electronic records

Using Default Options Within the Electronic Health Record to Increase the Prescribing of Generic-Equivalent Medications A Quasi-experimental Study Mitesh S. Patel, MD, MBA, MS*; Susan Day, MD, MPH; Dylan S. Small, PhD; John T. Howell III, MD; Gillian L. Lautenbach, MD; Eliot H. Nierman, MD; and Kevin G. Volpp, MD, PhD

Background: Low-value services, such as prescribing brand-name medications that have existing generic equivalents, contribute to unnecessary health care spending.

Objective: To evaluate the association of an intervention by using the electronic health record with provider prescription of generic- equivalent medications.

Design: Quasi-experimental study.

Setting: General internal medicine (IM) (n � 2) and family medi- cine (FM) (n � 2) clinics at the University of Pennsylvania from June 2011 to September 2012.

Participants: Attending physicians (IM, n � 38; FM, n � 17) and residents (IM, n � 166; FM, n � 34).

Intervention: In January 2012, the default in the electronic health record was changed for IM providers from displaying brand and generic medications to displaying initially only generics, with the ability to opt out.

Measurements: Monthly prescriptions of brand-name and generic- equivalent �-blockers, statins, and proton-pump inhibitors.

Results: During the preintervention period, FM providers had slightly higher rates of generic medication prescribing (range, 80.8% to 85.5%) than did IM providers (range, 75.4% to 79.6%),

but both groups had similar trends. In the postintervention period relative to the preintervention period, IM providers had an increase in generic prescribing compared with FM providers for all 3 med- ications combined (5.4 percentage points [95% CI, 2.2 to 8.7 percentage points]; P � 0.001), �-blockers (10.5 percentage points [CI, 5.8 to 15.2 percentage points]; P � 0.001), and statins (4.0 percentage points [CI, 0.4 to 7.6 percentage points]; P � 0.002). Results for proton-pump inhibitors (2.1 percentage points [CI, �3.7 to 8.0 percentage points]; P � 0.47) were not significant. Subset analyses revealed similar findings for attending physicians. Among residents, however, results were imprecise, with wide CIs.

Limitation: Observational single-center evaluation, comparison groups that represented different specialties, and a small subset of medication classes studied.

Conclusion: The use of default options was an effective method to increase the odds of prescribing generic medication equivalents for �-blockers and statins.

Primary Funding Source: U.S. Department of Veterans Affairs and Robert Wood Johnson Foundation.

Ann Intern Med. 2014;161:S44-S52. doi:10.7326/M13-3001 www.annals.org For author affiliations, see end of text. * Current Robert Wood Johnson Foundation Clinical Scholar.

Health care costs in the United States continue to riseand now account for more than $2.8 trillion annually (1). It is estimated that as much as one third of this spend- ing is due to unnecessary waste in the health care system (2). The Choosing Wisely campaign is a joint initiative launched in 2012 by the American Board of Internal Med- icine Foundation, Consumer Reports, and several medical societies (3). Its premise is to decrease unnecessary spend- ing by focusing on reducing the use of low-value services. Similarly, the American College of Physicians (ACP) has launched the High Value Care Initiative with the goal of helping physicians provide the best possible care for their patients while simultaneously reducing unnecessary costs to the health care system (4).

Prescribing brand-name medications that have existing generic equivalents is a prime example of a low-value ser- vice. These medications are often more expensive than

their generic equivalents, yet in most cases evidence sug- gests they are similar in effectiveness (5). A recent study of 20 popular multisource drugs found that in 2009, Medic- aid spent an additional $329 million that could have been saved by using existing generic equivalents instead of brand-name medications (6).

The use of default options has been recognized as a successful strategy to change behavior in many settings, including health care (7–10). Default options are effective because they are often viewed as an implicit recommenda- tion and because people tend to choose the path of least

This article is part of the Annals supplement “RWJF Clinical Scholars in Pursuit of the Value Proposition: Evaluations of Low-Cost Innovations for Prevention and Management of Conditions.” The Robert Wood Johnson Foundation provided funding for publication of this supplement, which is available online only at www.annals.org. Carol M. Mangione, MD, MPH (co-director of the RWJF Clinical Scholars Program at the University of California, Los Angeles); Jaya K. Rao, MD, MHS (Annals Deputy Editor); and Christine Laine, MD, MPH (Annals Editor in Chief), served as editors for this supplement.

See also:

Web-Only Data Supplement

Annals of Internal MedicineSupplement

S44 © 2014 American College of Physicians

resistance. If a decision maker does not opt out, the de- fault’s action takes place (7). Small changes in default set- tings can substantially affect medical decision making and provider behavior.

In January 2012, the Division of General Internal Medicine at the University of Pennsylvania in Philadelphia implemented an intervention to change the default in the electronic health record (EHR) medication prescriber with the goal of increasing the prescribing of generic equiv- alents when available. Before the intervention, if a provider searched for a brand-name medication, the options for pre- scribing brand names were shown near the top with generic equivalents listed below. After the intervention, if a pro- vider searched for a brand-name medication, only the ge- neric equivalent was listed. Providers still maintained the ability to opt out and conduct a broader search that listed the brand if warranted. The objective of this study was to evaluate the effect of this intervention on physician behav- ior related to the ordering of brand-name medications ver- sus existing generic equivalents.

METHODS The institutional review board at the University of

Pennsylvania approved this study.

