 To compare changes in vaccination rates (pneumococcal polysaccharide vaccine [PPSV]; tetanus, diphtheria, and pertussis [Tdap] vaccine; and influenza vaccine) among high-risk adults following an intervention (June 1, 2013, to January 31, 2015) that used the 4 Pillars Practice Transformation Program (4 Pillars Program). 

 Post hoc analysis of data from a randomized controlled cluster trial.

 Eighteen primary care practices received staff education, guidance for using the 4 Pillars Program, and support for a practice immunization champion. Paired 

 tests were used to compare vaccination rates separately for those with diabetes, chronic lung or chronic heart disease, or other high-risk conditions. Student’s 

 tests were used to compare vaccination rates across high-risk conditions. Generalized estimating equation modeling was used to determine the likelihood of vaccination. 

International Classification of Diseases, Ninth Revision, Clinical Modification

 codes, 4737 patients aged 18 to 64 years were identified as having diabetes (n = 1999), chronic heart disease (n = 658), chronic lung disease (n = 1682), or another high-risk condition (n = 764). PPSV uptake increased by 12.2 percentage points (PP), Tdap vaccination increased by 11.4 PP, and influenza vaccination increased by 4.8 PP. In regression analyses, patients with diabetes (odds ratio [OR], 2.2; 95% CI, 1.80-2.73), chronic lung disease (OR, 1.50; 95% CI, 1.21-1.87), or chronic heart disease (OR, 1.32; 95% CI, 1.02-1.71) were more likely to receive PPSV than those without the respective high-risk condition. Those with diabetes (OR, 1.14; 95% CI, 1.01-1.28) or chronic lung disease (OR, 1.14; 95% CI, 1.01-1.30) were more likely to receive an influenza vaccine than those without the respective condition. The likelihood of Tdap vaccination was not significantly associated with any of the chronic conditions tested.

 An intervention including the 4 Pillars Program was associated with significant increases in vaccination of high-risk adults. However, the overall uptake of recommended vaccines for those with high-risk conditions remained below national goals. 

Using the 4 Pillars Practice Transformation Program (4 Pillars Program), primary care practices can achieve meaningful improvements in adult vaccination rates among high-risk adults younger than 65 years, who are historically a group with low vaccine uptake. 

There remains a gap between current vaccine uptake and national goals for high-risk adults. 

The 4 Pillars Program provides step-by-step guidance for implementing evidence-based strategies to improve vaccine uptake. 

In the program, those with diabetes were more likely to receive the pneumococcal polysaccharide vaccine than those with other high-risk conditions.

Adults with certain chronic medical conditions are at higher risk of complications from some vaccine-preventable diseases because these conditions are known to compromise the immune response to infection or increase vulnerability to the effects of infection.

 For example, among adults aged 18 to 64 years, rates of pneumococcal pneumonia are 3.0 to 9.8 times higher for those with chronic heart disease, lung disease, or diabetes compared with healthy adults; for invasive pneumococcal disease, rates are 3.6 to 7.7 times higher.

 Not only are vaccination rates for this group woefully low—20.3% for pneumococcal polysaccharide vaccine (PPSV) in 2014

—and far from the Healthy People 2020 goal of 60%,

 there are significant disparities in rates by race,

 health insurance status, and frequency of contact with a medical provider.

 Although the 2013 influenza vaccination rate among high-risk adults (49.5%) was higher than that among those without high-risk conditions (32.9%),

 this value is also below the US goal of 70%.

 Tetanus, diphtheria, and pertussis (Tdap) vaccine uptake among all adults 19 years or older was 20.1% in 2014.

Recent research on interventions to improve vaccination among high-risk adults is scant. Two studies focused on specialized high-risk populations (patients on dialysis

PPSV uptake to 65.5% and 92%, respectively, through extensive provider and patient education and outreach to patients, including home vaccination visits. Among patients on dialysis

 multicomponent interventions resulted in increases in influenza vaccine uptake of 4 to 5 percentage points (PP). 

We undertook a 2-year study of 18 primary care practices to test the effectiveness of an intervention designed to increase uptake of adult vaccines using the 4 Pillars Practice Transformation Program (4 Pillars Program). This program is a step-by-step guide for medical practices to implement evidence-based strategies for increasing vaccination rates in primary care or other outpatient settings.

 These strategies are applicable to many practice settings and populations. Overall findings from the randomized controlled cluster trial (RCCT) and pre-post studies have been published.

 The purpose of this study was to compare the effect of the intervention on adult PPSV, influenza, and Tdap vaccination rates and likelihood of vaccination among adults aged 18 to 64 years with the 3 most common high-risk medical conditions (diabetes, chronic lung disease, and chronic heart disease) in a post hoc analysis. 

The trial was approved by the Human Research Protection Office of the University of Pittsburgh. The methods have been published previously

Eligible primary care family medicine (FM) and internal medicine (IM) practices from a practice-based research network in Pittsburgh (FM PittNet), a clinical network in southwestern Pennsylvania (Community Medicine, Inc), and a safety-net clinical network in Houston were solicited for participation. When 25 sites (a sufficient number per sample size calculations for an RCCT) had agreed to participate, solicitation ceased. All sites used a common electronic health record (EHR), EpicCare. Eligibility requirements included having at least 100 patients 18 years or older, preliminary baseline vaccination rates less than 50% for at least 1 adult vaccine (influenza, pneumococcal, Tdap), and a willingness to make office changes to increase vaccination rates. Participating practices were stratified by location (urban, suburban, or rural) and discipline (FM or IM), then randomized. The practices in this analysis were the 18 private practices or residency sites in southwestern Pennsylvania and did not include 1 site in Pittsburgh, which dropped out, and 6 publicly funded practices in Houston, from which data on high-risk conditions were not available. 

 key domains: Pillar 1: convenient vaccination services; Pillar 2: communication with patients about the importance of immunization and the availability of vaccines; Pillar 3: enhanced office systems to facilitate immunization; and Pillar 4: motivation through an office immunization champion (IC). The 4 Pillars Program includes background on the importance of protecting patients against vaccine-preventable diseases, barriers to increasing vaccination from both provider and patient perspectives, and strategies to eliminate those barriers. Practices were expected to implement strategies from each of the 4 pillars. 

