David Grande, MD, MPA

 Adequate access to primary and dental care is essential for population health, and some state Medicaid programs have expanded insurance coverage for both. However, there are few data on new Medicaid enrollees’ ability to access services. We examined the relationship between provider supply and enrollees’ identification of usual sources of care.

 Between November 2015 and February 2016, we surveyed low-income adults newly insured through Medicaid in Philadelphia, Pennsylvania, to determine if they had a usual source of care. Additionally, we used geospatial methods to calculate adult population per provider ratios by Census tract for primary and dental care providers who accepted Medicaid patients, then identified low-supply clusters.

 We used multivariable logistic regression models to describe the odds of identifying usual sources of care based on being in low- or high-supply clusters, adjusting for patient demographics.

 Of 1000 contacted individuals, 312 completed the survey. Among respondents, 168 were previously uninsured and newly enrolled in Medicaid; 66.7% of this group identified a usual primary care provider and 42.3% identified a usual dental care provider. In adjusted analyses, individuals living in low- and high-supply areas had similar likelihoods of identifying a usual source of primary or dental care.

 Many new Medicaid enrollees did not have usual sources of primary or dental care, regardless of nearby provider supply. Efforts to understand what improves access or engagement in healthcare among Medicaid enrollees are critical after low-income adults gain insurance.

Even after low-income adults gained insurance, barriers to accessing care persisted.

Nearby provider supply did not affect identification of usual sources of primary or dental care among new Medicaid enrollees.

Further efforts to understand what improves access or engagement in healthcare among Medicaid enrollees are critical.

Recent expansions in Medicaid have renewed policy debates on how to improve access to primary and dental care among low-income populations. Access to primary care is essential because it provides an entry point to other forms of healthcare, delivers preventive services, and treats chronic diseases. Better primary care access is also associated with improved health outcomes at a lower cost, including reductions in heart disease and cancer mortality.

 Better dental care access is associated with lower rates of dental caries,

 and low-value emergency department utilization.

 Low-income populations in particular experience more access barriers to these 2 types of care than other populations.

The Affordable Care Act (ACA) expanded coverage for primary and dental care services for low-income populations. As of 2016, when this study was conducted, 31 states and the District of Columbia had expanded Medicaid through the ACA, and 23 states included preventive dental service coverage for adults.

 However, in states that have expanded coverage, newly insured Medicaid enrollees still face access barriers. The ACA Medicaid expansion focused on addressing affordability

; however, health reforms have not equally prioritized provider availability

The relationship between provider supply and the identification of a usual source of care among low-income populations for both primary and dental care is poorly understood. Therefore, the objective of this study was to examine the association between the supply of nearby Medicaid-accepting primary and dental care providers and the likelihood of identifying a usual source of care among newly enrolled Medicaid beneficiaries in an urban environment. Prior to our study, Pennsylvania began administering Medicaid to all of its expansion population through managed care organizations. In addition to covering primary care visits, these plans covered a limited set of dental visits for preventive, diagnostic, and minor restorative needs up to $1000.

 The population sampled resides in Philadelphia, a racially diverse city with a high density of providers and the fifth-largest population, but also the highest poverty rate among the 10 largest cities in the United States.

The University of Pennsylvania Institutional Review Board approved this study.

We surveyed new Medicaid enrollees between November 2015 and February 2016. We identified Medicaid applicants through a partnership with Benefits Data Trust (BDT), a Philadelphia-based nonprofit organization that facilitates access to comprehensive public benefits and provides Medicaid application assistance. We randomly selected 1000 individuals to participate in a mailed survey if they received assistance from BDT within the year prior to November 2015, were aged 18 to 64 years, were able to read English, and had a Philadelphia mailing address. We excluded individuals who had not yet enrolled in Medicaid or had insurance before enrolling to focus on responses from newly insured adults. Individuals first received a letter informing them of the upcoming survey. One week later, we mailed the survey, consent information, a prepaid return envelope, and a $2 cash participation incentive. Nonrespondents received up to 2 additional reminders, then were called by phone to complete the survey. Respondents received an additional $10 gift card.

