SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

J Clin Hypertens (Greenwich). 2012; 14:435–446. ©2012 Wiley Periodicals, Inc.

The authors explored to what extent important medical decisions by practitioners can be influenced by pharmaceutical representatives and, in particular, whether restricting such access could delay appropriate changes in clinical practice. Medical practices were divided into four categories based on the degree of sales representative access to clinicians: very low, low, medium, and high from a database compiled by ZS Associates called AccessMonitor (Evanston, IL) used extensively by many pharmaceutical companies. Clinical decisions of 58,647 to 72,114 physicians were statistically analyzed using prescription data from IMS Health (Danbury, CT) in three critical areas: an innovative drug for type 2 diabetes (sitagliptin), an older diabetes drug with a new Food and Drug Administration–required black box warning for cardiovascular safety (rosiglitazone), and a combination lipid therapy that had reported negative outcomes in a clinical trial (simvastatin+ezetimbe). For the uptake of the new diabetes agent, the authors found that physicians with very low access to representatives had the lowest adoption of this new therapy and took 1.4 and 4.6 times longer to adopt than physicians in the low- and medium-access restriction categories, respectively. In responding to the black box warning for rosiglitazone, the authors found that physicians with very low access were 4.0 times slower to reduce their use of this treatment than those with low access. Likewise, there was significantly less response in terms of changing prescribing to the negative news with the lipid therapy for physicians in more access-restricted offices. Overall, cardiologists were the most responsive to information changes relative to primary care physicians. These findings emphasize that limiting access to pharmaceutical representatives can have the unintended effect of reducing appropriate responses to negative information about drugs just as much as responses to positive information about innovative drugs.

Pharmaceutical sales representative access to physicians has been getting increasingly difficult. Sales representatives have been the main channel for transmitting marketing information through “detailing” to physicians for the past 50 years,1 accounting for 60% of all sales and marketing expenditures.2 Currently, all marketing information by sales representatives to physicians are regulated and enforced by the Food and Drug Administration (FDA) under the newly named Office of Prescription Drug Promotion to ensure that all information delivered is consistent with FDA-approved product labeling. In 2010, about 11% of American physicians have “severe” or “no-see” (meaning sales representatives cannot see physicians in their office) pharmaceutical representative access limits and up to another 34% have some restrictions.3 Significant geographic variation in access limits exist around the United States.3 There has been progressive tightening of access-limit policies resulting in an increase in physicians having severe or no-see ratings over the past few years.3 These increasing access limits and outright bans of contacts with industry representatives have occurred as more academic articles, policies at medical schools, and physician anti-detailing campaigns call for such restrictions.4–16 Legal arguments have advocated limiting First Amendment protections of pharmaceutical commercial speech to physicians and banning the selling of physician prescription information.17,18 Challenging this research are arguments that limiting commercial speech and more tightly regulating academic-industry research relationships will do more harm than good.19–23 Empirical evidence exists showing that sales and marketing are associated with an increase in the adoption of new clinical evidence that is beneficial to patients.24 The industry has argued against sales force access limits and revised its code of conduct to correct for past abuses and to prevent future infractions.25,26 The US Supreme Court recently ruled 6–3 a Vermont law as unconstitutionally restricting the distribution of prescriber information to pharmaceutical companies and sales representatives as a way to limit detailing to physicians.27 The Court majority opinion also noted: “If pharmaceutical marketing affects treatment decisions, it does so because doctors find it persuasive.… Indeed the record demonstrates that some Vermont doctors view targeted detailing based on prescriber-identifying information as “very helpful” because it allows detailers to shape their messages to each doctor’s practice.”27 Critics have noted that the decision is a setback for efforts to curtail detailing in the pursuit of protecting patient health, reducing unnecessary expenditures on new drugs where therapeutically equivalent generics exist, and protecting physician and patient privacy.28,29

Regardless of position, no previous study has used measurements of sales representative access limits at the physician level with actual prescription behavior and analyzed on a scale large enough to measure the effect of access limits on physician clinical prescription decisions to new medical information. Specifically, this research analyzed whether physicians in more access-limited offices respond to different types of new medical information by changing their product prescription market share in a smaller and longer fashion than counterparts in more open-access offices, while controlling for other factors that affect their decision.

We conducted modeling work on physician prescription data applying previous literature on the analysis of medical information markets to three recent and well-known product cases.30,31 We estimated individual physician clinical responses, measured by prescription extent and speed of product prescription share changes, to positive and negative new medical information in three medical events: (1) launch in October 2006 of sitagliptin, a first-in-class selective dipeptidyl peptidase-4 inhibitor drug and recognized new product to treat type 2 diabetes32–37; (2) the release of the Effect of Combination Ezetimibe and High-Dose Simvastatin vs. Simvastatin Alone on the Atherosclerotic Process in Subjects with Heterozygous Familial Hypercholesterolemia (ENHANCE) negative clinical trial outcome results in January 2008 involving the combination drug simvastatin/ezetimbe, used to lower low-density lipoprotein cholesterol38; and (3) the effect of a critical New England Journal of Medicine (NEJM) study published in May 2007 on rosiglitazone, used to treat type 2 diabetes, and subsequent FDA-imposed black-box warning in August 2007, due to a higher risk of heart failure in patients taking the drug.39,40

These product cases were selected based on these factors: (1) events came after the development of physician-level access limit measurements, (2) events represented new medical information to the market that could affect physician clinical prescription decisions, (3) events represented negative and positive new medical information to determine whether physician clinical prescription decisions varied by type of new information, (4) availability of other physician-level measures consistent with the medical market literature needed for proper model estimation of access limit effects, and (5) product cases dealt with important and costly chronic conditions.41,42

