Search

Background: Health plans’ administrative claims data are known to be very useful in post-marketing drug or vaccine safety surveillance to detect certain adverse events that are difficult to capture during the preapproval clinical trails. To detect the existence of excess risk for an adverse drug event, hypothesis testing should be conducted whenever data is updated and thus appropriate group sequential or sequential analysis methods should be implemented to adjust for multiple testing and preserve the overall type I error rate. Methods: We propose a practical group sequential method, a conditional sequential sampling procedure, to derive valid inference on the parameter of interest, the relative risk for an adverse drug event between the drug of interest and the comparison drug. The method allows the information for both drug groups to be accumulated prospectively and thus, unlike the newly developed maximized sequential probability ratio test (MaxSPRT), doesn’t require the availability of a lot of historical data for the comparison drug. Moreover, the method automatically adjusts for population heterogeneity and temporal trend and requires no a priori assumptions on how the baseline incidence rates change across strata and over time. In addition, the method remains valid (i.e., preserves the nominal Type I error rate) even when the number of interim tests is large and/or there are a lot of strata defined based on several potential confounders. We will explain why the standard general group sequential theory method might not be appropriate in such settings. Results: We have conducted an extensive simulation study to evaluate the performance of the new method and compare that to the performance of the general group sequential theory method in a subset of scenarios when applicable. The power performance for both methods in the considered scenarios is very similar and our new method applies to much more general settings. Conclusions: We will also implement this method to the data collected in the HMO Research Network CERT II study. This method can be particularly useful in prospective drug surveillance studies in which both drugs are relative new and not enough historical data is available.

Background/Aims: Research and surveillance work addressing ectopic pregnancy (EP) rely largely on ICD diagnostic codes and CPT procedure codes available from automated data sources. However, cases identified in this way may not be true EP cases, as these codes may also be used to rule out, follow up or note a history of EP. Through the use of additional automated data on treatment, procedures, and other aspects of care, we developed a classification algorithm that could improve the accuracy of EP case identification. Methods: Using Group Health automated data files, we initially identified 2,632 potential EP episodes occurring in women aged 15–44 years during 1988–2007 using ICD and CPT codes for EP and for surgical or laparoscopic treatment of EP. Chart reviews to verify EP status were conducted on 130 potential EP cases in the algorithm development dataset and on 150 in the algorithm validation dataset. Using additional information on demographic factors, other diagnosis and procedure codes, treatment modalities, site of care (ambulatory vs. inpatient), and laboratory data available with each EP episode, we conducted a classification and regression tree (CART) analysis to create a case finding algorithm for EP. Results: From the CART analysis, the case-finding algorithm for EP contained three main predictors: at least two encounter dates with an EP diagnosis or procedure code during an episode; treatment with methotrexate; and presence of an ICD -9 code of 633.1, 633.10 or 633.11 for tubal pregnancy. Sensitivity for the development and validation sets, respectively, was 95% and 91% while specificity was 78% and 83%. The EP misclassification rate using the new algorithm was 10.7% in the development set and 11.5% in the validation set compared to 32% when EP cases were originally defined using EP diagnosis and procedure codes. Conclusions: The CART-derived algorithm for identifying EP cases was highly sensitive, had good specificity, and misclassification rates were notably improved over the original case identification techniques. Additional pharmacy and encounter data available in many health plan databases can markedly improve the accuracy of EP identification. In particular, the identified predictors in the algorithm are available in the CRN VDW, and it would be of interest to test this algorithm at other HMORN sites.