We developed the survey assessing usual sources of care based on questions used by the Agency for Healthcare and Research and Quality

]). Specifically, the primary care question asked, “Is there a particular doctor’s office, clinic, health center, or other place that you usually go if you are sick or need advice about your health?” The dental care question asked, “A regular dentist is the one you would go to for check-ups and cleanings or when you have a cavity or tooth pain. Do you have a regular dentist or dental clinic?” The survey confirmed whether respondents had ultimately enrolled in Medicaid and captured demographic and self-reported health measures. We geocoded survey respondent addresses using ArcGIS

10.3 (Environmental Systems Research Institute; Redlands, California) to identify their Census tract (a close approximation to neighborhoods in Philadelphia), which is more relevant than larger geographic areas, such as zip code, for policy makers and planners in dense urban environments.

Primary Care and Dental Practice Database

We constructed a database of Medicaid-participating primary and dental care providers in and around Philadelphia. We aggregated providers within office locations for the purposes of geospatial analyses. Primary care providers included physicians, nurse practitioners, and physician assistants. As described in a prior study,

 we used SK&A (2014), a proprietary database, to identify all primary care providers in and near Philadelphia participating in Medicaid. We supplemented this database with provider directories from the Philadelphia-based Medicaid plans and public lists of federally qualified health centers. Practices were contacted by phone to verify their address, the number of practicing clinicians, and each clinician’s full-time equivalent (FTE) workload to calculate an aggregate FTE for each office. Dental care providers consisted of general practice dentists who treated adults. We utilized the 2014 American Dental Association masterfile as the initial file and supplemented it with the National Provider Identifier dentist registry and the Medicaid provider file from Pennsylvania’s Department of Human Services. We defined the number of dental FTEs in a dental office based on the number of offices for which a dentist works.

Census Tract Primary Care and Dental Care Provider Supply 

Using the practice file described previously and population density data from the American Community Survey, we calculated a Medicaid adult population per provider ratio for each Census tract using the 2-step floating catchment area (2SFCA) method to estimate the ratio of Medicaid-enrolled adults per Medicaid-participating provider based on a 5-minute travel time. The 2SFCA method accounts for the providers in and around Philadelphia-based Census tracts and the population around a provider office.

 In addition, the 2SFCA method helps account for the modifiable areal unit problem, or the error introduced into spatial analyses by drawing unit borders and by aggregating units.

 We used ArcGIS 10.3 to account for traffic history and street restrictions in order to accurately measure travel times.

We defined low-provider areas as those with 5 or more contiguous Census tracts in the lowest quintile of supply—the population per provider ratio within a 5-minute travel time of a Census tract.

 Our goal was to identify geographic clusters with lower supply (as opposed to isolated Census tracts that may be adjacent to higher supply areas) to more accurately reflect experiences of patients seeking usual sources of care within and outside of their neighborhood. We utilized a relative instead of absolute measure of supply because there is no agreed-upon definition of the ideal adult population per provider ratio in urban areas for either primary or dental care.

For all analyses, we evaluated primary and dental care access separately. We assessed whether living in a low-provider supply area for primary or dental care was associated with having a usual source of primary or dental care, respectively. We used multivariable logistic regression models clustered at the level of Census tracts adjusting for age, gender, race/ethnicity, self-rated health, educational attainment, and employment status. In an additional analysis, we used the aforementioned models but changed the exposure of primary care and dental care supply to continuous measures of adult population per provider ratios. As a sensitivity analysis, we used 10-minute travel times to define supply. A 2-sided 

<.05 was considered statistically significant. All analyses were carried out using Stata version 14.0 (StataCorp, LLP; College Station, Texas).

Of 1000 contacted individuals, 312 completed the survey. Among respondents, 168 reported being enrolled in Medicaid and were previously uninsured. These individuals (

) were predominantly female (60.1%), 45 years or older (67.9%), African American (65.5%), and non-Hispanic (86.3%). A total of 59.5% had a high school education or less, and few (35.1%) were employed full time or part time. The majority (78.0%) reported having good to excellent health.

Within the study cohort, 112 of 168 (66.7%) had a usual source of primary care and 71 of 168 (42.3%) had a usual source of dental care. Those with a usual source of primary and dental care did not differ significantly from those without a usual source based on their sociodemographic characteristics. However, those with a usual source of dental care were more likely to have better self-reported health (excellent, very good, good) compared with respondents without a usual source of dental care (63/71 [88.7%] vs 56/84 [66.7%], respectively; 

 = .001). For primary care, there were no significant differences in self-reported health (

 shows the unadjusted and adjusted associations between the supply of providers for respondents’ Census tracts and respondents’ identification of a usual source of primary and dental care. Individuals living in lower-supply areas were as likely to identify a usual source of primary or dental care as those living in higher-supply areas. These findings were no different when supply was modeled as a continuous variable (adult population per provider ratio) or when using 10-minute travel times.