The intervention was designed using the diffusion of innovations theory

 and included the 4 Pillars Program, provider education, and 1-on-1 coaching of the IC for each practice. The IC was responsible for using the 4 Pillars Program to guide the practice’s intervention activities, participating in the biweekly telephone call with a research liaison for coaching, ensuring that chosen strategies were being implemented, and working to maintain motivation of the staff. 

The overall study included a 2-year RCCT in which the year 1 controls were crossed over into active intervention and the year 1 intervention groups became maintenance groups after the first year.

 In this analysis, all patients from the 18 southwestern Pennsylvania sites were combined and vaccination among eligible high-risk patients was examined at the end of baseline (May 31, 2013) and the end of the intervention (January 31, 2015), at which time all sites had completed the intervention. The effects of the intervention among the types of high-risk conditions were compared in a post hoc analysis. 

De-identified demographic data (date of birth, sex, race, health insurance coverage as a proxy for income); office visit dates; 

) codes for high-risk conditions, including immune and autoimmune diseases, cancers, chronic kidney diseases, diabetes, chronic lung diseases, and chronic heart diseases (codes 42, 135, 141-208.91, 250.0-250.93, 279-279.9, 282.6-284, 288-288.2, 393-398.99, 402.0-404.93, 410-412, 141-141.9, 416-416.9, 428-428.9, 438-438.9, 446-446.7, 491-496, 500-505, 506.4, 506.9, 508-508.9, 510-510.9, 513-519.9, 571-572.8, 585-586, 710-710.9, and 714-714.9) (see 

]); and vaccination data (vaccines given and dates) were derived from de-identified EHR data extractions. A longitudinal database was created with only those patients who were aged 18 to 64 years at baseline and who had a visit each year during the study period, creating the cohort of individuals for study. 

Descriptive analyses were performed for patient demographic characteristics (age, sex, race, health insurance, high-risk condition). Age was used as a continuous variable, and racial groupings were non-Hispanic white and nonwhite. Patients with more than 1 of the 3 high-risk conditions (diabetes, chronic lung disease, chronic heart disease) were included in each of their respective disease groups for analysis. PPSV and Tdap would typically be administered once during the project period; thus, PPSV and Tdap rates are presented as cumulative rates at the end of baseline (May 31, 2013) and end of the intervention (January 31, 2015). For influenza vaccination, the analytical periods were June 1, 2012, to May 31, 2013, for baseline, and June 1, 2014, to January 31, 2015, for the intervention year. Proportions were reported for categorical variables, and means and standard deviations were reported for continuous variables. The primary outcome measures were the cumulative PPSV and Tdap vaccination rates, influenza vaccination rates reported at the end of baseline and the end of the intervention, and PP differences. Student’s paired 

 tests were performed to test for 2-year differences in influenza vaccination rates and cumulative PPSV and Tdap vaccination rates. In addition, the weighted average vaccination rates were compared between high-risk conditions for each vaccine using Student’s 

Multilevel generalized estimating equation modeling, which accounts for the clustered nature of the data (ie, patients are clustered within practices), was conducted using vaccination status for each vaccine as the binary outcome variable. Those who received the PPSV or Tdap vaccine prior to the trial were excluded from the regression analyses. To determine which factors were related to PPSV, Tdap, and influenza vaccine uptake, the regression models also accounted for heterogeneity in demographic characteristics, including age, sex, race, and health insurance. Statistical significance of 2-sided tests was set at a type I error (alpha) equal to 0.05. All analytical procedures were performed using SAS version 9.4 (SAS Institute; Cary, North Carolina). 

Among the 4737 patients aged 18 to 64 years who had a high-risk condition, the average age was 52.1 ± 10.2 years, with 54.2% female patients, 8.2% nonwhite patients, and 65.4% who were privately insured (data not shown). In this cohort, 42.2% of patients had diabetes, 35.5% had chronic lung disease, 13.9% had chronic heart disease, and 16.1% had another high-risk condition. Overall, 366 (7.7%) had 2 or more high-risk conditions. 

Cumulative PPSV uptake at the end of intervention reached 55.7% for all high-risk patients. Specifically, 59% of those with chronic heart disease, 54% with chronic lung disease, 66% with diabetes, and 39% with another high-risk condition had received PPSV by the end of the intervention (

). Overall cumulative pneumococcal vaccination rates significantly increased 12.2 PP from baseline; patients with diabetes had larger increases than those with chronic lung disease (

= .032), or another high-risk condition (

= .009). Cumulative Tdap vaccination rates increased significantly for all high-risk patients by 11.4 PP from baseline, reaching nearly 50% at the end of the intervention. Vaccination rates for the various high-risk groups ranged from 46% to 51%. Only those with other high-risk conditions increased their rates significantly more than those with diabetes (12.7 PP vs 11.3 PP, respectively; 

= .04). Annual influenza vaccination also increased significantly from baseline for those with diabetes, chronic lung disease, and other high-risk conditions, reaching 57% for all high-risk patients. There were no differences among high-risk groups for PP increases in influenza vaccination rates.

), 2060 patients who had received PPSV before the study began (June 1, 2012) were excluded from the PPSV regression model; similarly, 1796 patients who had received the Tdap vaccine before the study began were excluded from the Tdap vaccine regression model. The odds of pneumococcal vaccination were significantly associated with older age (odds ratio [OR], 1.02; 95% CI, 1.01-1.02), white race (OR, 1.45; 95% CI, 1.02-2.06), and having diabetes (OR, 2.22; 95% CI, 1.80-2.73), chronic lung disease (OR, 1.50; 95% CI, 1.21-1.87), and chronic heart disease (OR, 1.32; 95% CI, 1.02-1.71). The odds of Tdap vaccination were significantly inversely associated with being female (OR, 0.83, 95% CI, 0.70-0.99; reference = males). The odds of influenza vaccination were associated with being female (OR, 1.24; 95% CI, 1.12-1.37), with being older (OR, 1.03; 95% CI, 1.03-1.04), with having diabetes (OR, 1.14; 95% CI, 1.01-1.28), and with having chronic lung disease (OR, 1.14; 95% CI, 1.01-1.30). 

