Development and external validation of prediction models for adverse health outcomes in rheumatoid arthritis: A multinational real-world cohort analysis.

BACKGROUND: Identification of rheumatoid arthritis (RA) patients at high risk of adverse health outcomes remains a major challenge. We aimed to develop and validate prediction models for a variety of adverse health outcomes in RA patients initiating first-line methotrexate (MTX) monotherapy. METHODS: Data from 15 claims and electronic health record databases across 9 countries were used. Models were developed and internally validated on Optum® De-identified Clinformatics® Data Mart Database using L1-regularized logistic regression to estimate the risk of adverse health outcomes within 3 months (leukopenia, pancytopenia, infection), 2 years (myocardial infarction (MI) and stroke), and 5 years (cancers [colorectal, breast, uterine] after treatment initiation. Candidate predictors included demographic variables and past medical history. Models were externally validated on all other databases. Performance was assessed using the area under the receiver operator characteristic curve (AUC) and calibration plots. FINDINGS: Models were developed and internally validated on 21,547 RA patients and externally validated on 131,928 RA patients. Models for serious infection (AUC: internal 0.74, external ranging from 0.62 to 0.83), MI (AUC: internal 0.76, external ranging from 0.56 to 0.82), and stroke (AUC: internal 0.77, external ranging from 0.63 to 0.95), showed good discrimination and adequate calibration. Models for the other outcomes showed modest internal discrimination (AUC < 0.65) and were not externally validated. INTERPRETATION: We developed and validated prediction models for a variety of adverse health outcomes in RA patients initiating first-line MTX monotherapy. Final models for serious infection, MI, and stroke demonstrated good performance across multiple databases and can be studied for clinical use. FUNDING: This activity under the European Health Data & Evidence Network (EHDEN) has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No 806968. This Joint Undertaking receives support from the European Union's Horizon 2020 research and innovation programme and EFPIA.

Predictors of chronic opioid therapy in Medicaid beneficiaries with HIV who initiated antiretroviral therapy.

The factors associated with chronic opioid therapy (COT) in patients with HIV is understudied. Using Medicaid data (2002-2009), this retrospective cohort study examines COT in beneficiaries with HIV who initiated standard combination anti-retroviral therapy (cART). We used generalized estimating equations on logistic regression models with backward selection to identify significant predictors of COT initiation. COT was initiated among 1014 out of 9615 beneficiaries with HIV (male: 10.4%; female: 10.7%). Those with older age, any malignancy, Hepatitis C infection, back pain, arthritis, neuropathy pain, substance use disorder, polypharmacy, (use of) benzodiazepines, gabapentinoids, antidepressants, and prior opioid therapies were positively associated with COT. In sex-stratified analyses, multiple predictors were shared between male and female beneficiaries; however, chronic obstructive pulmonary disease, liver disease, any malignancy, and antipsychotic therapy were unique to female beneficiaries. Comorbidities and polypharmacy were important predictors of COT in Medicaid beneficiaries with HIV who initiated cART.

Signal Detection and Methodological Limitations in a Real-World Registry: Learnings from the Evaluation of Long-Term Safety Analyses in PSOLAR.

INTRODUCTION: Psoriasis Longitudinal Assessment and Registry (PSOLAR) was designed in 2007 as the first disease-based registry for patients with psoriasis. OBJECTIVE: The aim of this study was to discuss methodological limitations and post hoc analyses in long-term safety registries using learnings from analyses of a potential safety risk for major adverse cardiovascular events (MACE) in PSOLAR. METHODS: PSOLAR is an international observational study of over 12,000 psoriasis patients that was conducted to meet postmarketing safety commitments for infliximab and ustekinumab. A recent annual review of registry data indicated a potential MACE risk for ustekinumab vs. non-biologics based on prespecified COX model regression analyses, which yielded an adjusted hazard ratio (HR) of 1.533 (95% confidence interval [CI] 1.103-2.131). Therefore, we conducted a comprehensive review of key statistical methodology and implemented post hoc analytical methods to address specific limitations. RESULTS: The following limiting factors were identified: (1) inclusion of both prevalent and incident (new) users of biologics; (2) unanticipated imbalances in patient characteristics between treatment cohorts at baseline; (3) limited availability of relevant clinical data after enrollment; and (4) divergence of characteristics associated with outcomes among comparator groups over time. The analysis was modified to include only incident users, propensity scores were used to weight HRs, and adalimumab was deemed a more clinically appropriate comparator. The revised HR was 0.820 (95% CI 0.532-1.265), indicating no meaningful increase in MACE risk for ustekinumab. CONCLUSION: Our results, which do not support a causal association between ustekinumab exposure and MACE risk, underscore the need for ongoing assessment of analytical methods in long-term observational studies.