Individual prescriber data were gathered from higher-volume prescribing physicians comprising 80% of all prescriptions dispensed per product event available through IMS Health (Danbury, CT), resulting in 58,647 to 72,114 physicians covered per case. Since the cases involved different products in different therapy classes, the list of physicians comprising 80% of prescriptions varied. IMS Health collects prescriber level data that accounts for approximately 70% of all dispensed prescriptions through pharmacies and mail order outlets in the United States. This research also leveraged a unique proprietary database called AccessMonitor from the global pharmaceutical sales and marketing consulting firm ZS Associates (Evanston, IL) that measured sales representative access limits to physicians. A major departure of this research from previous studies is our use of a data-driven physician-level sales representative access limit metric.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

Model Design

Seminal research work exists on the role of information in medical markets and the response by physicians and patients.30,31 Physicians make multiple complex clinical decisions daily for their patients where medical information is costly and time consuming to acquire. Optimal health decisions for patients are the result of physicians weighing the value and relevance of medical and marketing information from a variety of channels on top of their own training and experiential knowledge. Limits placed on the flow of relevant medical information from any channel will affect physician decision-making, unless that physician can easily replace that lost information through another channel. Drug sales representatives provide timely and convenient regulated clinical information used by physicians, nurses, and office staff.

The key question analyzed was: Did physician access limits restrict the flow of new medical information enough to affect both the speed and extent of physician clinical decisions as measured by their prescribing decisions? Measuring these relationships must account for other factors that can influence how physicians used medical and marketing information through other channels to mitigate the impact from sales representative access limits.

We approached this study by making the following research decisions. Major new medical information events were analyzed to determine their affect on physician clinical prescription decisions such as a new product launch, new drug clinical trial information, or FDA announcements such as a black-box warning. Positive and negative news items were selected and statistically tested to determine whether physician clinical prescription responses varied by type of new medical information. We used IMS Health data analyzing actual dispensed prescriptions to determine what physicians actually did in response to varying access limits. Physician clinical prescription decisions were measured on their extent and speed of product prescription market share adoption vs decline relative to variations in access limits. Analysis was done to control for physician and nonphysician factors that affected the variation of access limits. For example, physicians who are high-volume prescribers are the focus of many sales representative efforts, creating greater congestion in these offices while affecting physician time with patients. A possible response by physicians would be to erect greater access limits to counter congestion costs and protect their time devoted to patient care.

Data

The AccessMonitor database was created by the global sales and marketing consulting firm ZS Associates in 2006.3 Twice a year in 6-month semester (S) intervals, participating drug companies send data to ZS Associates listing the physicians sales representatives are expected to call, the frequency of planned expected calls, and the number of delivered calls to each physician.3 Calculation of sales representative calls delivered relative to plan is then aggregated across all sales representatives per physician.3 This comparison determines the industry-wide access limits by physician.3 Scores are then assessed per physician noting the degree of overall sales representative call achievement relative to plan ranging from 1 (most access restricted) to 10 (least access restricted).3 Measurements of access limits for this analysis were based on access_bucket definitions by physician using AccessMonitor ratings grouped by very low (1–3), low (4 or 5), medium (6 or 7), and high (8–10). Access scores can be aggregated by specialty and geography.3 ZS Associates does not release the members of its client database for legal reasons.3

Numerous indicators suggest that the AccessMonitor database is valid and representative of variations in access limits by physician for research analysis. Client pharmaceutical companies pay fees to use this information to make significant sales and marketing decisions for an efficient and effective means of distributing medical and marketing information to physicians and their patients.3 AccessMonitor covers a significant number of prescribers and percent of industry sales representatives, with the former metric rising from 240,000 to consistently over the 300,000 level, and the latter metric rising from 21% to 54%, respectively, from S1 2008 to S1 2011.3 AccessMonitor covers more than 200 unique sales teams, including emerging pharmaceutical companies, and provides excellent coverage across a wide range of physician specialties.3 Access trends reported in AccessMonitor are also affirmed elsewhere.43

The Figure depicts S1 2011 data from AccessMonitor revealing sales representative access challenges across the United States by metropolitan statistical area (MSA). Access challenges are most severe in New England, the Upper Midwest, parts of the Mountain West, and the West Coast.3 The rest of the country has relatively greater access, although pockets of severe access limits exist. A comparison of earlier AccessMonitor MSA access maps to S1 2011 reveals a significant growth in access limits, as measured by the proportion of high-access prescribers falling from approximately 74% in S1 2008 to about 55% in S1 2011.3Table I reveals that significant access variation exists among physician specialties, with nephrology (68%), oncology (57%), and cardiology (54%) being the most limited, and urology (21%), OB/GYN (21%), and allergy (28%) being the least limited, when adding up the percentage of severe and moderate-access limit data (noted in parentheses).3Table II provides an explanation of all variables used in the statistical model design, data sources, description of the trustworthiness of all pharmaceutical databases, ability to replicate study results, and how data limitations were addressed. Table III lists data descriptive statistics and physician count by variable.

image

Figure FIGURE.  Sales representative access limits for the United States by Metropolitan Statistical Area (MSA) for semester (S1) 2011. The map comes from the S1 2011 AccessMonitor report produced by ZS Associates. Among MSAs noted to have severe access restrictions, 20% or more of prescribers within such locations have severe access restrictions according to S1 2011 data contained in AccessMonitor from ZS Associates. Access measured at the MSA level as percentage of prescribers rated with an AccessMonitor “no-see rating” between 1 and 5. Non-MSA geographic areas are noted in white. Variations also exist in physician access by specialty expressed as percentage of prescribers for S1 2011, ranked by “Severe Access Restrictions” and organized by little or no access, moderate access, and severe access restrictions by AccessMonitor no-see ratings of 8 to 10, 4 to 7, and 1 to 3, respectively. Permission was received from ZS Associates to reproduce this Figure.