In this study of new Medicaid beneficiaries from Philadelphia, we report 3 key findings. First, only 66.7% of our respondents identified a usual source of primary care, whereas nationally 78% did in 2016.

 Second, only 42.3% identified a usual source of dental care, whereas nationally 64% did during a comparable time period. Third, the supply of nearby providers was not associated with a higher or lower likelihood of identifying a usual source of primary or dental care.

The high percentage of individuals without a usual source of primary or dental care could be due to lack of perceived need, difficulty finding a provider who accepts new Medicaid patients,

 costs not covered by insurance (eg, missed work to attend appointments), or lack of knowledge about the healthcare system.

 Although we hypothesized that individuals living in lower-supply areas would be less likely to find a source of care, Philadelphia is a uniquely high-supply market, potentially affecting our findings and limiting generalizability. For example, the Philadelphia hospital referral region has 87.5 primary care physicians per 100,000 residents compared with the national median of 73.5, and it is near the 90th percentile for physicians overall.

 Second, beneficiaries may seek sources of care that are not necessarily near their home,

 basing their decisions on convenience in relation to shopping centers, workplaces, or public transportation networks. Finally, the point estimates from our model suggest that a relationship between supply and having a usual source of care may exist and is worth exploring in larger data sets across multiple cities.

Our study should be interpreted in the context of several limitations. First, given the observational nature of our study, our results are measures of association, not causation. Second, our supply measure does not utilize detailed accounts of provider effort, productivity, services offered, or acceptance of new Medicaid patients. Similarly, our demand measure (population counts) does not consider factors such as health status and healthcare-seeking preferences. Third, our approach assumes that most people prefer to obtain primary care or dental care near their home. The optimal distance to primary or dental care in urban areas is not known and may vary based on individual preference or travel patterns. Fourth, our study population was identified through a benefits outreach organization in a single urban area and may not be representative of all Medicaid enrollees in Philadelphia or other urban environments. Fifth, we were unable to determine which of 4 Medicaid managed care plans patients were enrolled in. Variations in plan types could lead to plan-level differences in provider availability, Census tract differences in plan-specific network adequacy standards, and different likelihoods of identifying a usual source of care. As a result, we may have overestimated access to care because not all nearby providers will participate in all Medicaid managed care plans. In addition, some individuals may have been assigned a provider by their insurance plan but may not have actually established a provider—patient relationship. Sixth, our generalizability is limited because we do not know the addresses or demographics of nonrespondents and cannot compare them with respondents. Finally, the modest response rate could lead to nonresponse bias. However, our response rate (31%) is an improvement compared with recent surveys of Medicaid patients, who are historically difficult to contact.

Although we did not observe a relationship between nearby provider supply and the identification of a usual source of primary or dental care, there are several key issues moving forward. This relationship should be evaluated in other geographic areas (urban and rural) and in those with low provider supply. These evaluations may help determine whether characterizing access in terms of a population to provider ratio—a measure commonly used by policy makers and researchers&mdash;is useful.

 Our results ultimately suggest that policy makers and insurance plan managers should better understand what improves access or engagement in healthcare among Medicaid enrollees so that gains in insurance coverage can be translated into gains in health.

The authors thank Carolyn Cannuscio, ScD, for her critical input on survey development and Brittany Harrison, MA, for reviewing earlier versions of this manuscript. They also thank the staff at Benefits Data Trust for survey administration and for identifying individuals in Philadelphia whom they had helped apply for Medicaid.

Division of General Internal Medicine, Perelman School of Medicine (KHC, XC, AB, DG), and Leonard Davis Institute of Health Economics (KHC, DG), University of Pennsylvania, Philadelphia, PA; Department of Public Health (JKH), Philadelphia, PA; Department of Pediatrics (CW) and Duke-Margolis Center for Health Policy (CW), Duke University, Durham, NC; Health Policy Institute, American Dental Association (KN, MV), Chicago, IL; Benefits Data Trust (EZ), Philadelphia, PA.