Objectives: Osteonecrosis of the jaw (ONJ) has been recently associated with exposure to bisphosphonates (BPs). Incidence and risk factors are ill-defined. This Dental PBRN study (www.dpbrn.org; support: DE-16746, DE-16747) used electronic records from two participating health plans to estimate ONJ incidence and assess whether BP use increased the likelihood of ONJ. Methods: We used multiple search strategies and manual chart reviews to identify ONJ cases in members of Kaiser Permanente Northwest and HealthPartners of Minnesota. Our cohort was members age 35 and older with medical and pharmacy coverage for at least one year during 1994–2006. We used a keyword search of electronic physician notes to flag suspected cases among patients with a diagnosis or procedure indicative of ONJ and then confirmed cases through manual chart review. Oral surgeons and general dentists identified additional cases among cohort members. We used logistic regression to estimate the odds of ONJ from BP exposure, adjusting for other risks. Results: 37,290 of 572,602 cohort members had a diagnosis or procedure indicating the potential for ONJ. A key word search of physician notes and chart review found 14 ONJ cases; 5 more cases were found through dental providers. Six cases (31.6%) had an oral BP dispense prior to ONJ onset, compared to 21,163 cohort members (3.7%). Eight cases (42.1%) had a history of cancer, and 5 (26.3%) had osteoporosis, compared to 71,776 (12.5%) and 23,392 (4.1%), respectively, among all cohort members. Univariate and multivariate logistic regression were inconclusive due to the small number of confirmed ONJ cases. Conclusions: The incidence of ONJ in this cohort was very low and limited our ability to assess risks. When compared to the larger cohort, the higher frequencies of oral BPs, cancer, and osteoporosis among ONJ cases suggest these may be risk factors for ONJ. The low number of cases resulted in logistic regression outcome predictors that were no better than random. The rarity of ONJ cases in our study may be good news for the millions of adults currently taking oral bisphosphonates, although more data are needed before we know for sure.

Background and Aims: Thiazide diuretics are recommended alone or in combination for uncomplicated hypertension (HTN). Most patients require treatment with 2 or more drugs. Based on studies of mono-therapy, ACE-I are recommended for patients at risk of renal disease, including diabetics. Data are sparse regarding thiazide plus ACE-I combination therapy. We hypothesized that thiazide plus ACE-I is associated with a lower incidence of renal disease compared with other common thiazide combinations, but that confounding by indication for diabetes might attenuate this effect. Methods: We conducted a retrospective cohort study of thiazide combinations in a 41-site clinical practice that is the dominant provider in a large rural area. Data were extracted from an electronic medical record for all patients >60 years treated for HTN between 2001 and 2006. Patients with prevalent renal disease, or 2 measurements persisting >3 months. Results: Among 4700 patients (98% Caucasian, 69% female, mean age 70 yrs, mean follow-up 32.5 months), the incidence of renal disease was 22.7%. Five drug categories accounted for 97% of thiazide combinations: ACE-I, Angiotensin Receptor Blocker (ARB), Beta-blocker (BB), Calcium Channel Blocker (CCB) and Potassium-sparing diuretic (P-S). In Cox models with ACEI + thiazide as the reference group, adjusted for age, sex, and pretreatment blood pressure, patients who used BB + thiazide (HR 0.72, 95% CI, 0.600.86) and CCB + thiazide (HR 0.72, 0.55–0.96) had significantly lower hazard ratios for incident renal disease than those who used ACE-I + thiazide. In analyses stratified on diabetes status, results were generally similar for patients with and without diabetes except for the suggestion of a greater rate of incident renal disease in patients without diabetes who used P-S + thiazide (HR 1.23, 1.01–1.49) compared with ACE-I + thiazide. Conclusions: Contrary to expectation, ACE-I with thiazide was associated with an increased incidence of renal disease compared with all other groups except potassium-sparing diuretics. This risk was significantly greater than that observed with BB. The association was not meaningfully changed by accounting for diabetes.