Setting and Participants The sample comprised attending faculty and residents

at 2 ambulatory clinics in the Division of General Internal Medicine (IM) and 2 ambulatory clinics in the Depart- ment of Family Medicine (FM) practicing between June 2011 and September 2012 at the University of Pennsylva- nia. All of these clinics were teaching practices where at- tending physicians practiced independently and served as preceptors for residents. The practices were located within the same ZIP code, and all providers used the same EHR. We classified providers by training level as attending phy- sician or resident using publicly available listings of faculty and resident rosters (11–14), along with special request

from the Internal Medicine Residency Program, Family Medicine Residency Program, Division of General Internal Medicine, and Department of Family Medicine.

Intervention During the preintervention period (June to December

2011), both IM and FM providers used the same EHR medication prescriber and were shown the same resulting options. If a provider searched for a brand-name medica- tion, the dosing options for prescribing brand names were listed at the top, with the dosing options for generic- equivalent medications listed below (Figure 1). During the postintervention period (January to September 2012), this remained unchanged for FM providers. However, if IM providers searched for a brand-name medication, the re- sults listed only dosing options for generic-equivalent med- ications (Figure 2). The IM providers had the ability to opt out (by clicking on “Database Lookup” or pressing “F7” on the keyboard) and would then be shown the dosing options for brand names at the top, with the generic- equivalent medications listed below.

Data Provider prescription data were obtained by using

EPIC’s analytical reporting database, Clarity (EPIC Sys- tems). These data reported the number of prescriptions per month by each provider that were sent electronically to the pharmacy for each medication included in the sample. When an electronic prescription is sent to the pharmacy, a patient can arrive without a paper copy and pick up the medications after presenting appropriate identification. Our reports capture all electronic prescribing from June 2011 to September 2012. Handwritten or faxed prescrip- tions were not captured by the reports produced for this study. Reports were produced as Excel files (Microsoft), coded, and then analyzed with Stata software, version 12 (StataCorp).

Figure 1. Example of medication prescriber results before the intervention.

Browse (F4) Preference List (F5) Database Lookup (F7)

Brand

Generic

Name

COREG Search

Medications Procedures Order Panels Split

Sig Rx Type Code Type Pref List Qty Unit Formulary Coverage COREG 12.5 MG PO TABS COREG 25 MG PO TABS COREG 3.125 MG PO TABS COREG 6.25 MG PO TABS COREG CR 10 MG PO CP24 COREG CR 20 MG PO CP24 COREG CR 40 MG PO CP24 COREG CR 80 MG PO CP24 COREG 12.5 MG OR TABS (aka CARVED COREG 25 MG OR TABS (aka CARVEDIL COREG 3.125 MG OR TABS (aka CARVED COREG 6.25 MG OR TABS (aka CARVED COREG CR 10 MG OR CP24 (aka CARVE COREG CR 20 MG OR CP24 (aka CARVE COREG CR 40 MG OR CP24 (aka CARVE COREG CR 80 MG OR CP24 (aka CARVE

Brand R: Brand R: Brand R: Brand R: Brand R: Brand R: Brand R: Brand R: Generic Generic Generic Generic Generic Generic Generic Generic

Resulting medication dosing options before the intervention for internal medicine and family medicine providers when a search for “Coreg” was conducted. Brand-name options are listed first, followed by generic-equivalent options.

SupplementDefault Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

www.annals.org 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) S45

Selection of Medications We evaluated 3 classes of medications: �-blockers,

statins, and proton-pump inhibitors (PPIs). Because we did not have access to data that allowed adjustment for patient characteristics, we chose these 3 medication classes because patients with indications for these medications are similar between the 2 specialties in our study, thus provid- ing a better comparison between the intervention and con- trol groups.

Among these 3 classes of medications, we excluded prescriptions for which a generic equivalent did not exist (e.g., Crestor [AstraZeneca]) to remove bias due to scenar- ios in which it was unclear whether prescribing a non- equivalent generic medication of the same class was not feasible or had already been done but was ineffective. We excluded combination medications to remove any bias re- lated to scenarios in which the provider was attempting to decrease pill burden for the patient (for example, niacin– simvastatin extended-release) and an equivalent generic combination did not exist. We excluded prescriptions for Lipitor (Pfizer) and atorvastatin from the analysis because the generic equivalent had become available just 1 month before the intervention (30 November 2011) and there may have been alternative motives for switching between the brand and generic other than that related to the objec- tive of our study. While this represented a large proportion of statins in the original sample, prescribing trends in the preintervention period differed because of the lack of a generic equivalent (see Data Supplement, available at www .annals.org).

The following medications and their associated brand-name equivalents were included in our sample: �-blockers (atenolol, carvedilol, labetalol, metoprolol, na- dolol, and propranolol), statins (lovastatin, pravastatin, and simvastatin), and PPIs (lansoprazole, omeprazole, and pantoprazole).

Outcome Measures To evaluate the effect of the intervention for IM pro-

viders compared with FM providers, we examined monthly prescribing trends of generic medication equivalents in the pre- and postintervention periods. Outcomes were evalu-

ated for all providers (attending physicians and residents) for all medications in the sample and by medication class (�-blockers, statins, and PPIs). Subset analyses were con- ducted to separately examine trends for attending physi- cians and residents because evidence suggests that these 2 groups may have differing practice patterns (15–17). Res- idents are closer than attending physicians to medical school training, where generic medication names are more commonly taught and tested. Attending physicians are likely to be exposed to longer periods of industry detailing than are residents. In addition, interventions to change provider behavior may differ in effect among attending physicians, who have had longer to form practicing habits, than residents, who may be newly learning their practice style.