 This work was supported by the American Dental Association (ADA). Two of the authors (Drs Chaiyachati and Hom) received training support from the Veterans Health Administration (VHA) and Robert Wood Johnson Foundation (RWJF) during data collection and initial drafts of the manuscript. The ADA, VHA, and RWJF had no role in the study design; collection, analysis, or interpretation of the data; writing the report; or the decision to submit the report for publication.

 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 (KHC, JKH, CW, KN, EZ, MV, DG); acquisition of data (KHC, JKH, CW, EZ, MV, DG); analysis and interpretation of data (KHC, JKH, CW, KN, XC, MV, DG); drafting of the manuscript (KHC, CW, XC, AB, MV, DG); critical revision of the manuscript for important intellectual content (KHC, JKH, CW, KN, AB, MV, DG); statistical analysis (KHC, XC); obtaining funding (CW); administrative, technical, or logistic support (KHC, KN, AB, EZ); and supervision (MV).

 Krisda H. Chaiyachati, MD, MPH, MSHP, University of Pennsylvania, 423 Guardian Dr, 1313 Blockley Hall, Philadelphia, PA 19104. Email: kchai@pennmedicine.upenn.edu.

1. Starfield B, Shi L, Macinko J. Contribution of primary care to health systems and health. 

. 2005;83(3):457-502. doi: 10.1111/j.1468-0009.2005.00409.x.

2. Shi L, Macinko J, Starfield B, Politzer R, Wulu J, Xu J. Primary care, social inequalities, and all-cause, heart disease, and cancer mortality in US counties, 1990. 

. 2005;95(4):674-680. doi: 10.2105/AJPH.2003.031716.

3. Kronenfeld J. Access to dental care: a comparison of medicine/dentistry and the role of a regular source. 

4. Institute of Medicine; National Research Council. 

Improving Access to Oral Health Care for Vulnerable and Underserved Populations

. Washington, DC: The National Academies Press; 2011.

5. Chávarry NG, Vettore MV, Sansone C, Sheiham A. The relationship between diabetes mellitus and destructive periodontal disease: a meta-analysis. 

6. Offenbacher S, Beck JD, Moss K, et al. Results from the Periodontitis and Vascular Events (PAVE) Study: a pilot multicentered, randomized, controlled trial to study effects of periodontal therapy in a secondary prevention model of cardiovascular disease. 

. 2009;80(2):190-201. doi: 10.1902/jop.2009.080007.

7. Cohen LA, Bonito AJ, Eicheldinger C, et al. Comparison of patient visits to emergency departments, physician offices, and dental offices for dental problems and injuries. 

. 2011;71(1):13-22. doi: 10.1111/j.1752-7325.2010.00195.x.

8. Shoen C, Radley DC, Riley P, et al. Health care in the two Americas: findings from the Scorecard on State Health System Performance for Low-Income Populations, 2013. The Commonwealth Fund website. commonwealthfund.org/publications/fund-reports/2013/sep/health-care-two-americas-findings-scorecard-state-health-system. Published September 18, 2013. Accessed August 2, 2018.

Oral Health in America: A Report of the Surgeon General

. Rockville, MD: HHS, National Institute of Dental and Craniofacial Research, National Institutes of Health; 2000.

10. Dzau VJ, McClellan MB, McGinnis JM, et al. Vital directions for health and health care: priorities from a National Academy of Medicine initiative. 

. 2017;317(14):1461-1470. doi: 10.1001/jama.2017.1964.

11. Hinton E, Paradise J. Access to dental care in Medicaid: spotlight on nonelderly adults. Kaiser Family Foundation website. kff.org/medicaid/issue-brief/access-to-dental-care-in-medicaid-spotlight-on-nonelderly-adults. Published March 17, 2016. Accessed September 7, 2017.

12. Medicaid expansion enrollment. Kaiser Family Foundation website. kff.org/health-reform/state-indicator/medicaid-expansion-enrollment/?currentTimeframe=0&sortModel=%7B%22colId%22:%22Location%22,%22sort%22:%22asc%22%7D. Published 2016. Accessed August 30, 2017.

Articles

Nearby provider supply did not affect identification of usual sources of primary or dental care among new Medicaid enrollees. Strategies to improve access are needed.

Access to Primary and Dental Care Among Adults Newly Enrolled in Medicaid

Although health policy research should inform policy making, the communication gap between researchers and policy makers limits successful translation. Social media represents a new opportunity to connect researchers and policy makers. Our objective was to assess who Congressional health policy staff follow on a major social media platform. 