Alpha-1 Antitrypsin Deficiency-Associated Clinical Manifestations and Healthcare Resource Use in the United States.

Pulmonary events (PEs) associated with alpha-1 antitrypsin deficiency (AATD) can have a severe clinical course and increase healthcare resource use (HRU). However, AATD-associated HRU and healthcare costs have not been extensively described. This study describes and compares real-world HRU and healthcare costs among US patients with severe (requiring hospitalization after AATD-related PE) versus nonsevere AATD clinical course. Administrative healthcare claims for patients with a second primary AATD diagnosis between 6/1/2008 and 12/31/2017 were analyzed from 2 databases (requiring continuous enrollment 6 months preceding diagnosis). Patient baseline characteristics and AATD-associated PE incidence rates, HRU, and healthcare costs during follow-up were compared in patients with severe versus nonsevere AATD. Of 5109 patients with a second AATD diagnosis, 2674 (severe, n = 711 [26.6%]; nonsevere, n = 1963 [73.4%]) had ≥1 AATD-associated PE. PE incidence per 100 person-years was higher in patients with severe versus nonsevere AATD. Annual incidences (mean ± SD) of emergency department (1.2 ± 5.7 vs. 0.4 ± 1.2), inpatient (1.3 ± 2.5 vs. 0.1 ± 0.5), and outpatient (10.3 ± 15.9 vs. 5.7 ± 13.2) visits were higher in patients with severe versus nonsevere AATD. Median (interquartile range) annual costs were also higher for patients with severe versus nonsevere AATD for emergency department ($185 [$0-$1665] vs. $0 [$0-$264]), inpatient ($16,038 [$2968-$70,941] vs. $0 [$0-$0]), and outpatient ($2663 [$412-$10,277] vs. $1114 [$134-$4195]) visits. Higher percentages of patients with severe AATD were prescribed augmentation therapy, antibiotics, or corticosteroids. These findings suggest that patients with severe AATD have higher incidence of AATD-associated PEs, as well as higher HRU and healthcare costs.

Leveraging unstructured data to identify hereditary angioedema patients in electronic medical records.

BACKGROUND: The epidemiologic impact of hereditary angioedema (HAE) is difficult to quantify, due to misclassification in retrospective studies resulting from non-specific diagnostic coding. The aim of this study was to identify cohorts of patients with HAE-1/2 by evaluating structured and unstructured data in a US ambulatory electronic medical record (EMR) database. METHODS: A retrospective feasibility study was performed using the GE Centricity EMR Database (2006-2017). Patients with ≥ 1 diagnosis code for HAE-1/2 (International Classification of Diseases, Ninth Revision, Clinical Modification 277.6 or International Classification of Diseases, Tenth Revision, Clinical Modification D84.1) and/or ≥ 1 physician note regarding HAE-1/2 and ≥ 6 months' data before and after the earliest code or note (index date) were included. Two mutually exclusive cohorts were created: probable HAE (≥ 2 codes or ≥ 2 notes on separate days) and suspected HAE (only 1 code or note). The impact of manually reviewing physician notes on cohort formation was assessed, and demographic and clinical characteristics of the 2 final cohorts were described. RESULTS: Initially, 1691 patients were identified: 190 and 1501 in the probable and suspected HAE cohorts, respectively. After physician note review, the confirmed HAE cohort comprised 254 patients and the suspected HAE cohort decreased to 1299 patients; 138 patients were determined not to have HAE and were excluded. The overall false-positive rate for the initial algorithms was 8.2%. Across final cohorts, the median age was 50 years and > 60% of patients were female. HAE-specific prescriptions were identified for 31% and 2% of the confirmed and suspected HAE cohorts, respectively. CONCLUSIONS: Unstructured EMR data can provide valuable information for identifying patients with HAE-1/2. Further research is needed to develop algorithms for more representative HAE cohorts in retrospective studies.