With a concerted effort, primary care practices were capable of modifying their offices’ systems to significantly improve vaccination rates from baseline levels among high-risk adults younger than 65 years. For pneumococcal vaccine, these results (56%) are in stark contrast to the 2014 national rate of 20%,

 and among those with diabetes (66%), the rate surpasses the national goal of 60%.

 Moreover, the improvement of 12.2 PP is notably higher than secular trends of less than 2 PP per year recently observed among adults aged 19 to 64 years with high-risk conditions.

 Female sex, older age, and white race were related to higher likelihood of receipt of PPSV, similar to recent national data that indicate significantly lower rates among nonwhites compared with whites

 and higher rates among older than younger individuals.

 In this study, those with diabetes, chronic lung disease, or chronic heart disease were more likely to receive PPSV than patients without each respective high-risk condition. The risk of pneumococcal disease is increased for all 3 of these comorbidities

; thus, it is important to know if an intervention shown to be effective among all adults is similarly effective among high-risk adults or if a special intervention is necessary. These data indicate that high-risk adults do not require a separate intervention, as their increases in PPSV uptake approached increases reported in a study of all adults.

Tdap vaccine uptake also increased significantly from baseline (by 11.4 PP to 49.4%). These values exceeded the 2015 national rate (20.1%), recent secular trends of 3 PP per year increased uptake for all adults older than 19 years,

 and the increases among all adults (6.2 PP) shown in a previous study.

 Interestingly, in this study, men with high-risk conditions were more likely to receive the Tdap vaccine, whereas increased rates among women might be expected given the recommendation for pregnant women

 and others who care for infants to receive the Tdap vaccine. Influenza vaccination increased significantly from baseline (3.1-5.9 PP) for those with any high-risk condition. Those with diabetes and those with chronic lung disease were more likely to have received the influenza vaccine compared with those without these conditions, whereas those with chronic heart disease were not more likely to be vaccinated against influenza than those without. Influenza vaccination rates for all groups were still considerably below Healthy People 2020 goals of 70%,

 a troubling finding given their high risk of influenza complications. 

Barriers to adult vaccination include patient, provider, and health system issues, such as lack of awareness of the need for vaccination, competing priorities for the physician, and incomplete documentation of vaccination history.

 The Task Force on Community Preventive Services recommends provider reminders and a combination of interventions to increase vaccination coverage among high-risk adults.

 The 4 Pillars Program offers strategies to address each of these types of barriers, including assessing and communicating the need for vaccination by all members of the clinical staff, implementing best practice alerts in the EHR or other reminders to providers, offering simultaneous vaccination with other indicated vaccines, and using standing order protocols. In an RCCT, the 4 Pillars Program demonstrated modest improvements in vaccination rates for all 3 vaccines among all adults.

 Other studies have used similar multifaceted approaches to increasing pneumococcal and influenza vaccination

The pneumococcal vaccine is recommended for cigarette smokers.

 We did not specifically include smokers without high-risk medical conditions and therefore do not know how their inclusion would have changed the vaccination estimates. The completeness and accuracy of 

 coding was not verified, although EHRs were used. Other records (eg, pharmaceuticals as a proxy for diagnoses) were not evaluated to confirm or augment 

 codes were not done due to funding and time limitations. The population is limited to the greater Pittsburgh region and may not be generalizable to other populations. This is a post hoc analysis derived from an RCCT. The primary purpose of the analysis was to compare the effect of the intervention on groups of adults with common high-risk conditions rather than demonstrate its effectiveness against no program; hence, before-and-after analyses were conducted. 

An intervention including the 4 Pillars Program, staff education, and support for a practice-based IC was associated with significant increases in PPSV, Tdap, and influenza vaccination among high-risk adults aged 18 to 64 years over a 2-year study. These findings further support the use of evidence-based strategies as part of a comprehensive, practice-based effort to address low vaccination rates among adults with high-risk medical conditions. Providers should be aware that the systems that are being successfully used to improve vaccination of non—high-risk patients may be equally effective for vaccinating patients with high-risk conditions.

 Department of Family Medicine, University of Pittsburgh School of Medicine (MPN, KKM, SZ, JMR, RKZ, CJL), Pittsburgh, PA.

 This investigation was supported by a grant (U01 IP000662) from the CDC. The views expressed herein are those of those authors and not those of the CDC. The project described was also supported by the National Institutes of Health through Grant Numbers UL1 RR024153 and UL1TR000005.

 Dr Nowalk, Ms Zhang, and Mr Raviotta have received grants from Pfizer Inc and Merck & Co. Drs Zimmerman and Lin and Ms Moehling have received grants from Pfizer, Merck, and Sanofi. 

 Concept and design (MPN, JMR, RKZ); acquisition of data (MPN, KKM, SZ, JMR, RKZ); analysis and interpretation of data (MPN, SZ, RKZ, CJL); drafting of the manuscript (MPN, CJL); critical revision of the manuscript for important intellectual content (MPN, KKM, JMR, RKZ, CJL); statistical analysis (SZ, CJL); provision of patients or study materials (JMR); obtaining funding (RKZ); administrative, technical, or logistic support (KKM, JMR, RKZ); and supervision (MPN, RKZ). 

 Mary Patricia Nowalk, PhD, RD, University of Pittsburgh School of Medicine, 4420 Bayard St, Ste 520, Pittsburgh, PA 15260. E-mail: tnowalk@pitt.edu.

1. Grohskopf LA, Sokolow LZ, Olsen SJ, Bresee JS, Broder KR, Karron RA. Prevention and control of influenza with vaccines: recommendations of the Advisory Committee on Immunization Practices, United States, 2015-16 influenza season. 

3. Shea KM, Edelsberg J, Weycker D, Farkouh RA, Strutton DR, Pelton SI. Rates of pneumococcal disease in adults with chronic medical conditions. 