Download figure to PowerPoint

Table I.    Sales Representative Access Limits for the United States by Physician Specialty for S1 2011a
Physician SpecialtyFew or No Access Limits, %Moderate Access Limits, %Severe Access Limits, %
  1. aPrimary care physicians are defined here as family practice, general practice, doctor of osteopathy, and internal medicine. All prescribers represent the overall average across all physicians in the AccessMonitor database per access limit category for semester 1 (S1) 2011.

Primary care533314
Nephrology325513
All prescribers543411
Oncology434710
Neurology523810
Psychiatry523810
Cardiology46459
Endocrinology62326
Pediatrics62335
Allergy73244
Dermatology54433
Gastroenterology69292
Urology79192
OB/GYN79201
Table II.    Empirical Model Variable Names, Descriptions, and Data Sourcesa
Variable (Physician-Level Data)Description
  1. a1. All patient-identified information on each prescription has been eliminated to remain in compliance with all federal protections of personal information HIPAA Privacy Rule regulations under the Health Insurance Portability and Accountability Act of 1996. 2. While all the pharmaceutical data for this study came from proprietary databases such as IMS Health, ZS Associates, and internal pharmaceutical company information, the empirical results can be replicated if researchers made similar contractual and financial arrangements as we did to secure all data elements. 3. The hypothesized model design includes the following factors beyond sales representative access limits affecting physician clinical prescription decisions: (1) physician characteristics like specialty, age, and gender, (2) physician experience and knowledge treating patients in a particular disease category, (3) physician access to other drug promotional efforts and medical educational channels, (4) patient needs that could influence physician response, (5) the size of the group practice setting where physicians operate, providing opportunities to interact with colleagues for medical information and resources to use digital technologies to acquire medical information, and (6) the external environment and institutions affecting physician decision-making, such as managed care and healthcare system rules governing physician clinical prescription decisions. 4. N/A refers to the lack of a gender designation provided for a physician in the AMA database. About 10% to 11% of physicians in our analysis did not specify their sex (noted as N/A). We ran all the models against only those physicians who specified their sex and determined the cost of dropping so much data was not worth the cost in terms of losing model explanatory power. 5. Managed care control was defined as a plan’s ability to move market share. Plans use various means to exert control such as step therapy, prior authorization, quantity limits, and tier exclusivity. A view across drug plans revealed differences in plan control and the degree to which they are generic forcing over branded drugs. Each physician NRx was tied to a plan which in turn generated a metric on the ability of that plan to move product market share relative to the national average. The statin and proton pump inhibitor product markets were used through a pharmaceutical firm because of data availability and plan coverage. Then each plan control measure per NRx was rolled up by physician to generate an average managed care control metric for that physician. 6. Primary care physicians (PCPs) were defined as family practice, general practice, and doctor of osteopathy physicians. While we recognize internal medicine physicians generally comprise the fourth group of PCPs according to traditional academic studies, industry research of physician prescribing showed that internal medicine sometimes behaved like specialists depending on the therapy class. 7. The specialty designation per physician was that provided in the AMA database. Physicians may have multiple specialties, though the analysis used only the one specialty provided in the AMA database. 8. Data limitations were partially addressed by variables already collected. Competing product promotion data on all drugs per class per event at the physician level did not exist for the large number of physicians analyzed. Thus, we chose to analyze physicians comprising the top 80% of prescriptions per event, ensuring each physician received significant sales and marketing from all competing drug companies. Direct measurements on physicians using newer digital medical information technologies were not available. Measurements on physician age and gender, specialty, market volume of new prescriptions, and group practice size likely picked up physician preferences and uses of these technologies. Direct knowledge of physician-level training and attitudes towards each product was unavailable. Measures collected like physician specialty, graduation date, gender, and market volume prescription likely captured these influences. Managed care patient co-pay data at the physician level were unavailable. Effects of localized differential managed care contracting efforts between health insurance and pharmaceutical companies were captured through estimating local geography effects by MSA. Variations in the distribution of patient disease severity by physician were captured by specialty of the physician, assuming specialists having disproportionately sicker patients than primary care physicians. There is evidence that physicians may harbor lack of trust and have objectivity suspicions of pharmaceutical industry-supplied materials.44 Differences in these attitudes were likely partially captured by existing physician demographic and specialty measures. There was also no data available to distinguish between physicians in group vs academic practices. The urban vs rural and group practice size variables may have captured some of this effect. Physicians in urban vs rural practices may also differ in their group practice size and the level of promotion received. Also, physicians in group practices may have received or drawn on more information from their peers, compared to a rural physician who has no or fewer peers, especially in a solo practice. There may exist a bias in the level of information sources available across physicians. Thus the relative role a sales representative source of information may play that differs by physician was a hypothesized effect we wanted to account for in the model design.