Incidence of von Willebrand disease in Denmark, 1995-2016: A cohort study.

INTRODUCTION: Information about temporal development of von Willebrand disease (VWD) incidence at a population level is scarce. To our knowledge, no study has described the incidence of VWD at a population level. AIM: To estimate overall and annual incidence rates of hospital diagnosed VWD in Denmark between 1995 and 2016 as well as the frequency of hospital treated bleeding episodes before and after VWD diagnosis. METHODS: A registry-based cohort study that included all Danish patients with a first diagnosis of VWD in Denmark, identified in the Danish National Patient Registry through 1995-2016. RESULTS: We identified 1,035 patients with a diagnosis of VWD. The overall incidence rate of VWD in 1995-2016 was 8.6 (95% CI: 8.1-9.2). The annual age-standardized incidence rate per 100 000 person-years varied between 4.1 (95% CI: 2.4-5.9) in 1998 and 16.7 (95% CI: 13.1-20.3) in 2005. A prominent peak in rates appeared from 2002 to 2008. One and five years before VWD diagnosis, 6% and 11.5% of the patients had at least one hospital treated bleeding episode. One and five years after diagnosis, the corresponding percentages were 7.9% and 13.4%. CONCLUSION: These results are the first population-based estimates of VWD incidence. The incidence may be underestimated because asymptomatic individuals may not be diagnosed. The observed peak in incidence from 2002-2008 may be explained by increased medical attention, leading to more patients being diagnosed, rather than an actual increase in VWD incidence. However, overall, we observed no systematic changes in VWD incidence over the study period.

Masking in Pragmatic Trials: Who, What, and When to Blind.

Masking (or blinding) of treatment assignment is routinely implemented in classical randomized clinical trials (RCTs) to isolate the effect of the intervention itself and to minimize the potential for bias that could occur with traditional trials. Such biases could be introduced with the conduct, assessment of endpoints, management of conditions, analysis, and reporting when the treatment assignments are known. However, masking of treatments is not only complex but it hinders how generalizable the findings are to the "real world" clinical setting. Pragmatic RCTs (pRCTs) are intended to evaluate the effects of interventions within routine medical care, and as such, do not typically mask treatment groups; moreover, pRCTs assess comparators that are available in routine medical practice, not masked placebos. Whether pRCTs should be masked if intended for regulatory or other purposes has recently been questioned. The literature on pRCTs, while extensive, does not address how much actual benefit is gained from masking outcomes and how masking may affect the "real world" nature of a study. Here, we propose an approach to evaluate sources of bias, describe stakeholders in the conduct of pRCTs who are most likely affected, and offer a framework for considering how masking may be implemented effectively while maintaining generalizability.

Real-World Evidence: What It Is and What It Can Tell Us According to the International Society for Pharmacoepidemiology (ISPE) Comparative Effectiveness Research (CER) Special Interest Group (SIG).

On December 8, 2016, the New England Journal of Medicine published a sounding board on Real World Evidence (RWE)1 by the US Food and Drug Administration (FDA) leadership. While the value of RWE based on nonrandomized observational studies was appreciated, such as for hypothesis generating, safety, and measuring quality in healthcare delivery, the authors expressed concerns on the quality of data sources and the ability of methodologies to control for confounding. In response, we offer a few considerations regarding these concerns.