Statistical Analysis For each measure, a multivariate logistic regression

model was fit by using each prescription as the unit of analysis and clustering on provider. The effect of time was modeled by a dummy variable for each month. We evalu- ated the effect of the intervention by using an interaction term for provider specialty (IM vs. FM) and time to com- pare the proportion of generic medication prescriptions for each medication with itself before and after the interven- tion, contrasting the changes in IM to changes in FM. By using FM (not exposed to the intervention) as the control group in this study design, we reduce potential biases from unmeasured variables, such as general trends in prescribing behavior over time. Standard errors in the models were adjusted to account for clustering provider (18, 19). We estimated the odds of prescribing a generic medication for IM providers compared with FM providers in the postin- tervention period relative to the preintervention period. To assess the mean effect of the intervention in the postinter- vention period, we exponentiated the mean of the monthly log odds ratios (ORs) and calculated a 95% CI for this quantity (20, 21). We report the estimated probabilities of generic medication prescribing, which were derived from the logit model. In the Data Supplement, we report the odds ratios with 95% CIs for each month in the postin-

Figure 2. Example of medication prescriber results after the intervention.

Browse (F4) Preference List (F5) Database Lookup (F7)

Generic

Name

COREG Search

Medications Procedures Order Panels Split

Sig Rx Type Code Type Pref List Qty Unit Formulary Coverage CARVEDILOL 12.5 MG OR TABS (CORE CARVEDILOL 12.5 MG OR TABS (CORE CARVEDILOL 25 MG OR TABS (COREG CARVEDILOL 25 MG OR TABS (COREG CARVEDILOL 25 MG OR TABS (COREG CARVEDILOL 25 MG OR TABS (COREG CARVEDILOL 3.125 MG OR TABS (COR CARVEDILOL 3.125 MG OR TABS (COR CARVEDILOL 6.25 MG OR TABS (CORE CARVEDILOL 6.25 MG OR TABS (CORE

Generic Generic Generic Generic Generic Generic Generic Generic Generic Generic

One tab twice daily One tab twice daily One tab twice daily One tab twice daily 2 tabs twice daily 2 tabs twice daily One tab twice daily One tab twice daily One tab twice daily One tab twice daily

GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO GIM MEDICATIO

180 60 180 60 360 120 180 60 180 60

Tab Tab Tab Tab Tab Tab Tab Tab Tab Tab

Resulting medication dosing options after the intervention for internal medicine providers when a search for “Coreg” was conducted. Only the generic-equivalent options are listed; users have the ability to opt out and choose the brand-name if warranted.

Supplement Default Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

S46 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) www.annals.org

tervention period to evaluate the level and trajectory of changes in trend.

We performed a test of controls for the preinterven- tion period (June to December 2011) to evaluate whether the null hypothesis of parallel trends between IM and FM trends could be rejected (22).

Role of the Funding Source This study was not funded by external sources. The

lead investigator was supported by the Department of Vet- erans Affairs and the Robert Wood Johnson Foundation. The funding sources had no role in the design, conduct, or analysis of the study and had no role in the preparation, review, or approval of the manuscript.

RESULTS The study sample comprised 38 IM attending physi-

cians, 17 FM attending physicians, 166 IM residents, and 34 FM residents (Table 1). The rate of medications pre- scribed per month between the preintervention period and the postintervention period decreased slightly for IM at- tending physicians (21.1 vs. 18.7), increased slightly for FM attending physicians (14.9 vs. 17.0), and remained steady for IM residents (2.6 vs. 2.7) and FM residents (3.4 vs. 3.6).

During the preintervention period, FM providers had slightly higher rates of generic medication prescribing (range, 80.8% to 85.5%) than did IM providers (range, 75.4% to 79.6%) (Table 2). Test of controls for the pre- intervention period could not reject the null hypothesis of parallel trends between IM and FM providers (OR, 0.92 [95% CI, 0.71 to 1.20]; P � 0.56).

In the postintervention period relative to the preinter- vention period, IM providers had an increase in generic prescribing compared with FM providers for all 3 medica- tions (5.4 percentage points [CI, 2.2 to 8.7 percentage points]; P � 0.001), �-blockers (10.5 percentage points [CI, 5.8 to 15.2 percentage points]; P � 0.001), and statins (4.0 percentage points [CI, 0.4 to 7.6 percentage points]; P � 0.002). Results for PPIs were not significant

(2.1 percentage points [CI, �3.7 to 8.0 percentage points]; P � 0.47).

In subset analyses, IM attending physicians had an increase in generic prescribing compared with FM attend- ing physicians in the postintervention period relative to the preintervention period for all 3 medications (6.5 percent- age points [CI, 2.2 to 10.7 percentage points]; P � 0.001), �-blockers (11.4 percentage points [CI, 5.6 to 17.2 percentage points]; P � 0.001), and statins (5.5 percentage points [CI, 0.8 to 10.1 percentage points]; P � 0.001). Results for PPIs were not significant (4.7 percentage points [CI, �2.6 to 12.1 percentage points]; P � 0.19) (Table 3). Subset analyses between IM residents and FM residents were imprecise, with wide CIs (Table 4). This may have been due to the small sample size; therefore, we were un- able to rule out meaningful changes in prescribing rates among residents.

DISCUSSION We evaluated the effect of using default options within

the EHR to change provider prescribing behavior in the ambulatory setting. Our study found that this intervention led to an increase in prescribing generic medications in the postintervention period when comparing providers in the intervention group to control. When we evaluated by med- ication class, the intervention was more effective for �-blockers and statins than for PPIs. In subset analyses by training level, we found similar trends among attending physicians. We did not find meaningful changes among residents, and this was probably due to limitations in sam- ple size.