Background: The National Kidney Foundation’s Kidney Disease Outcomes Quality Initiative recommends the use of the Modification in Diet and Renal Disease (MDRD) equation to estimate glomerular filtration rate (eGFR) from serum creatinine, age, sex, and race. However, in many managed care organizations (MCOs) race is not routinely collected. The current analysis compares the use of areabased imputed race to actual race in using the MDRD equation to calculate eGFRs and to identify chronic kidney disease (CKD) in MCOs. Methods: The study population consisted of 9,196 Kaiser Permanente Georgia members with two eGFR measures

Background/Aims: Determination of hyperkalemia associated adverse outcomes incidence and risk assessment is complicated by lack of consistent hyperkalemia definitions across studies. Further, information about hyperkalemia from clinical trials, while reflecting the potential of renin-angiotensin-aldosterone system (RAAS) inhibitor treatment to increase serum potassium (K) concentration above a defined level, may not reflect risk or severity of outcomes. We sought to examine to what extent increasing levels of restriction influenced incidence estimates of hyperkalemia outcomes. Methods: The study cohort was drawn from a population of adult patients with diabetes at 3 HMORN sites. We identified all new users of a RAAS inhibitor between 01/01/2001 and 12/31/2006 and assessed hyperkalemia-associated outcomes within the first year of therapy. The initial definition of a hyperkalemia outcome to which other definitions were compared included any ambulatory (AV), emergency department (ED) or inpatient (IP) visit with a K level > 5.5 mmol/l or a coded hyperkalemia diagnosis within 7 days of the visit. The following restrictions were then applied: increasing minimum K concentration to > 6.0 mmol/l; reducing timeframe to 24 hours; and removing AV. Crude incidence rates of hyperkalemia-associated adverse outcomes were calculated using person years (p-y) determined as time from drug initiation to first outcome or other censoring event (e.g., drug discontinuation, end of study). Results: The cohort included 27,362 patients. Mean duration of initial therapy was 212 days. Hyperkalemia-associated outcome incidence estimates varied from 33.6 per 1000 p-y (defined as AV, ED or IP visit with K > 5.5 or a coded hyperkalemia diagnosis within 7 days) to 11.0 per 1000 p-y (defined as IP or ED visit with K > 6 or coded diagnosis within 24 hours). Removing AV had the greatest effect and reducing the timeframe to 24 hours had the least effect on incidence estimates. Conclusions: Modifying hyperkalemia definition criteria resulted in up to a 3-fold difference in by estimates of hyperkalemia-associated adverse outcomes. Further work linking these findings to adverse events is critical to decisions regarding appropriate definitions to answer specific study questions. We caution against comparing incidence estimates across published studies without considering the severity implications of differences in hyperkalemia outcomes definitions.

Conducting matched case/control studies can be difficult when matching on very large populations. We have developed an algorithm to identify a pair-wise matched case/control population for retrospective cohort studies. We have implemented this method using SAS code at two sites for the Vaccine Safety DataLink (VSD) project. This algorithm is superior as it works on certain extremely large populations such as those at certain Kaiser Permanente sites. The algorithm is thoroughly tested, robust, and effective for use on multiple sites projects to identify a pair wise matched case/control population, regardless of the dataset size each Medical Care Organization maintains.

Background: Heart Failure (HF) is one of the most common and serious progressive illnesses among elderly patients. It is usually detected at a relatively advanced stage, after irreversible damage has occurred. Early detection offers the potential to substantially reduce patient disability and health care costs. Objective: To develop a novel detection strategy, making use of longitudinal electronic health record (EHR) data to create an HF early detection prediction model. Methods: All data for this study were obtained from Geisinger Clinic’s EHR among patients who had a primary care provider. A prediction model was developed using a nested case control study design, where HF cases diagnosed between 2003–2006 were identified and controls were randomly selected matched on sex, age, and clinic. We used conditional logistic regression to model the relation between EHR data and detection of HF 6+ months and 18+ months before the actual date of diagnosis. Variables for the model included diagnoses, the most recent lab and clinical (e.g., SBP, DBP, pulse pressure) measures, medication orders and ambulatory care use in the previous two years, and smoking status. Data were only used if they occurred in the record either 6+ months or 18+ months before the diagnosis date, depending on the specific model. Model results were validated by combining a bootstrap re-sampling approach with a backwards elimination selection method. Results: We identified at least one matching control for 2,239 of the 2,764 cases; 9 or 10 controls were identified for 81% of the cases. A total of 24,249 controls were selected. The model for detecting HF 6+ months before usual diagnosis had an AUC of 0.80; the parallel model for detecting HF 18+ months before usual diagnosis had an AUC of 0.75 (95% CI: 0.73, 0.79). The AUC findings were similar for separate models completed on systolic HF and diastolic HF. Conclusions: In practice, clinicians do not have the time or ability to process seemingly disparate data points over a series of visits that might suggest a preclinical HF for a given patient. Our analysis of EHR data indicate that HF can be detected 6 or more months before usual diagnosis with good AUC and high specificity. These findings suggest that routine evaluation of EHR data may be useful in screening for patients at high risk of HF, creating numerous opportunities for early and aggressive intervention and potentially altering the natural history of heart failure for many patients.