Using administrative data for your research project: 10 considerations before you begin.

With increasing pressure to conduct research during residency training, and given the availability of administrative claims data, pharmacy residents will likely consider using large administrative databases for their research project. With competing time commitments and the short duration of residencies, residents and their preceptors must consider the 10 factors outlined above in order to produce a thoughtful, clinically relevant research project. While this discussion focused on the completion of a residency research project, these topics are also relevant to a broader pharmacy audience. Colleges of pharmacy are increasingly requiring research projects as part of their curriculum, and pharmacy students and practitioners often consider obtaining additional degrees requiring a research component. Both students and practitioners can use the guidance provided herein when planning research projects and investigations to aid in the successful completion of research using administrative claims data.

Evaluating the incident user design in the HIV population: incident use versus naive?

INTRODUCTION: The incident user design is the preferred study design in comparative effectiveness (CER) research. Usually, 180-365 days of exposure free time is adequate to remove biases associated with inclusion of prevalent users. In HIV research, the use of antiretrovirals (ARVs) at any time in the past may influence future treatment choices and CER results; thus, identifying naive as opposed to incident users is of importance. We examined misclassification of antiretroviral naive status based on Medicaid administrative data through linkage to the UNC CFAR HIV Clinical Cohort (UCHCC). METHODS: We identified Medicaid patients with incident exposure to common first-line ARV regimens between 2002 and 2008 that were also patients enrolled in the UCHCC. We calculated the proportion of antiretroviral naive patients based on the UCHCC, among patients identified as having incident exposure in Medicaid and examined factors associated with being antiretroviral naive in both data sources using logistic regression to generate prevalence odds ratios and associated 95% confidence intervals. RESULTS: Of the 3500 Medicaid patients with incident antiretroviral (ARV) exposure, 1344 were also enrolled in the UCHCC. In this sample, 34% were antiretroviral naive at the time of first exposure in the Medicaid data based on the UCHCC. In multivariable models, higher CD4 cell counts and log HIV RNA values were associated with being antiretroviral naive in both data sources. CONCLUSIONS: Administrative data are an important source of information related to HIV treatment. As the construction of a durable and long-lasting HIV treatment plan involves knowledge of current and past antiretroviral therapy, augmentation of this data with comprehensive clinical cohort information is necessary.

Validation of Medicaid Claims-based Diagnosis of Myocardial Infarction Using an HIV Clinical Cohort.

BACKGROUND: In nonexperimental comparative effectiveness research using health care databases, outcome measurements must be validated to evaluate and potentially adjust for misclassification bias. We aimed to validate claims-based myocardial infarction (MI) algorithms in a Medicaid population using an HIV clinical cohort as the gold standard. METHODS: Medicaid administrative data were obtained for the years 2002-2008 and linked to the UNC CFAR HIV Clinical Cohort based on social security number, first name, and last name and MI were adjudicated. Sensitivity, specificity, positive predictive value, and negative predictive value were calculated. RESULTS: There were 1063 individuals included in the study. Over a median observed time of 2.5 years, 17 had an MI. Specificity ranged from 0.979 to 0.993 with the highest specificity obtained using the ICD-9 code 410.xx in the primary or secondary position and a length of stay >3 days. Sensitivity of MI ascertainment varied from 0.588 to 0.824 depending on algorithm. CONCLUSIONS: Specificities of varying claims-based MI ascertainment criteria are high but small changes impact positive predictive value in a cohort with low incidence. Sensitivities vary based on ascertainment criteria. Type of algorithm used should be prioritized based on study question and maximization of specific validation parameters that will minimize bias while also considering precision.

Effects of combination antiretroviral therapies on the risk of myocardial infarction among HIV patients.