2014;1(1):ofu024. doi: 10.1093/ofid/ofu024.

4. Williams WW, Lu PJ, O’Halloran A, et al; CDC. Surveillance of vaccination coverage among adult populations—United States, 2014. 

2016;65(1):1-36. doi: 10.15585/mmwr.ss6501a1.

5. Immunization and infectious diseases. HealthyPeople.gov website. healthypeople.gov/2020/topics-objectives/topic/immunization-and-infectious-diseases?topicid=23. Published 2013. Accessed January 18, 2013. 

6. Lu PJ, O’Halloran A, Williams WW, Lindley MC, Farrall S, Bridges CB. Racial and ethnic disparities in vaccination coverage among adult populations in the U.S. 

7. Lu PJ, O’Halloran A, Williams WW. Impact of health insurance status on vaccination coverage among adult populations. 

2015;48(6):647-661. doi: 10.1016/j.amepre.2014.12.008.

8. Lu PJ, O’Halloran A, Ding H, Srivastav A, Williams WW. Uptake of influenza vaccination and missed opportunities among adults with high-risk conditions, United States, 2013. 

. 2016;129(6):636.e1-636.e11. doi: 10.1016/j.amjmed.2015.10.031.

Articles

CLINICAL

Pneumococcal; tetanus, diphtheria, and pertussis; and influenza vaccination increased among high-risk adults in a 2-year study.

Using the 4 Pillars to Increase Vaccination Among High-Risk Adults: Who Benefits?

 Alternative payment models, such as accountable care organizations, use financial incentives as levers for change to facilitate the transition from volume to value. However, implementation raises concerns about adverse changes in market competition and the resultant physician response. We sought to identify physician characteristics and market-level factors associated with variation in response to financial incentives for cancer care that may ultimately be leveraged in risk-shared payment models.

 Retrospective cohort study of physicians providing minimally invasive bladder cancer procedures to fee-for-service Medicare beneficiaries.

 We examined the relationship of between-group differences in market-level factors (competition [Herfindahl-Hirschman Index (HHI)] and provider density) and physician-level factors (use of unique billing codes, number of billing codes per patient, and competing financial interest) to responsiveness to financial incentives.

Incentive-responsive providers had increased odds (odds ratio [OR], 1.19; 95% CI, 1.04-1.35) of practicing in markets with the highest quartile of provider density but not HHI (OR, 0.96; 95% CI, 0.87-1.05). Incentive-responsive providers were more likely to bill in the highest quartile for unique codes (OR, 1.49; 95% CI, 1.32-1.69) and codes per patient (OR, 1.18; 95% CI, 1.11-1.25) and less likely to have a competing financial interest (OR, 0.76; 95% CI, 0.72-0.81).

 Responsiveness to financial incentives in cancer care is associated with high market provider density, profit-maximizing billing behavior, and lack of competing financial ownership interests. Identifying physicians and markets responsive to financial incentives may ultimately promote the successful implementation of alternative payment models in cancer care. 

Responsiveness to financial incentives in cancer care is associated with high market provider density, profit-maximizing billing behavior, and lack of competing financial ownership interests. 

Responsiveness to financial incentives is not associated with high levels of market competition. 

Identifying physicians and markets with this phenotype of responsiveness to financial incentives may ultimately be leveraged in identifying physicians and markets well suited to participate in and therefore optimize the successful implementation of alternative payment models.

Regionalization of healthcare is increasing as independent practices join hospitals and healthcare networks and larger groups and younger physicians favor employed positions.

 Furthermore, concentration of cancer care is advocated by several studies to improve medical and surgical management of multiple disease states.

 These changes have the potential to reduce variation in care delivery and increase care coordination, resulting in reductions in unnecessary spending and ultimately improved outcomes.

 Yet, such benefits have been balanced by concerns surrounding potential increases in cost associated with reductions in competition.

Patterns of cancer care delivery remain heterogeneous with inconsistent reaction to market forces.

 Although observed variation in cancer care is due, in part, to differences in patient and disease characteristics, there remains considerable unexplained variation at the physician level.

 The complex relationship between physician behavior and loco-regional market forces remains poorly characterized. 

Treatment of bladder cancer incurs remarkable costs and exhibits significant practice heterogeneity.

 Previous work has evaluated changes in diagnostic and treatment patterns of bladder cancer as a result of changes to physician payments in the Medicare fee schedule and demonstrated inefficient and costly alterations in practice.

 In brief, this work found that, despite payment incentives with strong face validity to shift bladder cancer care from high-cost facility-based locations (operating room or ambulatory surgery center) to physician offices, providers increased the total number of procedures performed as opposed to substituting site of service to optimize the value of bladder cancer care delivery. Additionally, this work demonstrated an increase in procedural redundancy and decreased diagnostic yield. 

We observed considerable variation in physician response to this financial incentive. To this end, in the current analysis, we sought to determine physician characteristics and market-level factors associated with responsiveness to financial incentives. We hypothesized that readily available physician- and market-level data would permit identification of a specialist physician phenotype particularly responsive to financial incentives that may ultimately be leveraged in identifying physicians and markets that may be particularly well suited to participate in early risk-bearing payment models. 

As described elsewhere, we used a 5% Medicare sample from 2001 through 2013 to identify beneficiaries enrolled in fee-for-service (FFS) Medicare Parts A and B who underwent a minor cystoscopic procedure associated with the diagnosis and/or management of bladder cancer (Current Procedural Technology [CPT] procedure codes 52204, 52214, 52224). There was a large increase in professional payment for these procedures beginning in 2005 when performed in an office-based location.

Performance of this group of minor cystoscopic procedures requires minimal additional equipment and expertise to perform in the office than is already established in most urology practices. Further, there are few (if any) contraindications to performing these procedures in the office rather than in dedicated operating rooms or surgical centers. 

We then identified individual providers who performed 1 or more of these procedures using their National Provider Identifier (NPI). We defined 2 groups of urologists, the first being those who did not change practice pattern after implementation of the change in Medicare fee schedule and the second being the group that increased the use of office-based procedures in response to the positive change in reimbursement.