Our study measured Congressional health policy staff’s use of Twitter and the types of individuals and organizations they follow. To focus on more influential Twitter accounts, we restricted our sample to those followed by at least 3 individual Congressional staff members. 

Of the 30,843 accounts followed by the 115 Congressional health policy staff, 1273 were potentially policy-related and followed by 3 or more staff. Of these, few were academically affiliated (2.4%) or explicitly health-related (5.6%) sites; many were general news media sources (50.9%) and political and governmental sources (36.4%). Health-focused accounts were frequently connected to the news media or government rather than academia. Top accounts followed (ie, highest quintile) were most likely to be national news organizations (odds ratio [OR], 5.88; 95% confidence interval [CI], 1.75-19.7) and elected officials (OR, 8.22; 95% CI, 1.75-38.6) compared with advocacy and interest groups.

Health-related and academic sources are largely absent from the Twitter conversations with US Congressional health policy staff. Even within social media, traditional and political news media are important information intermediaries that researchers and journals should target to disseminate health policy evidence. 

		Many Congressional health staff are Twitter users; however, they are much more likely to follow traditional news media rather than academic or health-specific sources. 

		Most of the sources they follow on Twitter are traditional news and political media rather than academic or health-specific sources. 

		Even within social media, traditional and political news media are important targets for research dissemination.

In recent years, researchers, international policy bodies, health funders, and other commentators have called for new efforts to improve the translation of research into health policy.

 Yet, there are significant challenges to translating research, particularly to policy-maker audiences.

Researchers often lack the skills, support, and time to effectively translate their work.

 Policy makers have very different incentives, values, timeframes, and competing priorities that challenge their efficient use of research evidence to set agendas and make decisions.

 Given that these barriers mean that research will not always be used in policy making, one necessary step in translation is effective communication of the research itself.

Social media offers opportunities to rapidly convey public health and health policy—relevant research to target audiences. For instance, researchers, journals, and institutions can use tools like Twitter to communicate findings and provide links to content.

 Blogs can “curate” policy-relevant research so that even older research findings can be made available to policy makers when windows of opportunity to use that research become apparent, and social networks can be used to disseminate these research findings.

 In 2013, 14% of health policy researchers reported using Twitter to disseminate research, but it is unknown whether they were connected via social media to policy makers—the target users of evidence.

Congressional staff are responsible for bringing analyses to their legislators and, as drafters of legislation, are direct users of policy evidence.

 Research suggests that staffers perceive Twitter as influential

; however, to our knowledge, no research is available about Congressional staffers’ use of Twitter and the types of individuals and organizations they follow as potential sources of information. Rather, prior studies of social media in the US political system have generally focused on its use by elected officials to communicate with constituents, or the manner in which political information spreads within social networks of constituents.

 This study was designed to fill this gap in the literature, with a focus on Congressional staff. The primary objective was to describe the Twitter accounts followed by Congressional health staff and to inform potential communication opportunities for researchers.

Social media is emerging as a tool used to communicate research evidence. In this study, we were interested in who a key audience of health policy research—Congressional staff&mdash;followed on social media. We focused on a widely used form of social media, specifically Twitter, a micro-blogging tool, and identified 3 key findings.

We obtained a proprietary list of Congressional staff members with responsibilities in health from Legistorm,

 a subscription service for government affairs professionals. They collect this individual-level data on Congressional staff from publicly available sources, including government reports and traditional and social media. We purchased a list of 558 staff members with responsibilities in health as determined by Legistorm in April 2014. Legistorm provided us with the Twitter account names (“handles”) of 84 staff members. We then conducted a search for the remaining 474 Congressional staff using Google and Twitter search engines and demographic characteristics provided by Legistorm to determine if they had an active or identifiable account. Because Twitter users can make their account private, all analyses described concerned the subset of staff with public accounts (n = 115).

Using the Twitter application program interface, we retrieved the list of the Twitter accounts followed by each Congressional staff member in our database. (The retrieval date was July 26, 2014.) We retrieved characteristics of these accounts, including their short biography/description, number of followers, and number of tweets. Because we were interested in potentially influential Twitter accounts, we restricted our sample to accounts followed by 3 or more staff.