BACKGROUND: Cohort studies have demonstrated greater risk of myocardial infarction (MI) associated with specific antiretroviral use, while meta-analyses of randomized controlled trials (RCTs) have not. These differences may be due to inherent biases in the observational study design or to the limited duration of randomized trials. We conducted a new-user, active-comparator cohort study emulating an RCT comparing the initiation of several antiretrovirals as part of combination antiretroviral therapy (cART) and MI. METHODS: We included North Carolina (NC) Medicaid beneficiaries infected with human immunodeficiency virus between 2002 and 2008 who were previously untreated with cART. We compared hazard ratios (HRs) and 95% confidence intervals (CIs) of MI between abacavir and tenofovir recipients, and lopinavir-ritonavir or atazanavir recipients and nonnucleoside reverse transcriptase inhibitor (NNRTI) recipients. We adjusted for confounding through inverse probability weighting methods. RESULTS: There were 3481 NC Medicaid new cART recipients who contributed 6399 person-years and experienced 38 MI events. Receiving abacavir compared with tenofovir as part of cART was associated with an increased rate of MI (unadjusted HR = 2.70 [95% CI = 1.24-5.91]; adjusted HR = 2.05 [0.72-5.86]). Point estimates also suggest a relationship between receipt of atazanavir or lopinavir-ritonavir compared with an NNRTI and MI, although estimates were imprecise. CONCLUSIONS: We found an increased rate of MI among patients initiating abacavir compared with tenofovir, although the association was decreased after confounding adjustment. Without a very large prospective comparative clinical trial, a much larger observational study of patients initiating cART would be needed to better define this apparent association.

Sex differences in benzodiazepine use in the HIV-infected population.

In the HIV-infected population there is a high prevalence of psychiatric disorders, conditions that often coexist with drug and alcohol dependence. Symptoms associated with psychiatric disorders are frequently managed with benzodiazepines, a class of medication often abused. We examined whether HIV-infected patients were more likely to fill a benzodiazepine prescription than their uninfected counterparts using a privately insured, nationally representative sample receiving clinical care between January 2007 and December 2009. Odds ratios (OR) and 95% confidence intervals (CI) to quantify the likelihood of receiving a benzodiazepine were calculated using multivariate logistic regression models. We examined the presence of interaction between HIV infection and sex using backwards elimination and by comparing stratum-specific OR to identify clinically meaningful differences. Overall, 323,796 beneficiaries were included in the sample, of which 723 were HIV infected. Bivariate analyses showed that compared to the uninfected sample, HIV-infected patients were more likely to have filled a benzodiazepine prescription (24% vs. 19%) during the study period. HIV-infected patients were also more likely to be male (80% vs. 44%), black (21% vs. 7%) and have a diagnosis of depression (12% vs. 8%) or insomnia (6% vs. 3%) than were uninfected patients. Adjusted for other covariates, HIV infection was associated with an increase (OR): 1.68, 95% CI: 1.39, 2.02) in the likelihood of filling a benzodiazepine prescription. When stratified by sex, HIV-infected males were more likely (OR: 1.68, 95% CI: 1.05, 2.67) than uninfected males to fill a benzodiazepine prescription while there was no observed difference in the likelihood of filling a benzodiazepine prescription between HIV-infected and uninfected females (OR: 1.12, 95% CI: 0.73, 1.70). Our findings suggest that HIV-infected patients, particularly HIV-infected males, are more likely to fill benzodiazepine prescriptions than their uninfected counterparts, highlighting the need for further research to investigate reasons for these observed differences.

Communicating quantitative risks and benefits in promotional prescription drug labeling or print advertising.