Access_Bucket (physician access limits to sales representatives by grouping)Four groupings based on ZS AccessMonitor rating. “0.N/A.” means prescriber is not in the AccessMonitor data and they were not used in the model. Bottom 3 ratings (1–3) are very low, next 2 are low (4–5), next 2 are medium (6–7), and the top 3 are high (8–10)
Access_ResidResidual measure from the ordered multinomial logistic model for Access_Bucket for that product case used to test for Access endogeneity
Adoption_ExtentSitagliptin new-product launch case (1): 80% of peak new prescription (NRx) share in the first 12 months of launch. Simvastatin/ezetimbe clinical trial case (2) and rosiglitazone New England Journal of Medicine (NEJM)/black-box warning case (3): the difference between preclinical trial or black-box warning share using a 3-month average before the event and the lowest share in the 12 months subsequent to the event. Source: IMS Health Xponent
Adoption_SpeedSitagliptin new product launch case (1): months taken for a 1%-point increase in share (computed as number of months taken to reach 80% of peak share in the first 12 months of launch)/(80% of peak share). Simvastatin/ezetimibe clinical trial case (2) and rosiglitazone NEJM/black-box warning case (3): months taken for a 1% point decrease in share computed as (number of months taken to reach the lowest share after decline)/(difference between pre-event share and the lowest share subsequent to the event). Source: IMS Health Xponent
Age_GradBased on number of years since graduation from medical school, American Medical Association (AMA) data
AMA_GenderMale, Female, or N/A (not specified) for each physician, from AMA data
Group_Practice_SizeNumber of physicians in affiliated group practice for that physician. Sources: IMS Health Healthcare Organization Services (HCOS) data and through internal adjustments made by a pharmaceutical company
MC Control_Bucket (managed care plan control by grouping)Five groupings based on cross-market average plan control for each physician. Control for each plan is based on how much the plan can change individual brand market share away from the national average. A high-controlling plan will have higher changes. Plan control is averaged for each physician. The 5 groupings are: very high, high, medium, low, very low. Sources: IMS Health Xponent and Xponent PlanTrak using data from February 2008 to July 2008
MSA_VectorVector of 1-0 dummy variables to specify location of the physician by metropolitan statistical area (MSA) with the residual category being a “Rural” designation. Source: U.S. Bureau of Census designations
Specialty_Group6 Physician specialty groupings based on IMS specialty codes: Allergists (ALG) & Pulmonologists (PUD), Cardiologists (CARD), Diabetologists (DIA) & Endocrinologists (ENDO), Internal Medicine (IM), Primary Care (PCP), and All Others (OTH). Sources: IMS Health Xponent and AMA data
Urban_RuralIf a physician prescriber lives in a MSA, is flagged as “Urban” otherwise “Rural.” Source: US Bureau of Census designations
Volume_GroupGrouping based on physician prescribing market deciles. Top 4 deciles are High, next 2 are Medium, next 2 are Low, and bottom 2 are Very Low. Only the top 8 deciles (top 3 volume groups of high, medium, and low) were used in the model. Source: IMS Health Xponent
Table III.    Sitagliptin (1), Simvastatin/Ezetimbe (2), and Rosiglitazone (3) Cases: Descriptive Statistics and Physician Count per Variable per Druga
 MinMaxMeanSD
  1. aThe total number of physicians per drug case (N): (1) 65,131 sitagliptin, (2) 72,161 simvastatin/ezetimbe, and (3) 58,651 rosiglitazone. The differences in sample size reflect differences in the set of physicians that are the top 8 decile prescribers (physicians comprising 80% of all dispense prescriptions) per product case. There is likely some small inaccuracy in the AMA data on the measure Age_Grad for the upper-age categories that are probably unrealistic, especially in the categories from 80 to 99 years, and as noted here in this table, a maximum age of 99. The AMA data contains the following breakdowns for the upper-age brackets as a percent of physicians in our sample: 0.5% for 90 to 99, 1.0% for 80 to 89, and 2.5% for 70 to 79 years. The number of physicians listed as 99 for Age_Grad in each product case are as follows: five for sitagliptin, four for simvastatin/ezetimbe, and five for rosiglitazone. Rather than arbitrarily decide an age to cut off the data from analysis, we recognize the issue exists. Given the very low number and percentage these data elements exist in the sample, it is likely not substantially affecting the Age_Grad estimates.

(1) Age_Grad0.09927.5814.64
(1) Group_Practice_Size0.09084176.36569.22
(1) Adoption_Extent0.00.80.06380.071
(1) Adoption_Speed0.0225.534.8285.546
(2) Age_Grad0.09927.5614.14
(2) Group_Practice_Size0.09084187.6605.1
(2) Adoption_Extent0.00.80.11470.081
(2) Adoption_Speed0.02388.150.7043.119
(3) Age_Grad0.09927.6814.57
(3) Group_Practice_Size0.09084175.97565.59
(3) Adoption_Extent0.01.00.07710.055
(3) Adoption_Speed0.01261.480.06331.445
 Physician Count per Variable
SitagliptinSimvastatin/EzetimbeRosiglitazone
Access bucket
 Very low178917431479
 Low351236472964
 Medium998411,3088716
 High49,84655,46345,492
Sex
 Female12,40413,17410,962
 Male45,79151,96841,424
 N/A693670196265
Geography
 Rural11,12610,2499625
 Urban54,00561,91249,026
Specialty
 ALG/PUD370404336
 CARD115186691043
 DIA/ENDO227412342122
 IM26,49827,91724,000
 OTH203019801730
 PCP32,80831,95729,420
Prescribing volume
 High15,77522,98615,269
 Medium19,86621,15218,325
 Low29,49028,02325,057
Managed care control
 Very low136629971223
 Low677686106130
 Medium18,88521,73517,050
 High20,21522,02218,154
 Very high17,88916,79716,094

Statistical Model Design

The statistical analysis for adoption_extent and adoption_speed per product case was done using a traditional two-stage statistical approach: (1) estimating the determinants of physician access limits, and (2) using the estimate for physician access limits in the first stage in the estimation of the affect of access limits on adoption_extent and adoption_speed in the second stage.45,46 The notes at the end of Table IV through Table VI provide details on how the estimations were conducted. The reason for this approach was to account for physician and/or institutional decisions that determined access limits in response to the effects of increasing sales representative effort.