We then linked the provider-level data from our index analysis to the 2013 National Downloadable File, a dataset maintained by CMS that includes years since graduation from medical school and gender.

 This file also contains data on firm affiliation, including number of physicians (regardless of specialty) within a given practice. Providers were excluded if their date of graduation was after 2008 to omit physicians that were highly likely to be in ongoing resident training. Provider location information was obtained from the NPI Dissemination File, which is a dataset also maintained by CMS and updated weekly with all known provider locations.

Proportion of effort was equally divided between each associated firm for providers listed as participating in multiple firms. 

We calculated 2 measures of loco-regional markets using zip codes to assign hospital referral regions (HRRs). HRRs were defined by the Dartmouth Institute of Health Policy and Clinical Practice and represent regional healthcare markets for tertiary medical care.

 Urologist density was calculated as the number of urologists per 10,000 Medicare beneficiaries within each HRR. Herfindahl-Hirschman Index (HHI) is a widely used measure of market competition and is the sum of the squared market shares of all the firms within the market multiplied by 10,000. We used the number of urologists in each firm as a measure of market share and HRR to define a market. Values of 10,000 represent a monopoly market, with values less than 2500 considered competitive markets.

Provider billing practice data was obtained from the publically available 2012 Medicare Provider Utilization and Payment Data. Using this dataset, we calculated 2 measures of billing practice as a representation of efficient coding behavior. First, we calculated the number of unique Healthcare Common Procedure Coding System (HCPCS) codes billed to Medicare during that year by each provider. Second, we calculated the number of codes billed per beneficiary. 

The structure of the data in this set does not allow for calculating the precise number of codes billed to CMS per beneficiary per visit. To estimate the number of unique codes per visit, we summed the number of times each HCPCS code was billed and divided this by the sum of the number of patients for which each HCPCS code was used. However, the denominator of this calculation contains duplicate counts for patients rather than counts of unique patients. As a result, this estimate will be conservative and should not be used as an assessment of the absolute number of codes per patient but in relative terms for between-group comparisons.

Some urologists have an ownership stake in a facility (ie, ambulatory surgery center) and may preferentially perform minor procedures in a facility that could otherwise be done in the office to capture not only the professional fee but also the facility fee associated with these procedures. Simple cystoscopy (CPT 52000) is one of the most common procedures performed by urologists and is most often conducted in the office. To account for this competing motivation for performing procedures in the facility regardless of the financial incentive associated with office-based management of bladder cancer, we identified the proportion of simple cystoscopy procedures performed by each urologist in the office compared with in the facility.

Data from the 2013 American Community Survey were used to construct additional covariates for HRR-level attributes, including median household income, race, and educational attainment.

Providers who increased the number of office-based minor cystoscopic procedures compared with the number of facility-based minor cystoscopic procedures after 2005 were considered incentive-responsive, and those who either continued to do the same proportion or decreased the proportion of office-based minor cystoscopic procedures compared with facility-based were considered incentive resistant. Differences in between-group provider, practice, and patient characteristics were assessed using the Wilcoxon rank sum and Pearson χ

The primary exposures of interest to explain provider response to the financial incentive were between-group differences in coding practice and market measures. Unadjusted comparisons were made using nonparametric statistics, and logistic regression models were used to adjust for patient (age, comorbidities, gender, and race), firm (number of physicians and number of urologists), and HRR (race, household income, and education) characteristics. Odds ratios were expressed as interquartile range values that reflect odds for the 75th percentile compared with the 25th percentile for each continuous exposure variable of interest.

Data management and statistical analyses were performed using R version 3.1.2 (The R Foundation; Vienna, Austria). 

 values below .05 were considered statistically significant. This study was approved by the institutional review board at Vanderbilt University.

Incentive-responsive urologists were more concentrated in the eastern United States but with other focused pockets throughout the country (

), and they practiced in smaller firms both in terms of number of total physicians and number of urologists (

). Although incentive-resistant providers cared for patients with a slightly lower number of comorbid conditions, this observed difference may reflect between-group variation in coding practices rather than management of more complex patients. There were no observed between-group differences in provider gender, time since graduation from medical school, or patient age, gender, and race. 

Loco-regional market characteristics, differences in facility-based preference for performing procedures, and measures of billing practices are summarized in 

. Incentive-responsive providers used more unique billing codes and more codes per beneficiary than incentive-resistant providers. They were also less likely to have a facility-based preference for performing cystoscopy, included as a proxy for competing financial interest. Incentive-resistant providers practiced in markets with both lower urologist density and competition (indicated by higher HHI). Of note, the majority of urologists practiced in areas with higher levels of competition, with the upper quartile in each group less than 2500.

After adjusting for patient, firm, and HRR characteristics and the other exposures of interest, there was no observed relationship between HHI and incentive-responsiveness. However, urologists in HRRs at the 75th percentile of provider density had a 19% (95% CI, 4%-35%) increased odds of incentive-responsiveness compared with those practicing in HRRs in the 25th percentile of provider density (

). Together, these data suggest that the impact of the market share of firms has less of an influence on individual provider response to incentives in FFS cancer care than does the density of providers in that market.

Providers who bill at the 75th percentile for number of unique codes had a 49% (CI, 32%-69%) increased odds of being incentive-responsive compared with providers at the 25th percentile. A higher number of billing codes

per patient was also associated with incentive-responsive behavior. These findings suggest that providers more facile with billing practices were more likely to respond to financial incentives. Incentive-responsive providers were also less likely to prefer performing cystoscopies in a facility, suggesting that a competing interest in opposition to the studied financial incentive determined physician behavior.

This study evaluated physician- and market-level characteristics associated with adoption of a financially incentivized change in practice for the diagnosis and management of bladder cancer. Several important relationships were identified that might characterize a phenotype for incentive responsiveness: providers in smaller groups, practicing in dense markets, who are also facile with billing and without competing financial ownership interests. 