PURPOSE: Under the Food, Drug, and Cosmetic Act, all promotional materials for prescription drugs must strike a fair balance in presentation of risks and benefits. How to best present this information is not clear. We sought to determine if the presentation of quantitative risk and benefit information in drug advertising and labeling influences consumers', patients', and clinicians' information processing, knowledge, and behavior by assessing available empirical evidence. METHODS: We used PubMed for a literature search, limiting to articles published in English from 1990 forward. Two reviewers independently reviewed the titles and abstracts for inclusion, after which we reviewed the full texts to determine if they communicated risk/benefit information either: (i) numerically (e.g., percent) versus non-numerically (e.g., using text such as "increased risk") or (ii) numerically using different formats (e.g., "25% of patients", "one in four patients", or use of pictographs). We abstracted information from included articles into standardized evidence tables. The research team identified a total of 674 relevant publications, of which 52 met our inclusion criteria. Of these, 37 focused on drugs. RESULTS AND CONCLUSIONS: Presenting numeric information appears to improve understanding of risks and benefits relative to non-numeric presentation; presenting both numeric and non-numeric information when possible may be best practice. No single specific format or graphical approach emerged as consistently superior. Numeracy and health literacy also deserve more empirical attention as moderators.

Comparative effectiveness research using electronic health records: impacts of oral antidiabetic drugs on the development of chronic kidney disease.

PURPOSE: Little is known about the comparative effects of common oral antidiabetic drugs ([OADs] metformin, sulfonylureas, or thiazolidinediones [THZs]) on chronic kidney disease (CKD) outcomes in patients newly diagnosed with type 2 diabetes (T2DM) and followed in community primary care practices. Electronic health records (EHRs) were used to evaluate the relationships between OAD class use and incident proteinuria and prevention of glomerular filtration rate decline. METHODS: A retrospective cohort study on newly diagnosed T2D cases requiring OADs documented in the EHRs of two primary care networks between 1998 and 2009 was conducted. CKD outcomes were new-onset proteinuria and estimated GFR (eGFR) falling below 60 ml/min/1.73 m(2). OAD exposures defined cohorts. Hazard ratios represent differential CKD outcome risk per year of OAD class use. RESULTS: A total of 798 and 977 patients qualified for proteinuria and eGFR outcome analyses, respectively. With metformin as the reference group, sulfonylurea exposure trended toward association with an increased risk of developing proteinuria ([adjusted hazard ratio; 95% CI] 1.27; 0.93, 1.74); proteinuria risk associated with THZ exposure (1.00; 0.70, 1.42) was similar to metformin. Compared with metformin, sulfonylurea exposure was associated with an increased risk of eGFR reduction to <60 ml/min/1.73 m(2) (1.41; 1.05, 1.91). THZ exposure (1.04; 0.71, 1.50) was not associated with change in the risk of eGFR decline. CONCLUSIONS: In a primary care population, metformin appeared to decrease the risk of CKD development compared with sulfonlyureas; risks of CKD development between metformin and THZs were similar. EHR use in pharmacotherapy comparative effectiveness research creates specific challenges and study limitations.

Initial and subsequent therapy for newly diagnosed type 2 diabetes patients treated in primary care using data from a vendor-based electronic health record.

BACKGROUND: Diabetes is a leading cause of death and disability, and its prevalence is increasing. When diet fails, patients with type 2 diabetes mellitus (T2DM) are prescribed oral hypoglycemics for glycemic control. Few studies have explored initial use or change from initial oral hypoglycemic therapy in the primary care setting. We aimed to describe the utilization of initial oral hypoglycemics among newly diagnosed patients with diabetes from 1998-2009 and changes from initial to subsequent therapy among patients prescribed older oral hypoglycemic agents using electronic health records. METHODS: This observational cohort study used electronic health records from newly diagnosed patients with T2DM between 1 January 1998 and 31 March 2009 at two large health systems in the USA. Oral hypoglycemics included older (biguanide, sulfonylurea, and thiazolidinedione) and newer agents (incretin mimetic agents, alpha-glucosidase inhibitors, and D-phenylalanine derivatives). Multinomial regression models were fit to evaluate initial older oral hypoglycemic medication. We used incidence density sampling and conditional logistic regression models to evaluate predictors of regimen change. RESULTS: Most patients were treated from the biguanide class of oral hypoglycemics (67%), but there were differences in initial prescribing by age and race. HbA1c (Odds Ratio for HbA1c 7.0-8.9 vs < 7.0, 5.87 [95% Confidence Interval: 3.62-9.52]; Odds Ratio for HbA1c ≥ 9 vs < 7.0, 20.25 [95% Confidence Interval: 8.32-49.29] and Black people (Odds Ratio, 0.29 [95% Confidence Interval: 0.14, 0.60]) versus White people were associated with regimen change in the adjusted analysis. CONCLUSIONS: Clinical and demographic characteristics influence choice and duration of initial oral hypoglycemic treatment as well as regimen changes.