Several factors may be contributing to the findings in our study. First, PPIs were available over the counter, and this may help explain why the intervention was less effec- tive for this medication class. Some patients may already have started a PPI and decided to use a brand or a generic equivalent. Second, residents had higher baseline rates of generic prescribing in the preintervention period than at- tending physicians, specifically among IM providers (Ta-

Table 1. Sample Provider and Prescription Characteristics in the Pre- and Postintervention Periods

Sample Characteristics Preintervention Period (June–December 2011) Postintervention Period (January–September 2012) Entire Study Period: Providers, nProviders, n Total

Prescriptions, n Prescriptions per Month per Provider, n

Providers, n Total Prescriptions, n

Prescriptions per Month per Provider, n

Internal medicine (intervention group)

Attending physicians 29 4289 21.1 33 5545 18.7 38 Residents 131 2409 2.6 142 3467 2.7 166 Total 160 6698 6.0 175 9012 5.7 204

Family medicine (control group)

Attending physicians 17 1772 14.9 16 2454 17.0 17 Residents 24 564 3.4 30 977 3.6 34 Total 41 2336 8.1 46 3431 8.3 51

SupplementDefault Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

www.annals.org 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) S47

bles 3 and 4). This may be explained by the fact that residents are closer to medical school training, where they are often taught only generic names of medications. In addition, because residents have been practicing for less time, they are less likely to have been exposed to as much pharmaceutical detailing as attending physicians. Further- more, although IM attending physicians in the sample were informed directly about the intervention, residents were not; this may have contributed to its effectiveness. Third, results did trend toward higher increases in generic prescribing among IM residents than among FM residents, but small sample sizes (particularly among FM residents) may have limited our power to detect statistically signifi- cant changes. Fourth, monthly trend analysis reveals that although the intervention occurred in January 2012, it was

not until March 2012 that provider behavior changed sig- nificantly. This may reflect a higher proportion of opt-outs immediately after the intervention was implemented that waned over time.

This intervention demonstrates the use of default op- tions within the EHR as a method applicable to reduce other low-value services that contribute to unnecessary health care spending (2). Default options are best used when there is a clear alternative that is dominant from a value perspective (9). Defaults have been demonstrated as a successful strategy to increase organ donor rates (23–25) and increase employee contributions to health savings ac- counts (26 –28). Default options within an EHR may be more meaningful in changing behavior than are active alerts because of the well-documented fatigue that occurs

Table 2. Estimated Probabilities of Prescribing Generic Medication Equivalents for General IM Providers Compared With FM Providers, by Medication Class*

All Providers All Medications �-Blockers Statins PPIs

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

2011 June 76.0 83.6 �7.6 68.7 77.8 �9.1 85.9 89.3 �3.4 71.9 79.4 �7.5 July 75.4 85.5 �10.1 67.0 90.1 �23.1 88.3 87.0 1.3 69.2 78.7 �9.5 August 79.6 81.9 �2.3 74.9 83.0 �8.1 88.5 88.3 0.2 73.7 72.5 1.2 September 78.9 84.3 �5.4 73.2 87.4 �14.2 86.1 95.4 �9.3 76.0 65.4 10.6 October 77.8 82.1 �4.3 74.0 75.8 �1.8 85.8 91.1 �5.3 72.6 76.6 �4.0 November 77.2 80.9 �3.7 71.7 76.5 �4.8 86.3 91.4 �5.1 72.3 72.0 0.3 December 76.4 80.8 �4.4 69.3 78.0 �8.7 87.5 92.5 �5.0 72.6 69.4 3.2 Mean 77.3 82.7 �5.4 71.3 81.2 �10.0 86.9 90.7 �3.8 72.6 73.4 �0.8

2012 January 80.6 83.6 �3.0 74.8 77.2 �2.4 85.9 94.7 �8.8 80.6 74.7 5.9 February 81.4 84.6 �3.2 77.0 74.6 2.4 89.9 91.2 �1.3 77.4 85.7 �8.3 March 83.2 83.1 0.1 78.7 82.7 �4.0 89.2 90.8 �1.6 81.3 71.9 9.4 April 85.5 82.4 3.1 83.3 81.8 1.5 90.4 87.7 2.7 82.1 74.7 7.4 May 84.0 80.6 3.4 79.4 77.5 1.9 90.9 87.3 3.6 81.5 76.0 5.5 June 83.9 81.0 2.9 79.7 78.1 1.6 91.7 83.7 8.0 80.2 81.0 �0.8 July 85.5 83.0 2.5 83.2 74.3 8.9 92.7 90.8 1.9 80.9 82.1 �1.2 August 82.9 83.8 �0.9 78.4 80.8 �2.4 91.5 89.7 1.8 78.3 79.3 �1.0 September 80.7 85.3 �4.6 76.9 79.9 �3.0 86.1 90.3 �4.2 80.1 85.0 �4.9 Mean 83.1 83.0 0.0 79.0 78.5 0.5 89.8 89.6 0.2 80.3 78.9 1.3

Year and Month

Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value

2012 January 2.4 0.38 7.6 0.161 �5.0 0.072 6.7 0.13 February 2.2 0.58 12.4 0.02 2.5 0.45 �7.5 0.17 March 5.5 0.04 6.0 0.34 2.2 0.44 10.2 0.07 April 8.5 0.001 11.5 0.07 6.5 0.02 8.2 0.18 May 8.8 0.002 11.9 0.003 7.4 0.01 6.3 0.23 June 8.3 0.004 11.6 0.02 11.8 0.004 0.0 0.96 July 7.9 0.01 18.9 �0.001 5.7 0.12 �0.4 0.88 August 4.5 0.05 7.6 0.14 5.6 0.03 �0.2 0.94 September 0.8 0.99 7.0 0.172 �0.4 0.94 �4.1 0.35 Mean 5.4 �0.001 10.5 �0.001 4.0 0.002 2.1 0.47 95% CI 2.2 to 8.7 5.8 to 15.2 0.4 to 7.6 �3.7 to 8.0

IM � general internal medicine; FM � family medicine; PPI � proton-pump inhibitor. * Test of controls for the preintervention period (June to December 2011) are as follows: all medications (odds ratio [OR], 0.92 [95% CI, 0.71–1.20]), �-blockers (OR, 0.86 [CI, 0.49 –1.52]), statins (OR, 1.15 [CI, 0.59 –2.23]), and PPIs (OR, 0.82 [CI, 0.52–1.29]).