The influence of firm size may reflect several possible phenomena. First, small firms are increasingly integrating with larger firms, and incentive-responsive behavior may be a resistive attempt to increase or maintain revenue in the face of consolidation pressures. Second, larger firms may be more likely to have an ownership stake in a facility that may mitigate the net benefit associated with the evaluated modification in the Medicare physician fee schedule. Finally, smaller organizations may be more nimble and adaptable to changes in price structure to maximize profitability.

Market forces appear to have a significant relationship with providers’ responses to changes in professional payment. Markets with a higher density of providers were associated with increased likelihood of incentive-responsive behavior, suggesting that providers in these markets may face disproportionate pressures to modify practice in order to maximize profit. Alternatively, denser provider markets may lead to more physician awareness of incentives.

Although unadjusted analyses suggest a relationship between HHI and incentive responsiveness, this relationship was not maintained after adjustment for patient, practice, and market characteristics including, most notably, provider density. This suggests that providers may more readily respond to financial incentives in highly dense markets irrespective of the structure of the firms within that market. The negotiating power of physicians in markets with higher HHI (ie, less competition) has been shown to result in leveraging increased payment from commercial insurers

; however, the payment set by CMS is not negotiable based upon firm size or market share. As a result, CMS may be shielded from this potential unintended consequence of novel payment models that foster healthcare concentration and regionalization. At the same time, providers in highly dense markets may be better suited to respond to changes in the realignment of incentives of novel payment models. In addition, price is not the only factor driving healthcare value. It is possible that providers working closely together to coordinate care can reduce overall spending despite lower competition, even if such declines in competition contribute to higher commercial prices.

Incentive-responsiveness was also associated with increased provider billing efficiency, as evidenced by utilization of a higher number of unique billing codes and more codes per beneficiary. This increase in billing efficiency, whether appropriate or not, as well as responsiveness to financial incentives may be a reaction to counter the pressure of lower prices in competitive markets.

This work has several limitations to consider. The primary data source regarding provider behavior was claims data, and we are unable to confirm important patient and disease characteristics from the medical record. Whether a given provider responded to incentives was determined by utilizing claims data spanning more than a decade (2001-2013), and data regarding loco-regional market and provider characteristics were based on data from 2011 to 2013. Therefore, we cannot determine the effects of changes in market and provider characteristics (eg, acquisition or relocation of physician firms) and assume that such events are random between groups. Finally, the Medicare claims data represents a 5% sample of patients, and the unit of analysis in this study was physicians. We cannot exclude nonrandom effects from this methodological consideration in sampling.

Managing bladder cancer through office-based, rather than facility-based, procedures should intuitively be in the interest of payers (more cost effective than facility-based management), physicians (more time efficient), and patients (more time efficient and cost effective), but in a FFS environment, payment incentives to adopt this practice have resulted in the delivery of low-value redundant cancer care. Our study results suggest that providers who adopt this practice may be influenced to do so by market forces and that they may also be inclined to implement other profit-maximizing behaviors. Approaches to counter this behavior might be to enact rules to govern the appropriateness of office-based procedures, commit significant resources to investigate and enforce laws against upcoding practices, or enact policies that reduce the variation in concentration of physicians.

approach will always face an uphill battle against the incentives in FFS medicine that reward such practices. Instead, adopting a payment model where physicians are paid for providing high-quality care rather than for the volume of billed procedures may result in better aligned incentives for all involved parties. While this is the direction cancer care is moving with the Oncology Care Model, surgeon behavior is largely excluded from such realignment of incentives.

That a large proportion of providers did not respond to this financial incentive with such strong face validity is notable. What is it about providers and firms that they did not or were not able to respond to this incentive that essentially left a large financial windfall on the table for taking care of patients with cancer? Are providers who are resistant to incentives simply unaffected due to status quo bias (ie, a preference for persisting in the current state of behavior) or are there alternative forces at play?

 It may be that identifying an incentive-responsive phenotype can be leveraged to provide high-quality, rather than low-quality, care. Recent work suggests that financial incentives for providers can be used to improve cardiovascular risk.

 Would similar results have been found selecting providers identified in this study as incentive-resistant? With Medicare seeking to tie payment to alternative structures instead of FFS,

 identifying markets and providers that remain responsive to incentives may be a crucial step to ensure provider behavior follows intended novel payment schemes. Or perhaps institutions might look to markets, firms, or providers that promote or match this incentive-responsive phenotype for clues to remain viable in the shifting sands of payment reform.

Department of Surgery, Division of Urology, Huntsman Cancer Institute, University of Utah (BO), Salt Lake City, UT; Department of Urology, Mayo Clinic-Scottsdale (MT), Scottsdale, AZ; Department of Urologic Surgery, Vanderbilt University (AJG, DAB, SSC, DFP, MJR), Nashville, TN; Geriatric Research Education and Clinical Center, VA Tennessee Valley Healthcare System (DFP, MJR), Nashville, TN.

 The project was supported by CTSA award No. UL1TR000445 from the National Center for Advancing Translational Sciences. Dr Resnick, the senior author, was also supported by the American Cancer Society (MSRG-15-103-01-CPHPS).

 The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article. 

 Concept and design (BO, MJR); acquisition of data (BO, AJG, DFP, MJR); analysis and interpretation of data (BO, MT, AJG, DAB, DFP, MJR); drafting of the manuscript (BO, MT, AJG, DAB, SSC, DFP, MJR); critical revision of the manuscript for important intellectual content (DAB, SSC); statistical analysis (BO, AJG, MJR); obtaining funding (BO, MJR); and supervision (MT, SSC, DFP, MJR). 

 Brock O’Neil, MD, 1950 Circle of Hope, N6405, Salt Lake City, UT 84112. E-mail: brock.oneil@hci.utah.edu. 

1. Liebhaber A, Grossman JM. Physicians moving to mid-sized, single-specialty practices. 

2. Welch WP, Cuellar AE, Stearns SC, Bindman AB. Proportion of physicians in large group practices continued to grow in 2009-11. 

. 2013;32(9):1659-1666. doi: 10.1377/hlthaff.2012.1256.