We categorized accounts along the following domains: individual versus organization, health versus nonhealth, and type (academic, political/government, news media, or interest group/advocacy, which included think tanks). The coding categories were not mutually exclusive. We created subcategories for the political/government category (elected official, partisan political organization, and government agency) and news media category (national, local, and political). Coders used the Twitter bio to classify accounts; for example, any entity that cited a “.edu” address in the bio or offered a university affiliation was classified as academic and any entity that used health-related terms anywhere in their bio was classified as health-related. Accounts on Twitter that could not be categorized into this scheme were determined to be unlikely to be related to professional duties (eg, local restaurants, non—news-focused television programs) and excluded from our analysis. For the purposes of this study, we defined the remaining Twitter accounts as policy-related. We pilot tested the coding instrument with 2 trained coders. Each coder then classified all Twitter accounts. If a coding category had an inter-rater reliability below 0.61, a third coder and co-investigator (JS) adjudicated after discussing with the study team.

We calculated frequencies for each Twitter account followed to determine the number of Congressional staff following each, and limited our analyses to those accounts followed by 3 or more staff (n = 2203). We modeled the characteristics of the most-followed accounts (ie, those falling in the top quintile of sources) using multivariable logistic regression.

The study was reviewed and approved by the institutional review board at the University of Pennsylvania. Data collection and analyses were done in compliance with Twitter’s terms of service.

From a database of 558 Congressional staff, we identified 33% (184) with Twitter accounts. Of the 184 with Twitter accounts, 69 had private accounts and 115 had public accounts. The 115 Congressional staff with public accounts collectively followed 30,843 different individual or organization Twitter accounts. Three or more Congressional staff followed 2203 of these Twitter accounts.

Of the 2203 individuals and organizations followed by 3 or more Congressional staff, 1273 (57.8%) were policy related (Table 1). Of these 1273, just 71 (5.6%) were health related. Very few (30; 2.4%) of the Twitter accounts followed were affiliated with academic institutions. Congressional staff most commonly followed news media accounts; 648 (50.9%) were affiliated with news media organizations, primarily national news organizations and political news organizations (eg, Politico); local news organizations were less commonly followed. Political and governmental accounts were the second most common category, representing 464 (36.4%) of accounts followed.

The top accounts were followed by 40 to 67 Congressional staff (Table 2). These accounts also had a large number of followers, ranging from 62,876 to 44,129,382. The top 6 included political news organizations (Politico, The Hill), national news organizations (

, elected officials (The White House, Barack Obama), and a political satirist (Stephen Colbert). The other most frequently followed accounts were predominantly related to news media organizations.

In a multivariable logistic regression model, we modeled the association of the characteristics of Twitter accounts with their likelihood of being highly followed (top quintile). We restricted this analysis to the 1273 accounts followed that were coded as policy related. Although there were more Twitter accounts in our sample representing individuals, individual accounts were less likely to be frequently followed by Congressional staff than those representing organizations (odds ratio [OR], 0.30; 95% confidence interval [CI], 0.10-0.92; 

= .035). Accounts representing national news organizations (OR, 5.88; 95% CI, 1.75-19.7; 

= .004) and elected officials (OR, 8.22; 95% CI, 1.75-38.6; 

= .008) were the most frequently followed by Congressional staff.

Top Health-Related Twitter Accounts Followed

Among the 71 health-related accounts followed by Congressional staffers (top accounts shown in Table 2), the most popular account was that of Sarah Kliff, a journalist and blogger with a national media organization (Vox.com, formerly of the

 Health) and governmental and political accounts (eg, HHS) were frequently represented among the top accounts. 

 was the only health-related academic journal followed by 3 or more Congressional staff.

Using Twitter, we identified 3 key findings. First, many Congressional staff used Twitter and followed accounts that may be related to their professional roles. In our sample of health-related Congressional staff, we were able to successfully identify one-third of targeted individuals with Twitter accounts. This is likely an underestimate given that we were unable to search for individuals who preferred to obscure their identity on Twitter (eg, use of pseudonyms). Second, even among Congressional staff with a health focus, very few of the accounts followed were health or health-policy related (just 71 of 1273; 5.6%). The accounts that were explicitly health-focused were often associated with major news media outlets. Third, beyond health-focused accounts, Congressional staff were most likely to follow national and political news media organizations and second most likely to follow political and governmental individuals and agencies.

Academic sources were almost completely absent from Twitter accounts frequently followed by Congressional staff.

 The few academic accounts represented in the sample were typically individuals with faculty appointments who had once served in a political or governmental role. It has been well documented that traditional news media can shape policy-makers’ agendas

 and that information sources from government and interest groups are important to decision makers.