Supplement Default Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

S48 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) www.annals.org

over time with alerts (29 –31). Changes in default options, such as the one in this study, can provide implicit recom- mendations to providers without giving more information (for example, an alert describing how generics are less ex- pensive that brand names) or requiring additional steps to provide care (for example, an alert requiring an extra click or override).

The evaluation of this intervention has important pol- icy and clinical implications. Although all 50 states have some form of generic substitution laws, mandatory substi- tution laws exist in only 15 states (32). Among the remain- ing 35 states, 80% require patient consent before generic substitution is allowed. A recent study found that states requiring patient consent had a 25% lower conversion to generic than states without such a law (33). Several recent studies demonstrated continued unnecessary spending on

brand-name medications for which generic equivalents ex- ist. An evaluation of 20 popular multisource drugs found that in 2009, Medicaid spent an additional $329 million that could have been saved by using existing generic equivalents instead of brand-name medications (6). In an- other study, prescriptions for equivalent medication classes among diabetic patients were compared between Medicare enrollees and patients receiving care in the Veteran Affairs health system (34). After adjustment for patient character- istics and geographic variation, Medicare was observed to spend $1.4 billion more per year on brand-name medica- tions than the Veterans Affairs health system. Even if man- datory generic substitutions laws were expanded to all 50 states, there is still concern that this interaction at the phar- macy might lead to confusion for the patient. Many pa- tients require guidance to make medical decisions, and ge-

Table 3. Estimated Probabilities of Increased Prescribing of Generic Medication Equivalents for General IM Attendings Compared With FM Attendings, by Medication Class*

Attendings All Medications �-Blockers Statins PPIs

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

2011 June 73.7 84.4 �10.7 66.2 76.5 �10.3 84.2 91.0 �6.8 69.8 80.4 �10.6 July 73.7 84.3 �10.6 66.5 87.2 �20.7 86.7 87.8 �1.1 65.6 76.1 �10.5 August 76.5 81.3 �4.8 72.2 81.9 �9.7 86.2 90.1 �3.9 69.3 69.4 �0.1 September 73.8 83.8 �10.0 70.3 87.0 �16.7 81.7 95.1 �13.4 68.5 64.4 4.1 October 74.8 80.0 �5.2 71.1 73.0 �1.9 82.3 91.1 �8.8 70.1 71.6 �1.5 November 73.3 74.9 �1.6 70.7 74.7 �4.0 82.6 91.2 �8.6 64.4 65.6 �1.2 December 72.6 80.7 �8.1 65.1 77.5 �12.4 84.2 92.7 �8.5 68.5 66.7 1.8 Mean 74.1 81.3 �7.3 68.9 79.7 �10.8 84.0 91.3 �7.3 68.0 70.6 �2.6

2012 January 76.4 82.6 �6.2 72.5 74.0 �1.5 81.1 95.7 �14.6 75.0 72.5 2.5 February 78.3 84.4 �6.1 73.3 73.5 �0.2 87.2 91.1 �3.9 74.8 87.8 �13.0 March 82.0 82.5 �0.5 76.8 82.5 �5.7 88.8 92.6 �3.8 80.2 62.9 17.3 April 83.4 81.1 2.3 81.4 78.0 3.4 88.1 87.9 0.2 80.0 72.1 7.9 May 81.1 79.2 1.9 76.4 74.0 2.4 88.9 88.6 0.3 77.0 73.0 4.0 June 82.0 78.6 3.4 78.5 76.6 1.9 88.4 82.2 6.2 79.0 75.5 3.5 July 84.2 79.6 4.6 82.6 70.4 12.2 91.4 88.8 2.6 79.1 78.1 1.0 August 80.3 81.9 �1.6 75.0 77.1 �2.1 89.9 90.3 �0.4 75.3 74.7 0.6 September 78.4 83.7 �5.3 74.8 79.8 �5.0 85.0 88.2 �3.2 76.8 81.0 �4.2 Mean 80.7 81.5 �0.8 76.8 76.2 0.6 87.6 89.5 �1.8 77.5 75.3 2.2

Month Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value

2012 January 1.1 0.65 9.3 0.14 �7.3 0.03 5.1 0.36 February 1.2 0.82 10.6 0.10 3.4 0.43 �10.4 0.09 March 6.8 0.03 5.1 0.57 3.5 0.50 19.9 0.002 April 9.6 0.001 14.2 0.05 7.5 0.02 10.5 0.21 May 9.2 0.01 13.2 0.01 7.6 0.01 6.6 0.35 June 10.7 0.003 12.7 0.06 13.5 0.004 6.1 0.34 July 11.9 �0.001 23.0 �0.001 9.9 0.02 3.6 0.62 August 5.7 0.03 8.7 0.14 6.9 0.01 3.2 0.63 September 2.0 0.58 5.8 0.37 4.1 0.26 �1.6 0.73 Mean 6.5 �0.001 11.4 �0.001 5.5 �0.001 4.7 0.19 95% CI 2.2 to 10.7 5.6 to 17.2 0.8 to 10.1 �2.6 to 12.1

IM � general internal medicine; FM � family medicine; PPI � proton-pump inhibitor. * Test of controls for the preintervention period (June to December 2011) are as follows: all medications (odds ratio [OR], 0.99 [95% CI, 0.74 –1.31]), �-blockers (OR, 1.01 [CI, 0.53–1.91]), statins (OR, 1.15 [CI, 0.56 –2.40]), and PPIs (OR, 0.80 [CI, 0.46 –1.39]). Test of controls for August to December 2011 for �-blockers (OR, 0.71 [CI, 0.26 –1.92]).