We determine a specialist physician phenotype responsive to financial incentives that may be leveraged to identify physicians and markets well-suited for participation in alternative payment models.

The Influence of Provider Characteristics and Market Forces on Response to Financial Incentives

 To assess the impact of the Affordable Care Act dependent coverage provision on outpatient prescription expenditures among young adults and to characterize medication types that are prescribed for young adults. 

 Using data from the Medical Expenditure Panel Survey from 2007 to 2009 and 2011 to 2013, difference-in-differences (DID) analyses estimated the provision’s impact among young adults aged 19 to 25 years versus those aged 26 to 34 years. Five outcomes related to prescription medication access and utilization were evaluated, including whether young adults were unable to get necessary medications in the past year due to cost and whether there were changes in total, out-of-pocket (OOP), and private expenditures and the share of total expenditures paid OOP. DID analyses were also carried out for medication expenditures by class. 

 There were a total of 19,165 young adults aged 19 to 25 years and 23,892 aged 26 to 34 years. After adjusting for demographic and socioeconomic factors, provision implementation (DID coefficient 

 ≤.05) was associated with an increase in total expenditures (by 3.8 percentage points), including those paid by private insurance (7.6 percentage points), and decreases in OOP expenditures and the share of total expenditures paid OOP (4.4 and 1.2 percentage points, respectively). Commonly used medications were anti-infectives, central nervous system agents, and hormones. Expenditures significantly increased for anti-infectives and decreased for hormones and psychotherapeutics. 

 The dependent coverage provision was associated with an increase in prescription medication expenditures, especially for anti-infectives, among young adults. The amount of expenditures paid by private insurers increased, whereas they decreased for OOP spending. 

The dependent coverage provision was associated with increased prescription total expenditures by 3.8 percentage points among young adults aged 19 to 25 years in the 3-year time period following implementation versus prior. This included an increase in expenditures paid by private insurance (7.6 percentage points) and decreases in out-of-pocket (OOP) expenditures (4.4 percentage points) and share of total expenditures paid OOP (1.2 percentage points). 

Provision effects were especially noted among young adults with higher household income levels. For total expenditures, there was a significant 11.3 percentage-point increase among young adults in the highest income group (≥300% of federal poverty level [FPL]), while significant decreases of 13.2 and 13.0 percentage points were noted among the ≤100% and 101% to 299% of FPL groups, respectively. Among young adults, 6 therapeutic classes composed more than 80% of all outpatient prescription medications: anti-infectives; central nervous system agents; and hormone, respiratory, psychotherapeutic, and topical medications.

Among the major objectives of the Affordable Care Act (ACA) legislation, passed in March 2010, was greater access to healthcare services through increasing insurance coverage and decreasing the number of uninsured Americans—estimated to be more than 41 million before enactment.

 The consequences of being uninsured are well documented and, besides financial considerations, include delays in care for serious conditions and receiving fewer preventive services.

 Greater access was accomplished through several provisions, including expansion of Medicaid and establishment of insurance marketplaces. Another strategy was the dependent coverage provision that specified that young adults, who had a high uninsured rate prior to the ACA,

 were eligible to stay on their parents’ plans until the start of the first plan year after they turn 26, regardless of residency, marital status, and financial dependency, effective September 23, 2010. 

Previously, investigators showed that the provision led to an increase in young adults who were insured, had a primary care doctor, and had coverage for emergency care.

 Others have documented the provision’s spillover effects in terms of increasing dental insurance rates.

 Chua and colleagues evaluated the provision’s effects on medical spending, healthcare use, and overall health.

 They found significant increases in the probabilities of insurance coverage and the reporting of excellent physical and mental health and a significant decrease in out-of-pocket (OOP) expenditures among young adults aged 19 to 25 years in the first whole-implementation year, 2011, compared with a group aged 26 to 34 years. However, they did not find an increase in healthcare utilization

in the form of primary care, emergency department, or hospital visits, nor changes in prescription medications filled. Look et al

 evaluated the ACA's impact on health and medication insurance coverage, medication utilization, and expenditures among young adults. They found that in 2011, compared with a control group, health and medication coverage increased by 4.9 and 5.5 percentage points, respectively, among young adults. There was no change in total medication utilization as measured by fills, although there was a 30% decrease in OOP medication expenditures. Shane et al evaluated changes in health insurance and prescription medication utilization through 2012 and found increases in insurance coverage among young adults, but no significant changes in fills in 2012 compared with 2011.

Using a nationally representative sample, the Medical Expenditure Panel Survey (MEPS), this study builds on the work of others

in evaluating changes in medication utilization among young adults in the longer term. Specifically, using a difference-in-differences (DID) approach, this study assessed the ACA’s impact on young adult healthcare services utilization regarding facets of outpatient medication utilization, including overall expenditures, through 2013. MEPS includes data regarding the medication name and expenditure amounts by payer, offering a dataset well suited to investigate this topic. Further, as there are few data that assess medication use patterns among young adults, a secondary aim was to examine the types that are commonly prescribed and whether or not expenditures by type were affected by the ACA. 

The data source for this study was the Household Component Public Use Files of MEPS, a nationally representative sample of the noninstitutionalized civilian population. In 2013, close to 14,000 households were interviewed, representing 35,068 individuals.

 The MEPS data include healthcare utilization, expenditures, medications, insurance coverage, and demographic and health characteristics.

This was a retrospective analysis that used data pooled from 2007 to 2009 and from 2011 to 2013 representing pre- and postimplementation periods of the dependent coverage provision, respectively. 2010 was not used because the dependent coverage provision was implemented during this year. Young adults aged 19 to 25 years served as the target group, whereas those aged 26 to 34 years were the comparison group, as defined previously.

 This age group is close in age to the target group and presumably similar on aspects related to health utilization, but is not affected by the provision. 

Prescription Medicine Utilization and Expenditures

During each survey round, respondents were asked about outpatient medications they obtained, both new and refills; this information is included in the Prescribed Medicines file. Data collected included medication name, quantity dispensed, and expenditure amounts. Information was verified by pharmacy providers (provided that interviewees give permission to contact their pharmacies and the pharmacies respond to requests; about three-fourths of pharmacies responded in 2011) who relay information concerning medication fill/refill dates, the National Drug Code, medication name, strength and quantity dispensed, total expenditures, and payment sources.

