| Becker 1972[34] |
Grants from the National Center for Health Services Research and Development USPHS &from National Institute of Health DHEW |
Low: All actively practicing primary physicians in the county included with a response rate of 84% (37 out of 44) |
High: Self-reported scale to collect data on physician’s use of detail men as sources of prescribing information. |
Low for prescription pattern outcome (computer-generated profiles of actual prescribing of chloramphenicol); High for prescription appropriateness (Two panels of experts evaluated physician's general prescribing behavior relative to five common complaints and five common illnesses) |
Unclear: Controlled for variance in numbers and types of patients. |
Unclear: Author did not comment on completeness of data |
| Haayer, 1982[31] |
Ziekenfondsraad (Health Insurance Fund) |
Low: Population-based study targeting all GPs (148) in Twente, resulting in a final list of 131 GPs (118 GPs agreed to participate) |
High: Study interview pilot tested but not validated. Risk of recall bias and/or social desirability bias since it is self-reported by physician |
High: Study questionnaire pilot tested but not validated. A panel of experts rated the rationality of prescribing on four different scales that have then been combined. |
Low: Authors conducted stepwise multiple regression to adjust for potential confounders. |
Unclear: Author did not comment on completeness of data |
| Bowman 1988[27] |
Not reported |
High: Sampling method not explained; low response rate (49%); characteristics of non-responders and responders were not compared |
Low: No reason to suspect that measurement of attendance of course was not valid |
High: Non validated self-report survey was used. |
High: Control for confounding variables not reported (Analyses did not take into account other predictors of prescription behavior) |
High: Response rate for different courses varied between 43% and 76%) |
| Peay, 1988[37] |
Australian Research Grants Scheme and the Flinders University Research Budget |
Low: Clear sample selection and eligibility criteria with 60% response rate. |
High: Survey/interview method used to measure the exposure. Authors did not report on validity and reliability of the interview guide. |
High: No objective measurement of outcome (survey/interview method used to measure behavior. Also, authors did not report on the validity and reliability of the interview guide) |
Low: Additional multivariate analyses were carried out. |
Unclear: The author did not mention any missing data |
| Orlowski, 1992[29] |
Not reported |
Unclear: No clearly defined eligibility criteria or sampling method (physicians who had accepted invitations to attend symposia were identified by general questioning of colleagues and were affiliated with one institution) |
Low: No reason to suspect that measurement of attendance of symposia was not valid |
Low: Objective measurement of outcome; prescribing pattern was tracked retrospectively using the hospital pharmacy inventory usage reports; both drugs were used only in hospitalized physicians; National usage data for the 2 drugs was obtained from Pharmaceutical Data Services |
High: Study focuses on the symposia and ignores the impact of other approaches to marketing including advertisements, salesman contacts, and journal articles. Also, analysis did not adjust for the fact that the second course was offered approximately 20 months after drug B had been added to the hospital formulary |
Unclear: Author did not comment on completeness of data |
| Chren, 1994 |
National Institute of Arthritis, USA; Skin Diseases Research Center and Clinical Analysis Project, University Hospitals of Cleveland, |
Low: Clear sample selection and eligibility criteria. Response rate was 88% |
High: Survey pilot tested but not validated. Risk of recall bias and/or social desirability bias since it is self-reported by physician |
Low: Objective measurement of outcome by reviewing the standard formulary request forms |
Low: Multivariable logistic regression models controlled for physician age, gender, departmental appointment, and number of patients seen per week |
Unclear: The author did not mention any missing data |
| Figueiras 2000 and Caamano, 2002[25, 26] |
Spanish Ministry of Health and Consumption |
Low: Clear eligibility criteria with random selection of subjects. Exposed and control subjects were from the same population. Response rate was 75% |
Low: A “valid and reliable” self-administered mailed questionnaire was used to collect information on exposure |
Low: Objective measurement of outcome using “the database of the accounting archives of the National Health Service, which includes all prescriptions served in all the pharmacies in Galicia.” |