SupplementDefault Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

www.annals.org 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) S49

neric substitution at the pharmacy leads to a situation in which the pharmacist is challenging the provider’s implicit recommendation. Some patients may perceive that because generics are cheaper, they must be lower quality (35). By using defaults within the EHR, providers may be more likely to prescribe generics to patients. This allows patients to have the opportunity to discuss brands versus generics while still in the clinic with the provider. It also helps avoid any unnecessary confusion or misperception on quality at the pharmacy.

Our study is subject to several limitations. First, any observational study is susceptible to unmeasured con- founding. However, by using FM as a comparison group, we reduce potential biases from unmeasured variables, such as general trends in prescribing behavior over time. There-

fore, confounding would occur only if the prevalence of unmeasured risk factors changed at different rates between IM and FM providers. Second, the intervention and our data were limited to the setting of electronic prescribing, which accounted for most prescriptions in these practices. We did not have data on handwritten or faxed prescrip- tions and therefore could not evaluate whether differential trends existed in these settings. Third, our study excluded Lipitor and atorvastatin, about 30% of statins prescribed in the original sample (see Data Supplement), because the generic equivalent did not become available until 1 month before the intervention. While generic prescribing behavior for atorvastatin probably reflected different factors than other statins, we cannot rule out that generic availability of atorvastatin did not increase general awareness of generic

Table 4. Estimated Probabilities of Increased Prescribing of Generic Medication Equivalents for General IM Residents Compared With FM Residents, by Medication Class*

Residents All Medications �-Blockers Statins PPIs

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

IM, % FM, % Difference (IM � FM), percentage points

2011 June 82.5 80.9 1.6 75.6 81.8 �6.2 90.6 82.8 7.8 79.6 76.5 3.1 July 79.1 89.7 �10.6 68.0 100.0 �32.0 91.7 83.3 8.4 76.2 85.2 �9.0 August 84.6 83.3 1.3 79.7 86.2 �6.5 92.2 83.8 8.4 80.3 80.0 0.3 September 87.1 85.9 1.2 78.4 88.9 �10.5 92.3 96.3 �4.0 88.9 68.4 20.5 October 82.7 89.7 �7.0 78.8 87.5 �8.7 91.5 90.9 0.6 76.6 90.0 �13.4 November 83.3 85.3 �2.0 72.8 81.8 �9.0 92.6 92.3 0.3 86.1 83.3 2.8 December 82.9 81.4 1.5 76.4 79.3 �2.9 93.1 91.7 1.4 79.4 75.8 3.6 Mean 83.2 85.2 �2.0 75.7 86.5 �10.8 92.0 88.7 3.3 81.0 79.9 1.1

2012 January 87.0 86.8 0.2 74.8 87.1 �12.3 93.1 91.2 1.9 88.7 80.8 7.9 February 86.5 85.0 1.5 84.3 78.6 5.7 94.2 91.7 2.5 81.1 80.6 0.5 March 85.2 84.6 0.6 81.6 83.3 �1.7 89.9 83.8 6.1 83.1 86.3 �3.2 April 88.9 86.8 2.1 86.4 92.9 �6.5 94.1 86.3 7.8 85.5 80.8 4.7 May 87.9 84.5 3.4 82.7 86.8 �4.1 93.9 82.8 11.1 88.0 83.3 4.7 June 87.8 86.5 1.3 82.2 81.1 1.1 98.0 87.9 10.1 82.6 92.3 �9.7 July 87.7 90.2 �2.5 84.3 87.5 �3.2 94.5 95.0 �0.5 83.9 87.5 �3.6 August 86.6 87.3 �0.7 83.7 88.0 �4.3 93.9 88.2 5.7 82.2 85.7 �3.5 September 84.8 88.7 �3.9 81.6 80.0 1.6 88.2 97.3 �9.1 85.2 90.5 �5.3 Mean 86.9 86.7 0.2 82.4 85.0 �2.6 93.3 89.4 4.0 84.5 85.3 �0.8

Month Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value Difference-in-Difference (Relative to 2011 Mean), percentage points

P Value

2012 January 2.2 0.58 �1.5 0.91 �1.4 0.88 6.8 0.32 February 3.5 0.68 16.5 0.09 �0.8 0.99 �0.6 0.99 March 2.6 0.64 9.1 0.27 2.8 0.75 �4.3 0.68 April 4.1 0.44 4.3 0.97 4.5 0.53 3.6 0.58 May 5.4 0.25 6.7 0.50 7.8 0.18 3.6 0.56 June 3.3 0.33 11.9 0.05 6.8 0.15 �10.8 0.22 July �0.5 0.84 7.6 0.56 �3.8 0.67 �4.7 0.60 August 1.3 0.74 6.5 0.50 2.4 0.58 �4.6 0.57 September �1.9 0.63 12.4 0.14 �12.4 0.07 �6.4 0.32 Mean 2.2 0.34 8.2 0.063 0.7 0.79 �2.0 0.70 95% CI �2.2 to 6.7 0 to 16.3 �5.6 to 7.0 �10.7 to 6.8

IM � general internal medicine; FM � family medicine; PPI � proton-pump inhibitor. * Test of controls for the preintervention period (June to December 2011) are as follows: all medications (odds ratio [OR], 0.76 [95% CI, 0.42–1.40]), �-blockers (OR, 0.09 [CI, 0.03– 0.26]), statins (OR, 1.06 [CI, 0.22–5.23]), and PPIs (OR, 0.85 [CI, 0.34 –2.13]).

