medications during pregnancy (2628), and the use of medications that carry a black box warning (29). Such studies provide an important evidence development function that feeds back to providers, payers, and patients. Policy Issues Development and implementation of a multipurpose, multi-institutional distributed health data network requires substantial stakeholder engagement and dedicated software development. On the basis of the previously described research studies, we recommend incremental implementation with a limited set of data holders and data types. Begin with information about eligibility for health care (such as health plan enrollment data); this would allow identification of defined populations, which are important for many uses. Initial data should also include demographic characteristics; diagnosis, procedure, and pharmacy dispensing data (30); and, potentially, electronic health record data, such as vital signs. During initial implementation, pilot testing is needed to assess network design, software development, and development and implementation of the common data model. A distributed networks viability depends on both its governance mechanisms and sustained funding. A governance institution is needed to develop and oversee procedures for requesting use of the network; to set priorities; and to audit use for compliance with various security, privacy, human subject research, and proprietary concerns. Such an institution should also monitor research integrity, data integrity, conflict of interest policies, transparency of activity and results, policies related to access and use, reproducibility, publishing rights, and dispute resolution. Annual development and maintenance costs would probably be several tens of millions of dollars for an initial system that covers up to 100 million persons. This would be similar to the 3-year startup cost for the National Cancer Institutes Cancer Biomedical Informatics Grid, which totaled

Textbook of pharmacoepidemiology

60 million for fiscal years 2004 to 2006 (31). The National Cancer Institute fiscal year 2010 budget requests