Is HIV infection a risk factor for multi-drug resistant tuberculosis? A systematic review.

BACKGROUND: Tuberculosis (TB) is an important cause of human suffering and death. Human immunodeficiency virus (HIV), multi-drug resistant TB (MDR-TB), and extensive drug resistant tuberculosis (XDR-TB) have emerged as threats to TB control. The association between MDR-TB and HIV infection has not yet been fully investigated. We conducted a systematic review and meta-analysis to summarize the evidence on the association between HIV infection and MDR-TB. METHODS AND RESULTS: Original studies providing Mycobacterium tuberculosis resistance data stratified by HIV status were identified using MEDLINE and ISI Web of Science. Crude MDR-TB prevalence ratios were calculated and analyzed by type of TB (primary or acquired), region and study period. Heterogeneity across studies was assessed, and pooled prevalence ratios were generated if appropriate. No clear association was found between MDR-TB and HIV infection across time and geographic locations. MDR-TB prevalence ratios in the 32 eligible studies, comparing MDR-TB prevalence by HIV status, ranged from 0.21 to 41.45. Assessment by geographical region or study period did not reveal noticeable patterns. The summary prevalence ratios for acquired and primary MDR-TB were 1.17 (95% CI 0.86, 1.6) and 2.72 (95% CI 2.03, 3.66), respectively. Studies eligible for review were few considering the size of the epidemics. Most studies were not adjusted for confounders and the heterogeneity across studies precluded the calculation of a meaningful overall summary measure. CONCLUSIONS: We could not demonstrate an overall association between MDR-TB and HIV or acquired MDR-TB and HIV, but our results suggest that HIV infection is associated with primary MDR-TB. Future well-designed studies and surveillance in all regions of the world are needed to better clarify the relationship between HIV infection and MDR-TB.

Geographic Information System mapping as a tool to assess nonresponse bias in survey research.

BACKGROUND: Surveys are a useful tool for assessing professional practice patterns, although declining response rates have caused concern over external validity. This is particularly relevant to Web-based surveys, where response rates traditionally have been lower than with paper mail surveys. In a 2005 survey of North Carolina community pharmacy managers using a Web-based data collection instrument, we achieved an overall response rate of 23%. OBJECTIVE: To explore nonresponse bias using accepted methods and to test whether Geographic Information System mapping is a useful tool for assessing response bias. METHODS: Cross-sectional survey of 1593 community pharmacy managers in North Carolina using a Web-based tool. Nonresponse bias was assessed quantitatively by comparing early responders with late responders (ie, wave analysis) and by comparing respondents with nonrespondents with regard to known pharmacy, pharmacist, and population characteristics. Significant variables from these analyses were then mapped using ArcGIS 9.1. RESULTS: Pharmacy type was identified as a predictor of response, with independent pharmacies less likely to respond than chain pharmacies (odds ratio 0.75; 95% confidence interval 0.59-0.95). This conclusion was consistent in the wave analysis and the analysis of known population characteristics. Other county-level variables such as the number of physicians per capita, income, and the percentage of residents eligible for Medicaid showed trends but were not statistically significant (P<.1). Geographic Information System mapping was able to descriptively illustrate nonresponse bias for pharmacy type but trends were more difficult to detect for statistically insignificant trends. CONCLUSION: The best way to avoid nonresponse bias is to improve response rates. When this is not possible, Geographic Information System mapping has some utility for assessing nonresponse bias, and for aggregating known population characteristics based on location. It is most useful in conjunction with other accepted techniques such as wave analysis and analysis of known population characteristics.