Supplement Default Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

S50 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) www.annals.org

availability of statins or affect statin prescribing in other ways. Fourth, our study did not evaluate trends beyond 9 months after the intervention. While rates of generic pre- scribing for IM providers were lower than those for FM providers in the last month of our study (September 2012), this change was not statistically significant in our model and may reflect usual variation. However, longer follow-up would be required for further evaluation. Finally, we did not have access to patient characteristics or pharmacy data and therefore were not able to adjust our model for these factors.

In conclusion, the use of an intervention to change default options within the EHR was an effective method to increase generic prescribing of �-blockers and statins when compared with a control group. These findings offer valu- able insights for clinical decision-support teams evaluating interventions that could lead to sustained changes in pro- vider behavior. Lessons from behavioral economics and specifically the use of default options within the EHR could be leveraged in other contexts to reduce unnecessary waste and improve the value of health care delivered to patients.

From the University of Pennsylvania and the Center for Health Equity Research and Promotion, Veterans Affairs Medical Center, Philadelphia, Pennsylvania.

Grant Support: Dr. Patel was supported by the Department of Veteran Affairs and the Robert Wood Johnson Foundation.

Disclosures: Disclosures can be viewed at www.acponline.org/authors /icmje/ConflictOfInterestForms.do?msNum�M13-3001.

Reproducible Research Statement: Study protocol: Available from Dr. Patel (e-mail, [email protected]). Statistical code: Parts available from Dr. Patel (e-mail, [email protected]). Data set: Not available.

Requests for Single Reprints: Mitesh S. Patel, MD, MBA, MS, Divi- sion of General Internal Medicine, Perelman School of Medicine at the University of Pennsylvania, 423 Guardian Drive, 12th Floor, Blockley Hall, Philadelphia, PA 19104; e-mail, [email protected].

Current Author Addresses: Dr. Patel: Division of General Internal Medicine, Perelman School of Medicine at the University of Pennsylva- nia, 423 Guardian Drive, 12th Floor Blockley Hall, Philadelphia, PA 19104. Drs. Day, Lautenbach, and Nierman: Division of General Internal Med- icine, Perelman School of Medicine at the University of Pennsylvania, 3701 Market Street, 7th Floor, Philadelphia, PA 19104. Dr. Small: Department of Statistics, The Wharton School at the Uni- versity of Pennsylvania, 464 Huntsman Hall, 3730 Walnut Street, Phil- adelphia, PA 19104. Dr. Howell: Office of the Chief Medical Information Officer at the University of Pennsylvania, 3001 Market Street, 4th Floor, Philadelphia, PA 19104. Dr. Volpp: Division of General Internal Medicine, Perelman School of Medicine at the University of Pennsylvania, 1120 Blockley Hall, 423 Guardian Drive, Philadelphia, PA 19104-6021.

Author Contributions: Conception and design: M.S. Patel, S. Day, J.T. Howell, G.L. Lautenbach, E.H. Nierman, K.G. Volpp.

Analysis and interpretation of the data: M.S. Patel, S. Day, D.S. Small, K.G. Volpp. Drafting of the article: M.S. Patel, S. Day. Critical revision of the article for important intellectual content: M.S. Patel, S. Day, D.S. Small, K.G. Volpp. Final approval of the article: M.S. Patel, E.H. Nierman, K.G. Volpp. Provision of study materials or patients: M.S. Patel, S. Day, G.L. Lautenbach. Statistical expertise: M.S. Patel, D.S. Small. Obtaining of funding: M.S. Patel. Administrative, technical, or logistic support: M.S. Patel, S. Day, K.G. Volpp. Collection and assembly of data: M.S. Patel.