Table IV.    Sitagliptin (1), Simvastatin/Ezetimbe (2), and Rosiglitazone (3) Cases: Adoption_Extent Regression Model Results per Drug Casea
VariableClass/ValueDependent Variable=Adoption_Extent
SitagliptinSimvastatin/EzetimbeRosiglitazone
  1. aValues represent coefficient estimates. Actual standard errors are available upon request. ProbChiSq: b<.0001. All blank rows indicate residual category. Physicians (MDs) with a market decile below 3, without an AccessMonitor rating, or with IMSIDs that start with 95 or higher are excluded. SAS 9.1 PROC LOGISTIC was used to estimate the adoption_extent models using a binary logistic with a weighted vector to reduce heteroskedasticity regression approach, with the relative contribution estimates in Table VI coming from these model estimates. All model results are available upon request. Global tests for model fit show each regression equation having a ProbChiSq<.0001. Access_Resid comes from the first stage model to estimate variations in Access_Bucket for each drug case across a set of explanatory variables to account for potential endogeneity bias. While the access_resid variable is still statistically significant, it has a very low relative contribution measure as seen from Table VI, meaning the approach taken to account for this bias as modeled in this table was mostly addressed. The variables used were AMA_Gender, Age_Grad, Specialty, Group_Practice_Size, MC Control, Volume, and MSA_Vector. The number of MSA_Vector areas employed per drug case: (1) 367; (2) 372; (3) 367 with an overall sample size of physicians per case of (1) 65,088; (2) 72,114; and (3) 58,647.

Intercept −2.659b−1.921b−2.716b
Access_Resid −0.291b−0.161b0.0603b
AMA_GenderFemale−0.0258b0.00245b0.0254b
AMA_GenderMale−0.0605b−0.0332b0.0142b
AMA_GenderN/A   
SpecialtyALG/PUD−0.0717b−0.0783b0.0986b
SpecialtyCARD0.391b0.243b0.204b
SpecialtyDIA/ENDO0.395b0.219b−0.503b
SpecialtyIM−0.153b−0.213b0.00335b
SpecialtyOTH−0.274b−0.0242b0.109b
SpecialtyPCP   
VolumeHigh−0.166b−0.105b−0.0986b
VolumeLow0.144b0.0957b0.101b
VolumeMedium   
AccessVery low−0.298b−0.126b0.00748b
AccessLow−0.081b0.0293b−0.0429b
AccessMedium0.140b0.0519b0.0154b
AccessHigh   
MC ControlVery low−0.0104b0.046b−0.144b
MC ControlLow0.0623b0.0219b0.00175b
MC ControlMedium0.0981b0.0247b0.0161b
MC ControlHigh0.0194b−0.0273b0.0228b
MC ControlVery high   
Group_Practice_Size −0.00003b−0.00002b−0.00005b
Age_Grad −0.00055b−0.00077b0.00179b
Specialty*AccessALG/PUD – Very Low−0.144b−0.108b−0.0693b
Specialty*AccessALG/PUD – Low−0.0347b−0.00246b−0.017b
Specialty*AccessALG/PUD – Medium0.092b0.0516b0.101b
Specialty*AccessALG/PUD – High   
Specialty*AccessCARD – Very low0.295b0.204b0.127b
Specialty*AccessCARD – Low0.153b−0.0518b−0.0151b
Specialty*AccessCARD – Medium−0.263b−0.0694b−0.134b
Specialty*AccessCARD – High   
Specialty*AccessDIA/ENDO – Very low0.243b0.0639b0.159b
Specialty*AccessDIA/ENDO – Low0.287b0.121b−0.0812b
Specialty*AccessDIA/ENDO – Medium−0.152b−0.0634b−0.0317b
Specialty*AccessDIA/ENDO – High   
Specialty*AccessIM – Very low−0.152b−0.0699b−0.0788b
Specialty*AccessIM – Low−0.0543b−0.0762b0.0171b
Specialty*AccessIM – Medium0.0714b0.0478b0.0254b
Specialty*AccessIM – High   
Specialty*AccessOTH – Very low−0.107b−0.0357b−0.0791b
Specialty*AccessOTH – Low−0.215b0.0607b0.042b
Specialty*AccessOTH – Medium0.170b0.00893b0.0462b
Specialty*AccessOTH – High   
Specialty*AccessPCP – Very low   
Specialty*AccessPCP – Low   
Specialty*AccessPCP – Medium   
Specialty*AccessPCP – High   
Urban_RuralRURAL0.0542d0.00571b−0.0148b
Urban_RuralURBAN   
N (number of physicians) 65,08872,11458,647
Table V.    Sitagliptin (1), Simvastatin/Ezetimbe (2), and Rosiglitazone (3) Cases: Adoption_Speed Regression Model Results per Drug Casea
VariableClass/ValueDependent Variable=Adoption_Speed
SitagliptinSimvastatin/EzetimbeRosiglitazone
  1. aValues represent coefficient estimates. Actual standard errors are available upon request. ProbChiSq: b<.0001; c<.01; d<.05. Weibull distribution was used given rejection of normality of the error structure as is customary to do for parametric duration models per regression equation. Therefore each regression model was estimated using a parametric duration model using a Weibull distribution with a weighted vector to reduce heteroskedasticity, with the relative contribution estimates in Table VI coming from these model estimates. All model results are available upon request. All blank rows indicate residual category. Physicians (MDs) with a market decile below 3, without an AccessMonitor rating, or with IMSIDs that start with 95 or higher are excluded. SAS 9.1 PROC LIFEREG was used to estimate the adoption_speed regression models. Global tests for model fit show each regression equation having a ProbChiSq<.0001. Access_Resid comes from the first stage model to estimate variations in Access_Bucket for each drug case across a set of explanatory variables to account for potential endogeneity bias. While the access_resid variable is still statistically significant, it has a very low relative contribution measure as seen from Table VI, meaning the approach taken to account for this bias as modeled in Table V was mostly addressed.