Background: Hypertension (HTN) control may be highly dependent on patient behaviors, including filling of antihypertensive medication prescriptions (AHM Rxs). We examined electronic health records (EHR) for patients of Geisinger Clinic who had prescription benefits through Geisinger Health Plan. Methods: Prescriptions in the EHR were defined as filled if a claim was paid within 90 days. We used logistic regression to test if a patient’s filling rate (FR) was associated with gender, age, economic status (measured as median household income by ZIP code) and number of vascular comorbidities (angina, MI, PVD, stroke). We then tested whether a higher categorical FR (0, 1–33, 33–67, 67–99 or 100%) resulted in a greater percent of office visits with blood pressure (BP) controlled to JNC7 standards (%BPc). Results: All results are reported as Adjusted Odds Ratios (AOR) [95% CI]. Among 20,161 patients who met JNC7 criteria for HTN and had >3 AHM Rxs during plan enrollment from 2002 to 2006, 99% were Caucasian and 53% were female with a mean age of 58.5 years (SD 15.8). Of 191,120 AHM Rxs (mean per patient 9.5, SD 6.5), 49%, 59% and 62% were filled within 30, 60 and 90 days. Based on a 90-day FR, 14.7% of patients filled no Rxs and 36.2% filled all. Factors negatively associated with FR were female gender (AOR 0.89 [CI 0.870.90]), increasing age (nonlinear, 0.30 [0.290.30] for age 60–80 years vs. age 40–60 years), and number of vascular comorbidities (nonlinear, 0.79 [0.770.81] for 1 vs. 0 comorbidities, 0.66 [0.610.71] for 2 vs. 0 comorbidities). Higher economic status was positively associated with FR (1.05 [1.031.07] per $10k estimated annual income). In 19,500 patients who had >3 office visits, the mean number of BP readings per patient was 19.6 (SD 10.6), and 60% of readings were under control. After adjusting for the factors above and baseline blood pressures, there was a positive nonlinear association between FR and %BPc (1.08 [1.06–1.13] for 33–67% vs. 0% FR, 1.09 [1.08–1.16] for 67–99% vs. 0% FR, and 1.34 [1.07–1.46] for 100% vs. 0% FR). Results were similar if we excluded patients with 0% FR who may have used alternate insurance for Rxs. Conclusions: These findings suggest that prescription filling differs by gender, age and comorbidity status and that higher Rx filling rates are associated with better BP control.

Background: Heart failure (HF) disproportionately affects older adults and previous studies have suggested that the demographic as well as clinical profile of older patients with HF is different from that of younger patients. However, generalizable, population-based data on the clinical, treatment, and prognostic profile of older as compared to middle aged and younger patients with HF are lacking. Methods and Results: Residents of the Worcester (MA) metropolitan area hospitalized for de-compensated HF at 11 greater Worcester medical centers during 1995 and 2000 (n=4,534) were compared according to 4 age groups ( 85 years). The mean age of patients with acute HF was 76 years and 24% were >85 years. Older patients (>75) were more likely to be female and to have higher ejection fraction findings, multiple comorbidities, and a lower body mass index. Older patients were significantly more likely to receive symptom modifying medications and less likely to receive disease modifying medications than younger patients. Advanced age was directly associated with increased in- hospital, 30-day, and 1-year death rates in both crude and adjusted analyses. Conclusions: The results of this communitywide study suggest that clinical, treatment, and prognostic factors differ by age in patients hospitalized for de-compensated HF. These high-risk patients warrant special attention in future studies in order to improve their management and long-term survival.