References 1. Centers for Medicare & Medicaid. National Health Expenditure Projec- tions 2010-2020. Accessed at https://www.cms.gov/NationalHealthExpendData /downloads/proj2010.pdf on 1 May 2013. 2. Berwick DM, Hackbarth AD. Eliminating waste in US health care. JAMA. 2012;307:1513-6. [PMID: 22419800] 3. Cassel CK, Guest JA. Choosing wisely: helping physicians and patients make smart decisions about their care. JAMA. 2012;307:1801-2. [PMID: 22492759] 4. American College of Physicians. High value care initiative. Accessed at https: //hvc.acponline.org on 15 March 2014 5. Kesselheim AS, Misono AS, Lee JL, Stedman MR, Brookhart MA, Choudhry NK, et al. Clinical equivalence of generic and brand-name drugs used in cardiovascular disease: a systematic review and meta-analysis. JAMA. 2008; 300:2514-26. [PMID: 19050195] 6. Brill A. Overspending on multi-source drugs in Medicaid. American Enter- prise Institute for Public Policy Research. 28 March 2011. Accessed at www.aei .org/papers/health/entitlements/overspending-on-multi-source-drugs-in-medicaid / on 31 January 2013. 7. Halpern SD, Ubel PA, Asch DA. Harnessing the power of default options to improve health care. N Engl J Med. 2007;357:1340-4. [PMID: 17898105] 8. Loewenstein G, Brennan T, Volpp KG. Asymmetric paternalism to improve health behaviors. JAMA. 2007;298:2415-7. [PMID: 18042920] 9. Patel MS, Volpp KG. Leveraging insights from behavioral economics to in- crease the value of health-care service provision. J Gen Intern Med. 2012;27: 1544-7. [PMID: 22549296] 10. Halpern SD, Loewenstein G, Volpp KG, Cooney E, Vranas K, Quill CM, et al. Default options in advance directives influence how patients set goals for end-of-life care. Health Aff (Millwood). 2013;32:408-17. [PMID: 23381535] 11. Penn Medicine Family Medicine. Current residents. Accessed at www.uphs .upenn.edu/family-medicine/education/residency/current-residents.html on 31 January 2013. 12. Penn Medicine Family Medicine. Resident placements. Accessed at www.uphs.upenn.edu/family-medicine/education/residency/resident-placements -alumni.html on 31 January 2013. 13. Penn Medicine Family Medicine. Physicians & team. Accessed at www .pennmedicine.org/family-medicine/patient/team.html on 31 January 2013. 14. Penn Medicine Division of General Internal Medicine. Faculty and staff. Accessed at www.uphs.upenn.edu/dgim/faculty at 31 January 2013. 15. Karlinsky JB, Barrera CM, Goodman DM, Abramson MH, Kaufman JS, Creech DF. Differences in outpatient corticosteroid prescribing patterns between attending and house staff physicians as an indicator of the quality of supervision. Am J Med Qual. 2000;15:26-33. [PMID: 10680222] 16. Boltri JM, Gordon ER, Vogel RL. Effect of antihypertensive samples on physician prescribing patterns. Fam Med. 2002;34:729-31. [PMID: 12448641] 17. Charlson ME, Karnik J, Wong M, McCulloch CE, Hollenberg JP. Does experience matter? A comparison of the practice of attendings and residents. J Gen Intern Med. 2005;20:497-503. [PMID: 15987323] 18. Rogers WH. Regression standard errors in clustered samples. Stata Technical Bulletin Reports. 1993;3:88-94. 19. Williams RL. A note on robust variance estimation for cluster-correlated data. Biometrics. 2000;56:645-6. [PMID: 10877330]

SupplementDefault Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

www.annals.org 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) S51

20. Newman SC. Biostatistical Methods in Epidemiology. New York: Wiley; 2001:136. 21. Agresti A. An Introduction to Categorical Data Analysis. New York: Wiley; 2007:108-9. 22. Meyer BD. Natural and quasi-experiments in economics. Journal of Business and Economic Statistics. 1995;13:151-61. 23. Gimbel RW, Strosberg MA, Lehrman SE, Gefenas E, Taft F. Presumed consent and other predictors of cadaveric organ donation in Europe. Prog Trans- plant. 2003;13:17-23. [PMID: 12688644] 24. Johnson EJ, Goldstein D. Medicine. Do defaults save lives? Science. 2003; 302:1338-9. [PMID: 14631022] 25. Johnson EJ, Goldstein DG. Defaults and donation decisions. Transplanta- tion. 2004;78:1713-6. [PMID: 15614141] 26. Schweitzer M, Hershey JC, Asch DA. Individual choice in spending ac- counts. Can we rely on employees to choose well? Med Care. 1996;34:583-93. [PMID: 8656724] 27. Madrian BC, Shea DF. The power of suggestion: inertia in 401(k) partici- pation and savings behavior. Q J Econ. 2001;116:1149-87. 28. Choi JJ, Laibson D, Mardian BC, Metrick A. Defined contribution pen- sions: plan rules, participant decisions, and the path of least resistance. In: Poterba JM, ed. Tax Policy and the Economy. Vol 16. Cambridge, MA: MIT Pr; 2002: 67-113. 29. Black AD, Car J, Pagliari C, Anandan C, Cresswell K, Bokun T, et al. The impact of eHealth on the quality and safety of health care: a systematic

overview. PLoS Med. 2011;8:e1000387. [PMID: 21267058] doi:10.1371 /journal.pmed.1000387 30. van der Sijs H, Aarts J, Vulto A, Berg M. Overriding of drug safety alerts in computerized physician order entry. J Am Med Inform Assoc. 2006;13:138-47. [PMID: 16357358] 31. Avery AJ, Savelyich BS, Sheikh A, Cantrill J, Morris CJ, Fernando B, et al. Identifying and establishing consensus on the most important safety features of GP computer systems: e-Delphi study. Inform Prim Care. 2005;13:3-12. [PMID: 15949170] 32. Mazer JS. Generic substitution: the science and saving. D.C. governor’s staff briefing. Pharmaceutical Care Management Association. 9 March 2011. Accessed at www.amcp.org/WorkArea/DownloadAsset.aspx?id�10530 on 15 March 2014. 33. Shrank WH, Choudhry NK, Agnew-Blais J, Federman AD, Liberman JN, Liu J, et al. State generic substitution laws can lower drug outlays under Medic- aid. Health Aff (Millwood). 2010;29:1383-90. [PMID: 20606192] 34. Gellad WF, Donohue JM, Zhao X, Mor MK, Thorpe CT, Smith J, et al. Brand-name prescription drug use among Veterans Affairs and Medicare Part D patients with diabetes: a national cohort comparison. Ann Intern Med. 2013;159: 105-14. [PMID: 23752663] 35. Volpp KG, Loewenstein G, Asch DA. Choosing wisely: low-value services, utilization, and patient cost sharing. JAMA. 2012;308:1635-6. [PMID: 23093160]

Supplement Default Options in the EHR to Increase Prescribing of Generic-Equivalent Medications

S52 18 November 2014 Annals of Internal Medicine Volume 161 • Number 10 (Supplement) www.annals.org

Copyright © American College of Physicians 2014.