Intercept 1.345b−1.921b−0.461b
Access_Resid 0.957b0.1360.040
AMA_GenderFemale0.335c−0.043−0.044c
AMA_GenderMale0.333b0.0160.020
AMA_GenderN/A   
SpecialtyALG/PUD−0.024−0.022−0.159b
SpecialtyCARD−0.393c−0.171b−0.141b
SpecialtyDIA/ENDO−1.115b−0.196d0.579b
SpecialtyIM−0.184b0.0020.021c
SpecialtyOTH0.219−0.132−0.072b
SpecialtyPCP   
VolumeHigh−0.247b0.237b0.366b
VolumeLow0.355b−0.299b−0.318b
VolumeMedium   
AccessVery low1.649b0.0070.277b
AccessLow1.203b−0.038−0.090c
AccessMedium0.364c−0.001−0.070c
AccessHigh   
MC ControlVery low−0.392d0.1120.006
MC ControlLow−0.557b0.0180.047b
MC ControlMedium−0.649b−0.071d0.019d
MC ControlHigh−0.416b−0.068d0.006
MC ControlVery high   
Group_Practice_Size 0.00010.0000.000b
Age_Grad 0.00260.00010.000
Urban_RuralRURAL0.111**−0.046−0.057b
Urban_RuralURBAN   
N (number of physicians) 65,08872,11458,647
Table VI.    Sitagliptin (1), Simvastatin/ezetimbe (2), and Rosiglitazone (3) Cases – Relative Contribution Analysis from the Adoption_Extent and Adoption Speed Regression Model Results per Drug Casea
EffectDependent Variable=Adoption_Extent Dependent Variable=Adoption Speed
SitagliptinSimvastatin/ezetimbeRosiglitazone
  1. aProbChiSq: b<.0001; c<.01; d<.05. Physicians (MDs) with a market decile below 3, without an AccessMonitor rating, or with IMSIDs that start with 95 or higher are excluded. Not applicable (NA) indicates that the interaction terms for Specialty*Access were not specified in the adoption_speed regression models. SAS 9.1 PROC LOGISTIC was used to estimate the adoption_extent regression models and PROC LIFEREG for the adoption_speed regression models. The relative contribution estimates came from each set of regression results. All model results are available upon request. Italics distinguish between the upper number for each effect representing adoption_extent vs the lower number representing adoption_speed.

Access_Resid1.7%b1.9%b0.3%b
4.6%b0.4%0.0%
Access1.7%b0.3%b0.1%b
11.1%b0.2%1.0%b
Specialty*Access14.3%b4.7%b1.8%b
NANANA
Specialty17.5%b22.0%b17.1%b
24.1%b6.4%b10.5%b
Volume31.8%b47.1%b48.2%b
26.3%b86.7%b85.9%b
MC Control24.4%b19.3%b21.4%b
28.6%b4.3%c0.3%c
Group_Practice_Size0.7%b1.1%b5.1%b
0.5%0.4%0.5%b
Age_Grad0.1%b0.5%b3.3%b
0.5%0.0%0.0%
AMA_Gender3.4%b3.0%b1.5%b
3.4%c1.0%1.0%b
Urban_Rural4.4%b0.1%b1.1%b
0.9%d0.5%0.7%b
Physicians, No.65,08872,11458,647

Results

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

Determinants of Sales Representative Access Limits

Table IV provides an explanation of the access_bucket regression equations per case. All key model expectations were verified. The strongest influence on access limits per case was MSA_Vector, with a relative influence of 57.1% to 64.7%. This affirmed the depiction from the Figure that access limits were a locally derived phenomenon.3,43 Physician market volume (Volume) of new prescriptions had the second greatest influence, with a 25.0% relative influence per case, with higher-volume prescribing physicians associated with greater access limits. This affirmed expectations of increasing levels of details to higher-prescribing physicians, thereby causing physicians to make decisions limiting their access to sales representatives vs time spent with patients. Greater managed care control (MC Control) was associated with greater access restrictions, with a relative influence ranging from 5.1% to 15.9%. This result was consistent with the objective of managed care plans to influence the creation of higher-access limits as a way to reduce branded drug prescribing and expenditures. The remaining variables of physician age, sex, and specialty were all statistically significant but relatively low in importance, while group practice size was insignificant in two cases.

Sales Representative Access Limits and Physician Prescribing Behavior

Table IV lists the empirical results for the adoption_extent models. All the empirical results showed increasing sales representative access limits significantly affected physician responses to new medical and marketing information consistent with expectations. All the variables and model fit in each case were highly significant (P<.0001).

Results for the sitagliptin new-product launch case revealed adoption_extent prescribing market share change for the very low-access physicians had a log-likelihood estimate 3.7 times greater than the low access category. Physicians in very low-access offices had the smallest adoption of sitagliptin, followed by physicians in low-access offices, relative to high-access physicians. The interaction of physician specialty and access limits (specialty*access) showed that effects from access limits were less on knowledgeable specialists such as cardiologists (CARDs) and diabetologists (DIAs), and endocrinologists (ENDOs) relative to primary care physicians (PCPs). CARDs and DIAs/ENDOs in very low and low-access offices had greater adoption (positive coefficients) of the sitagliptin than PCPs. Physicians with high-market volume of new prescriptions (Volume) had lower sitagliptin adoption than low-volume physicians compared with the medium-volume category. Physicians facing very low-managed care control had lower adoption_extent than physicians in low to very high control. Physician-specialty CARDs and DIA/ENDOs had larger adoption than PCPs. Larger group practice size and higher age of physician since graduation had a depressing effect on adoption_extent. Greater adoption was shown by physicians in rural areas relative to urban MSAs.