Low: Analyses controlled for confounding variables such as type of practice, number of identification cards, number of patients seen per day, the service accessibility\ for the patients, unemployment rate and population distribution. |
Low: Missing data were controlled for by carrying out multiple imputation |
| Mizik, 2004[35] |
Institute for the Study of Business Markets (ISBM) at Pennsylvania State University |
Unclear: No clearly defined eligibility criteria or sampling method |
Low: Objective measurement of exposure using panel data from US pharmaceutical manufacturer |
Low: Objective measurement of outcome using panel data from a U.S. pharmaceutical manufacturer |
Low: Dynamic fixed-effects distributed lag regression model controlled for a range of potential confounding factors |
Unclear: Author did not mention any missing data |
| Muijrers, 2005[36] |
Dutch Pharmacist’s Association and the CZ health insurance company |
Low: Clear selection criteria. Studied general practitioners in south of the Netherland. Response rate was 71%. |
High: Self-reported survey used to measure exposure. Authors did not report on validity and reliability of survey tool. |
Low: Objective measurement of prescribing indicators using a prescription database compiled by linking pharmacy databases from 379 pharmacies. |
Low: A multiple regression analysis included the a range of predictors |
Unclear: Author did not mention any missing data |
| Symm, 2006[32] |
Scott & White Institutional Research Fund |
Low: Clear selection criteria. The 25 sample medications selected comprised 84% of samples dispensed during study period |
Low: Objective measurement of exposure using the 2003 sample logs which were reported to be 95% to 100% accurate |
Low: Objective measurement of outcome using Scott & White Health Plan prescription claims data. |
High: Although case-mix adjustment data indicates very similar practices among the 3 clinics, “there may still have been differences that we overlooked or were unable to measure” |
Unclear: Author did not mention any missing data |
| Miller, 2008[28] |
Not reported |
Low: Clear eligibility criteria with all 10 attending physicians in a large resident-faculty practice selected. |
Low: Objective measurement of exposure (observation of existing sample cabinet being discontinued) |
Low: “Presence of an electronic pharmacy database allowed to abstract accurate data for a wide assortment of medications prescribed by a variety of physicians.” |
Low: Controlled for a range of potential confounders. Also conducted sensitivity analysis which did not significantly affect results |
Unclear: Author did not mention any missing data |
| Anderson 2009[33] |
Office of Medical Applications of Research, National Institutes of Health, and Maternal and Child Health Bureau, Health Resources and Services Administration, USA |
Low: Random selection of participants from a nationally representative database. Response rate of 49%; however, factors for which responders and non-responders differed were not associated with industry attitudes and interaction |
High: Study questionnaire pilot tested but not validated. Risk of recall bias and/or social desirability bias since it is self-reported by physician |
High: Study questionnaire pilot tested but not validated. Risk of recall bias and/or social desirability bias since it is self-reported by physician |
Low: Authors constructed three linear regression models controlling for: reading guidelines on physician-pharma interactions, physician characteristics, physician practice, physician perceived value of industry drug information. |
Unclear: Authors did not comment on completeness of data |
| Søndergaard, 2009[17] |
AstraZeneca funded the study through a grant to the Research Unit for General Practice in Odense |
Low: Population-based study targeting all GPs (191) in the county, resulting in a final list of 165 GPs |
Low: Objective measurement of exposure using AstraZeneca’s database |
Low: “Outcome data based on a highly valid and complete register covering all prescribed asthma drugs” |
Unclear: While authors controlled for calendar time and device preferences, they did not control for other factors such as competing firm’s drug marketing effort, which can also affect drug preference. |
Low: Only two of the requested to be withdrawn from the analysis |
| Pinckney, 2011[19] |
Freeman Medical Scholars Program, The Champlain Valley Area Health Education Center, and the Attorney General Consumer and Prescriber Grant Program |
High: Low response rate was low (35% of all 631 primary care clinicians practicing in the state of Vermont) |
Low: Absence or presence of sample closet in clinic measured using “several items from a survey developed and validated by Chew et al”. |
High: Prescription preference was based on a hypothetical scenario and not actual behavior. Attitude measurement used a scale that was not reported as validated |