 For nonresponses, expenditures were imputed from pharmacy data for another person’s obtainment of the same medication. Using a Generic Product Identifier code, MEPS coders classified medications into 16 major categories according to Lexicon Plus (Cerner Multum Inc; Denver, Colorado):

anti-infective, antineoplastic, central nervous system (CNS), hormonal, topical, cardiovascular, gastrointestinal, respiratory, nutritional, metabolic, psychotherapeutic, genitourinary, coagulation, immunologic, alternative, and miscellaneous agents.

The MEPS medication data have been validated. When MEPS data were matched with Medicare Part D claims data to validate prescription data, it was found that household respondents were good at reporting the total number of fills/refills and that they overreported the number of fills per drug, but underreported the number of drugs filled.

 Concordance between sources was greater for chronic as opposed to intermittently used medications, such as anti-infective and pain medications. 

There were 5 dependent variables related to medications. These included 1 binary (yes/no) variable related to medication access (whether or not there was a time in the past year that they were unable to get necessary prescribed medications due to cost) and 4 continuous variables related to utilization: total prescription expenditures, total paid by self/family (OOP), total paid by private insurance, and share of total expenditures paid OOP. 

DID methods tested the impact of the dependent coverage provision. Specifically, a multivariate DID model was used:

(group) × (time period) + other covariates

 group represents the target group, young adults aged 19 to 25 years, and B

, the postpolicy period, 2011 to 2013. The B

 coefficient is the output of interest, representing an interaction term that captures the difference of the provision effect on the target and comparison groups. 

Other demographic and health status factors that may affect medication utilization were included: age (continuous), gender, race/ethnicity (white non-Hispanic, Hispanic, black non-Hispanic, Asian/Pacific Islander, or other), region (Northeast, Midwest, South, or West), marital status (married, single/never married, widowed, or divorced/separated), household income (poor: ≤100% of the federal poverty level [FPL]; near-poor to low income: 101%-299% of the FPL; or middle to high income: ≥300% of FPL), self-perceived health and mental health status (excellent, very good, good, fair, or poor), education (less than high school diploma, high school diploma or General Educational Development credential, some college, or degree), health insurance (none, private, or public), having usual source of healthcare (yes/no), having chronic conditions (yes/no; asthma, high blood pressure, high cholesterol, or arthritis), and employment status (yes/no). 

Separate models were constructed for the different outcomes. For expenditures, log transformed ordinary least squares regressions were conducted, adjusting for covariates. Analyses were also carried out in which individual postpolicy years (2011, 2012, 2013) were treated as an interaction term [(group) × (year)] to assess whether the ACA’s impact on outcomes studied increased, decreased, or leveled off throughout postpolicy years. To discern whether the ACA had a differential impact depending upon household income, the outcomes were stratified by income levels, as outlined previously, in additional models. Multivariate DID models were also carried out for expenditures by medication class. 

Descriptive analyses characterized medication use patterns by quantifying the proportion of different classes out of total medications prescribed to target and comparison group members before and after dependent coverage provision implementation. Chi-square and Student’s 

 tests were used, as appropriate, to ascertain for differences in demographic and socioeconomic factors between groups. Given that the members of the target population of the provision were those covered by private insurance, sensitivity analyses were carried out whereby all individuals who were publicly insured were removed from the sample to evaluate result robustness.

To obtain nationally representative estimates, appropriate survey-weighting procedures accounting for the MEPS national probability design were used, as well as procedures to generate robust standard errors and estimates to take into account the complex survey design. Expenditures were adjusted to 2013 US$ based on the MEPS Personal Healthcare Expenditure Component Index, as recommended when pooling prescription medication expenditures.

 Statistical tests were 2-tailed with alpha level of 0.05. 

For all years, there were 19,165 (weighted = 177,426,653) young adults aged 19 to 25 years and 23,892 (weighted = 221,590,555) aged 26 to 34 years. 

) displays population characteristics for both groups. There were several significant differences across marital status, education, and income groups. For example, there was a lower proportion of comparison group members who were single/never married (42%), who had less than a high school diploma (11%), and who had an income ≤100% FPL (14%) compared with target group members (85%, 22%, and 19%, respectively; 

 <.001). There were also differences across health measures, with more target individuals having self-perceived health and mental health statuses of “excellent” (42% and 52% vs 33% and 47%, respectively; 

 displays adjusted differences in outcomes between the groups (full regression results are available in 

). For total expenditures, by the DID estimate, there was a significant increase of 3.8 percentage points among young adults aged 19 to 25 years compared with those aged 26 to 34 years. There was also a significant increase in private expenditures (7.6 percentage points) among young adults compared with those aged 26 to 34 years. The provision was associated with significant decreases in OOP expenditures (4.4 percentage points) and in the share of total expenditures paid OOP (1.2 percentage points). No differences were found in regard to whether or not there was a time in the past year they were unable to get medications due to cost. 

The dependent coverage provision was associated with an increase in total and private expenditures and a decrease in out-of-pocket medication expenditures paid, especially among higher-income groups.

November 2017

Using the 4 Pillars to Increase Vaccination Among High-Risk Adults: Who Benefits?

The Influence of Provider Characteristics and Market Forces on Response to Financial Incentives

Impact of Health Reform on Young Adult Prescription Medication Utilization

Patients' Perspectives of Care Management: A Qualitative Study

Reframing the Unaffordability Debate: Patient Responsibility for Physician Care

Electronic Reminder's Role in Promoting Human Papillomavirus Vaccine Use

Improving Antibiotic Stewardship: A Stepped-Wedge Cluster Randomized Trial

Changes in Cardiovascular Care Provision After the Affordable Care Act

Diabetes Care Improvement in Pharmacist- Versus Nurse-Supported Patient-Centered Medical Homes

Validation of a Claims-Based Algorithm to Characterize Episodes of Care