 However, social media does not appear to circumvent traditional pathways of information dissemination (ie, the news media), but rather serve as an alternative channel to connect entities with influence that predated social media.

Our study has several important implications. First, although policy makers’ staff were common users of social media, they did not appear to be using the technology in a way that directly connected researchers and policy makers. This suggests that academic researchers and institutions should not overlook traditional news media targets in their social media dissemination strategies. These findings may also point to the need for researchers and academic institutions to build relationships with trusted information intermediaries (eg, advocacy organizations/think tanks) that could curate new research findings and deliver information through social media that is timely and salient to policy makers. Second, although the pathways may be similar to traditional channels, the nature or influence of the information communicated may nonetheless be different. Future research should examine how the information communicated through social media differs from that provided by traditional news sources (eg, newspapers), and whether it produces alterations in terms of influence. News outlets increasingly have health-focused Twitter feeds that are narrow in focus. Future research should also examine how the information-seeking habits of various types of policy staff differ. For example, staff in Senate member offices, on Congressional committees, or in executive branch agencies have more specialized roles than staff in House member offices. Therefore, they may rely on different sources through social media.

Our study has several limitations. First, we obtained a list of health-related Congressional staff from a company that gathers this information and sells it to third parties. We were not able to validate that the database included all health-related Congressional staff or if those identified were actively working on health policy in a Congressional office. We also could not verify that the individuals used Twitter for work-related purposes. By restricting the search to accounts followed by 3 or more staff, we identified accounts that were more likely to be related to their professional roles. In addition, we were unable to extract the Twitter accounts followed by Congressional staff if their own account was private (n = 69). Another possibility is that some staff maintain 2 Twitter accounts—a private account (or public account with a pseudonym) for work-related use and a public account for personal use. If this were the case, our analysis may be missing some Twitter accounts followed by 3 or more staff. Given these limitations, our sample is unlikely to be representative of all health policy staff in the Congress, and may be missing some Twitter accounts followed by staff.

Second, we coded each Twitter account using the bio within the Twitter account. In some cases, the bio provided limited information and thus, there may be some misclassification of Twitter accounts that could not be classified within our coding scheme. Third, we did not attempt to characterize the content of information being communicated on Twitter or the relative contributions of different accounts. Fourth, Twitter users sometimes followed “lists” of Twitter accounts followed by others. This allowed a user to follow another user’s list of curated accounts, which could have been either public or private. Our analysis did not account for this. Finally, our study focused on Twitter and our results were only generalizable to this social media platform.

Improving the translation and dissemination of health policy and health service research is essential to achieving evidence-influenced policy making. Social media represents a highly touted opportunity to communicate research evidence. However, it does not appear to circumvent the traditional news media as an important intermediary that provides information to policy makers.

 As academic researchers and institutions use social media to disseminate research, our findings suggest that they should focus their communication strategies on news media targets if the goal is to reach networks policy makers already use. 

The authors thank Maarten Sap for his technical assistance on this project. This study was supported by the Robert Wood Johnson Foundation Health & Society Scholars Program at the University of Pennsylvania. The funder had no role in the study design; collection, analysis and interpretation of data; writing of the manuscript; and the decision to submit the article for publication.

 Division of General Internal Medicine (DG), Leonard Davis Institute of Health Economics (DG, ZFM, RMM), and Department of Emergency Medicine (ZFM, RMM), University of Pennsylvania, Philadelphia, PA; School of Public Health, Boston University (JS), (Boston, MA); School of Public Health, University of Minnesota (SEG), (Minneapolis, MN).

 Robert Wood Johnson Foundation Health & Society Scholars Program at the University of Pennsylvania.

 Concept and design (DG, ZFM, RMM, SEG); acquisition of data (DG, ZFM, RMM); analysis and interpretation of data (DG, ZFM, RMM, JS, SEG); drafting of the manuscript (DG, RMM, JS, SEG); critical revision of the manuscript for important intellectual content (DG, ZFM, RMM, SEG); statistical analysis (DG); obtaining funding (DG); administrative, technical, or logistic support (ZFM, JS).

 David Grande, MD, MPA, University of Pennsylvania, 3641 Locust Walk &shy;— 407, Philadelphia, PA 19104. E-mail: dgrande@wharton.upenn.edu. 