Low: Authors used multivariable regression models to adjust for potential confounders. |
Low: Exclusion due to incomplete data less than 6% |
| Pedan 2011[12] |
Inventiv Health |
Low: Clear eligibility criteria. Results were “robust to alternative … sample selection criteria”. The panel is geographically and socioeconomically representative. |
Low: Objective measurement of exposure using the unique representative dataset. |
Low: Objective measure of outcome using the dispensing records from a large number of nationwide and regional pharmacy chains (inVentiv Health computerized pharmacy prescription database) |
Low: Authors accounted for competitive promotions, various physicians, practice settings, patient base, and market dynamic characteristics. |
Unclear: Authors did not comment on completeness of data |
| Lieb, 2014[14] |
No support or funding to report |
High: No clear eligibility criteria; Low response rate 11.5% (n = 160) |
High: No objective measurement of exposure; doctors completed an online questionnaire. Also, discrepancy in categorizing exposure and control group |
Low: Objective measurement using prescribing data over a year for all on-patent branded, off-patent branded, and generic drugs from the Bavarian Association of Statutory Health Insurance Physicians |
High: Control for confounding variables not reported “We have not recorded or taken into consideration any other factors that could influence the prescribing habits of doctors and may interact with the PSR visits” |
Unclear: Authors did not comment on completeness of data |
| Hurley, 2014[18] |
National Heart, Lung, and Blood Institute & National Center for Research Resources and the National Center for Advancing Translational Sciences, National Institutes of Health |
Unclear: While national data was obtained from the National Disease and Therapeutic Index (NDTI), physicians were selected from master lists of the American Medical Association (AMC) and the American Osteopathic Association through random sampling. Differences in demographics between patients at the AMC and on a national level |
Low: Objective measurement of outcome using data from a large academic medical center without samples extracted from Stanford University’s Epic electronic database via the Center for Clinical Informatics |
Low: Objective measurement of outcome using national data obtained from the National Disease and Therapeutic Index (NDTI). Drug prices were directly quoted from customer service representatives of a major pharmacy in July 2013. |
High: “The observed differences in prescribing habits may be attributed to other forms of pharmaceutical marketing that were not adequately captured in our study, such as the number of visits by or gifts from pharmaceutical representatives or the use of co-payment discount cards, which can also influence prescribing patterns |
Unclear: Authors did not comment on completeness of data |
| Dejong 2016 [15] |
National Center for Advancing Translational Sciences, National Institutes of Health; and by the Hawaii Medical Service Association Endowed Chair in Health Services and Quality Research at University of Hawaii |
Low: Clear eligibility criteria. The study population included 279 669 physicians. Of these, 155 849 physicians wrote more than 20 prescriptions in 1 of the 4 target drug classes and were assigned to study groups. |
Low: Objective measurement of exposure using the 2013 Open Payments database which describes the value and the drug or device being promoted for all payments to physicians from August through December 2013 |
Low: Objective measurement of prescribing data for individual physicians from Medicare Part D |
Low: Multivariable grouped logistic regression models with binomial physician-level prescribing data, and adjusting for a number of covariates |
Low: 5% of payments promoting the target drugs were excluded from the regression analysis |
| Yeh 2016 [16] |
Not reported |
Low: “From 363653 physicians in the Medicare Part D prescription claims database, we identified 9628 with a business address in Massachusetts, of whom 2444 had associated statin prescriptions covered by Medicare.” |
Unclear: Although exposure was measured using Massachusetts physicians payment database compiled by Massachusetts Department of Health, the authors were unable to determine the frequency of misattribution of the payment category or underreporting of payment |
Low: Objective measurement of outcome using Part D Medicare prescriptions claims data prepared by the Centers for Medicare and Medicaid Services (CMS). |
Unclear: Although authors mentioned conducting linear regression models, they stated that they were not able to control for certain physician characteristics (e.g., practice characteristics, level of experience) which may have an impact on prescribing patterns. |
Unclear: Authors did not mention any missing data |