Results for the simvastatin/ezetimbe clinical trial case revealed similar effects to sitagliptin for the results from access, volume, physician specialty, interaction effects of specialty*access, age, group practice size, and rural location. Physicians facing medium to low managed care control generated greater product declines than physicians facing high and very high control. Smaller product market share declines occurred with older physicians and those in larger group practices.

Results for the rosiglitazone NEJM/black-box warning case revealed physicians in low-access offices had a lower decline than physicians in the other categories. CARDs and IMs showed larger share declines while DIA/ENDOs revealed smaller declines relative to PCPs. The effects of the interaction terms specialty*access showed CARDs and DIA/ENDOs in very low offices having higher switching changes than PCPs. Physicians with high-market volume, in larger-group practices, and in rural areas were associated with smaller declines, while older physicians experienced larger share declines.

Table V provides the adoption_speed product market share model results. Key findings in the sitagliptin case showed physicians in very low-access offices took 1.4 and 4.6 times longer to adopt than physicians in the low- and medium-access categories, respectively. CARDs, DIA/ENDOs, and IMs took less time to adopt relative to PCPs as did physicians in the high-volume category. Physicians facing very low to high managed care control took less time to adopt than those in the very high category.

Key findings in the rosiglitazone case showed physicians in very low-access offices with a speed change four times longer than physicians in low-access offices. CARDs took less time, but DIA/ENDOs and IMs took longer to switch than PCPs. Physicians facing very low to high managed care control took more time to switch relative to those in the very high control category, although statistical significance was weaker. The adoption_speed model broke down for the simvastatin/ezetimbe case as explained in the next section.

Table VI lists relative influence of the key variables in the adoption_extent and adoption_speed models per case that affirmed model expectations. Factors with the greatest relative contribution in explaining adoption_extent changes were physician volume (31.8% to 48.2%), managed care (19.3% to 24.4%), specialty (17.1% to 22.0%), and access plus the interaction of specialty*access (16.0% to 0.5%). The same general pattern was true for the adoption_speed models, less the specialty*access effects, although the effects from volume-dominated contribution relative to all other factors. Access plus specialty*access restriction effects mattered more for the new product sitagliptin case (16.0%) than the negative information cases of simvastatin/ezetimbe (5.0%) and rosiglitazone (1.9%) in the adoption_extent models. Lastly, while specialty*access interaction effects were significant in the adoption_extent models, this was not the case for the adoption_speed models.

Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

Access Limits and Clinical Practice

This study estimated the effects of increasing sales representative access limits to physicians on their response to new medical information. The statistical analysis produced robust and significant estimates on access limit effects on physician prescription decision-making. Physicians in more access-limited offices adopted the first-in-class new drug much less and at a slower speed than their counterpart physicians in more open-access offices. The relative contribution results in Table VI showed that sales representatives had greater effects on physicians during the drug launch case where they represent a key source of new medical information. Regarding negative medical news cases, physicians derive this information from other sources such as medical journals and conferences, clinical practice, FDA announcements, and discussions with colleagues.

The interpretation of the model results for sitagliptin offers different explanations. We found that physicians facing increasing access limits that restricted the flow of medical information reduced their market share adoption patterns relative to counterparts in more open-access offices. The challenge here for physicians was knowing whether a new drug would produce expected clinical benefits at launch. Confidence in a new drug is acquired over time from experience in clinical practice and follow-up studies that are reported in the medical literature. A reduction in the extent and speed of adoption market share for sitagliptin in the first 6 months may be a wise decision by all physicians as greater knowledge is acquired in actual practice in a much larger patient population than seen in clinical trials. The example of the COX-II inhibitor rofecoxib withdrawal from the market in 2004 comes to mind.47 Physician-delayed response would have even more validity for a first-in-class drug where no prior actual clinical practice is available to guide physician decisions.

Our empirical findings on sitagliptin, however, contradict the view that all physicians found delay of adoption to be a preferred strategy. We found that the effects of increasing access limits on sitagliptin adoption significantly varied by physician specialty, controlling for other factors that affect adoption patterns. The empirical results found specialists such as CARDs and ENDOs/DIAs demonstrating greater and faster adoption market share patterns than PCPs. We infer from this result that physicians having greater insight about the relative benefits and costs of sitagliptin and/or treating more chronically ill patients in this disease area saw the benefits of faster and greater adoption vs PCPs. Thus, if delaying the adoption of sitagliptin was the preferred strategy in an effort to acquire more clinical information as it became known, how do we explain empirical variations in adoption patterns by physician specialty? The more informed physicians as inferred by specialty adopted sitagliptin in greater amounts and at a faster rate than PCPs. Increasing sales representative access limits restrict the flow of medical and marketing information needed by physicians to make better-informed clinical prescription decisions for their patients.