Background: Researchers often consider mixing modes for data collection to increase response rate. However, it is possible that people respond differently when asked sensitive questions (like questions about depression) by mail compared to phone. Methods: A random sample of 4000 adolescent health plan enrollees (ages 13–17 years) were invited to participate in a brief survey examining exercise, sedentary behaviors, and depressive symptoms. Mailings were directed at parents and included an invitation letter, a consent form, survey, and a $2 pre-incentive for the youth. Parents were instructed to complete the consent form and give the survey to their child to complete and return. Non-responders were sent a second survey. Among remaining non- responders, 60% of the adults were phoned, asking them to review and sign the consent form and to ask their child to complete the screener. Forty percent of remaining non-responders were phoned asking the adults’ permission to complete the screener with the child by phone. In a second study, 9,838 adults with a history of diabetes or heart disease were invited to participate in a randomized controlled trial. The initial contact was a mailed 5-item survey that included a depression (PHQ2) screener and eligibility questions, along with a study information sheet and a $2 pre-incentive. Approximately 12 days later, non-responders were sent a second survey, and 10 days later, non- responders were called to complete the screener by phone. Results: Data collection is being completed as this abstract is submitted. This poster will examine whether there are differences in depression scores among those who are early responders (complete after one mailing), those who complete after a second mailing, and those who complete by telephone. We will also examine whether there are demographic differences at each step. In addition, we will demonstrate the increases in response rate by adding a phone call to complete a mailed survey. Conclusions: This poster will indicate whether this mixing of modes compromises data quality or produces different responses than if only one mode was used.

Background and Aims: Angiotensin converting enzyme (ACE) inhibitors are widely used in patients with chronic kidney disease (CKD) because the drugs slow disease progression. But some physicians are reluctant to prescribe ACE inhibitors “or use higher doses” because of concerns about the risk of hyperkalemia (high potassium), a potentially fatal adverse effect. Physicians need a tool to help them predict the risk of hyperkalemia; for example, high risk patients could be targeted for intensive laboratory monitoring. We synthesized known predictors of hyperkalemia into a prognostic risk score to predict the risk of hyperkalemia. Methods: We assembled a retrospective cohort of adult patients with possible CKD (at least one recently estimated glomerular filtration rate (eGFR) value less than 60 mL/min/1.73m2) who started an ACE inhibitor between 1998 and 2006 at Kaiser Permanente Northwest, a health maintenance organization (HMO). We followed patients for the earliest evidence of hyperkalemia: potassium value > 5.5 mmol/L; diagnosis code for hyperkalemia (ICD-9-CM 276.7). Cox regression synthesized known predictors of hyperkalemia that were recorded in the electronic medical record or KPNW databases into a risk score to predict the absolute risk of hyperkalemia 90 days after starting therapy. Results: We followed 5,097 patients who experienced 145 hyperkalemic events, a 90-day risk of 2.1% (95% CI 1.9% to 2.4%). The following baseline characteristics predicted hyperkalemia and contributed to the risk score: age, eGFR, diabetes, heart failure, current use of potassium supplements, current use of potassium sparing diuretics (e.g., spironolactone), and a high starting dose for the ACEinhibitor. The risk score discriminated high-risk patients (top quintile, observed risk of 7%) from low risk patients (bottom quintile, observed risk of 0.7%). Predicted and observed risks agreed closely (within 1%) for each quintile. Conclusions: The risk score separated high-risk patients (who may need more intensive laboratory monitoring) from low risk patients (whose physicians could use ACE-inhibitors more confidently). If validated in other clinical settings, the risk score could improve the efficiency of laboratory monitoring for hyperkalemia by focusing attention on higher risk patients.