The effects of increasing access limits were weaker, although statistically significant, when physicians responded to negative medical information to the clinical trial announcement and NEJM/black-box warning cases. Increasing access limits decreased the competition of medical ideas from different sales representatives, thus preventing physicians from seeing a more complete picture on the pros and cons of various drug treatment options. Access limits were statistically significant in affecting the speed of physician response in the new-product launch and NEJM/FDA black-box warning cases but not in the negative clinical trial information case. We surmise that the failure of the model to produce significant access effects for the simvastatin/ezetimbe case in the adoption_speed model was due to the 2-year delay of Merck announcing the Effect of Combination Ezetimibe and High-Dose Simvastatin vs. Simvastatin Alone on the Atherosclerotic Process in Subjects with Heterozygous Familial Hypercholesterolemia (ENHANCE) clinical trial results, with physicians already expecting negative news prior to the January 2008 announcement. The longer time response for DIA/ENDOs in the rosiglitazone adoption_speed case may be the result of these physicians treating the most severe type 2 diabetic patients. Thus, even with the NEJM article and subsequent black-box warning, these specialists may have seen benefits of continued use that outweighed the risks for these specific patients. The effects of increasing access limits worked to interact with physician specialty, with specialists such as CARDs and DIA/ENDOs being less affected compared with PCPs. This suggests that PCPs relied more heavily on sales representatives as a source for new medical information than specialists in treating patients. This result has significant clinical implications given that PCPs represent the first line of medical treatment for patients in our health care system.

Conclusions

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

This research investigated and revealed the effects of increasing sales representative access limits on physician prescription responses to new medical information in ways not analyzed in previous studies. Supporting conclusions were new data on access limits at the physician level, large physician sample sizes, actual physician prescriptions, managed care control at the physician level, and other relevant measures. Quantitative research methods were employed to determine the proper effect of access restrictions affecting both extent and speed of physician product prescription market share change.

Directions for Future Research

More needs to be understood regarding the potential unintended effects of sales representative access limits to physicians, the consequences these limits can have on physician clinical decisions, and patient health. Policies that promote physician ignorance of new medical information resulting from access limits runs counter to protecting patient health. This study did not have measures of patient health. However, results for the negative news cases provide potential insights. PCPs in restricted-access environments had slower and lower prescription product share reductions than counterparts in more open access offices. This means patients of these physicians were put through greater potential risks that were less prevalent for patients being treated by more open-access limit physicians. The intuitive insights from the negative news cases deserve further investigation.

In addition, what are the clinical decision and patient health effects of physicians using newer digital technologies to convey and receive medical information as a substitute for information supplied by sales representatives as access limits increase?48 Pharmaceutical companies are increasingly using digital technologies to connect with physicians as a consequence of factors such as changes in ways physicians are accessing medical information, cost pressures of maintaining large sales forces, and increasing access limits.48 Not well-known are the longer-term effects on clinical practice and patient health of physicians who focus on digital channels to acquire medical information relative to physicians who rely on more traditional channels. Analysis is needed on the longer-term clinical and patient health effects of policies that ban sales representatives from hospitals and medical schools as the training grounds for future physicians.

Lastly, a more complete evaluation is needed on overall health care costs and benefits of increasing physician access limits vs granting more open access. Further analysis may show increasing access limits resulting in additional patient health care costs. This may be outweighed by the costs of allowing greater access to physicians. These costs include greater congestion and disruption of physician and office practices that work against protecting patient health, encouraging greater branded drug use when appropriate lower-cost generic drugs are available for patients, and spending on sales and marketing activities that could be better spent elsewhere to drive higher benefits to patients. Extensions of this research can lead to practices by sales representatives to find ways to bring greater value to physicians in their clinical practice of treating patients. More study is needed to determine whether the effects found here for diabetes and lipid drugs are extended to other drug classes and specialties. Thus far, as we can determine from the literature, this complete empirical health economic and outcome analysis of the pros vs cons of varying access limits has not been done. Rather than assume the result, we hope this study encourages other researchers to look more rigorously into this important drug policy and clinical practice matter. The goal is to help determine the right clinical and policy path that is best for physician decision-making while protecting patient health.

Acknowledgments and disclosures:

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References

Appreciation is extended to AstraZeneca Pharmaceuticals LP (AZ) for their support and allowing this analysis to continue through an independent academic research program for public dissemination. Appreciation is also given for comments received from the following sources: members of the Ad Hoc Working Group on the Economics of the Pharmaceutical Industry; seminar participants at the University of Chicago, Lehigh University, Temple University, and Ball State University; participants at the 2011 annual conferences of the AcademyHealth Annual Research Meeting, International Health Economics Association World Congress, and Association of Clinical Researchers and Educators; and helpful reviews of past manuscript versions from Thomas Huddle, Andrew Sfekas, Mark Showalter, Lance Stell, Thomas Stossel, Michael Weber, and Jacqueline Zinn. All views expressed here and any errors that may exist remain solely those of the authors. George Chressanthis is a former employee of AZ and was project director of this research while employed there. He still retains stock grants valued at $21,474 as of March 8, 2012 in the company. The stock grants vested on March 27, 2012. Michael Seiders is a former employee of AZ effective February 1, 2012. He retains stock options and stock grants in the company. Pratap Khedkar, Nitin Jain, and Prashant Poddar acted as compensated project consultants from ZS Associates to help complete this research. All research produced from this study is the exclusive property of Temple University per intellectual property agreement with AZ. Temple University also exercises independent control over the design and implementation of the study, analysis, writing of the manuscript, and information dissemination. The lead author collaborated and cooperated with AZ per agreement in ensuring that no market-sensitive and/or AZ-confidential/proprietary information was publicly revealed in any research publication or presentation. Agreements like this to protect commercial confidential/proprietary information are common with academic-corporate research relationships. This arrangement has not altered in any way the research approach and conclusions reached by the researchers. Oversight by Temple University officials of any work completed by George Chressanthis in connection with AZ is done to ensure all research meets rigorous academic standards as an added precaution against claims of conflict of interest. Rhonda Croxton and Helen Yeh from AZ provided clinical research publication reference assistance. All views expressed here and any errors that may exist remain solely those of the authors. This research originated at and was funded by AZ.

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments and disclosures:
  8. References