1. Prewitt K, Schwandt TA, Straf ML; National Research Council. Using science as evidence in public policy. The National Academies Press website. https://www.nap.edu/catalog/13460/using-science-as-evidence-in-public-policy. Published 2012. Accessed July 12, 2017.

2. World report on knowledge for better health: strengthening health systems. World Health Organization website. http://www.who.int/rpc/meetings/en/world_report_on_knowledge_for_better_health2.pdf. Published 2004. Accessed July 12, 2017).

3. Bogenschneider K, Corbett TJ. Evidence-Based Policymaking: Insights From Policy-Minded Researchers and Research-Minded Policymakers. New York: Routledge; 2011.

Disseminating timely and relevant research findings to policy makers is a national priority to inform health policy decisions. Social media is a novel tool to bridge the communication gap.  

Twitter Accounts Followed by Congressional Health Staff

Given that many patients with chronic diseases face cost-related barriers to care, we evaluated patients’ views on which providers (both physicians and nonphysicians) to involve and which methods to use to screen for those barriers. We also examined patients’ preferences for how physicians consider cost-efficacy trade-offs in decisions.

A national survey of 1400 randomly sampled adults with a chronic disease seeking financial assistance (842 respondents).

Participants rated their comfort with various providers and tools for identifying cost barriers. Then they rated a randomly assigned clinical vignette that described how a clinical decision was made in the context of a cost-efficacy tradeoff. Vignettes depicted 3 decision types: cost-conscious physician, cost-indifferent physician, or patient-directed. Comfort was rated from 1 to 10—ratings above 7 indicated high comfort.

More respondents reported high comfort with physicians screening for cost barriers (81.1%) than with pharmacists (74.8%; 

<.001). Regarding screening for cost barriers using administrative records, more respondents reported higher comfort with doctors’ offices (58.8%) than with insurance companies (53.3%; 

= .03), but similar levels of comfort compared to pharmacies (62.1%; 

= .17). Participants favored “patient-directed” decisions with physician input (odds ratio, 4.64; 95% CI, 3.14-6.84; 

<.001) compared with “cost-conscious” decisions in which physicians unilaterally decided how to manage cost-efficacy tradeoffs.

Patients were open to a range of cost-barrier screening approaches, but most favor direct conversations with their doctor and shared decision making in decisions involving cost-efficacy trade-offs.

Prior studies show that patients frequently encounter cost barriers leading to costrelated nonadherence with medications, but rarely do patients raise these concerns with their doctors nor do their doctors initiate discussions on the subject. Our study finds:

		Many patients are open to a range of cost-barrier screening approaches outside of face-to-face visits with their doctors, including screening by nonphysician members of the healthcare team and administrative mechanisms such as insurer claims reviews or alerts from pharmacies to physician offices.

		Even when patients face cost barriers, they generally want a participatory decision-making process to address cost-efficacy trade-offs. Physicians need tools to facilitate these discussions.

Financial stress associated with healthcare imposes a substantial burden on US families, including a high rate of cost-related nonadherence to prescription medications.

 For example, 20% of seniors in fair or poor health report cost-related nonadherence despite a high rate of Medicare prescription drug coverage.

Cost-related nonadherence leads to higher rates of morbidity and mortality, including elevated rates of cardiovascular events.

A number of factors suggest that cost barriers will continue to increase. First, despite a recent slowdown, the rate of growth of healthcare costs continues to exceed that of economic growth.

 Second, high-deductible health plans, which were further stimulated by the creation of tax-advantaged health savings accounts through the Medicare Modernization Act a decade ago, are growing rapidly, with 19% of workers with employer-sponsored insurance now enrolled.

 Third, although the Affordable Care Act will significantly expand insurance coverage, the minimum coverage requirements will still permit a high level of cost sharing.

As cost barriers grow, patients will increasingly need assistance in making healthcare decisions that involve trade-offs between the efficacy and affordability of treatment regimens. (We will refer to these decisions as cost-efficacy trade-offs.)

Prior studies have examined attitudes and behaviors regarding physician-patient cost discussions. Those studies showed that patients wanted to discuss costs with their physician and would invite discussions about affordability.

 However, just 35% of patients experiencing cost-related nonadherence reported having had a cost discussion with their doctor.

Although physicians report that cost discussions are important, they often do not recognize when their own patients are experiencing cost barriers.

 Physicians also perceive that they lack the knowledge to discuss treatment costs, and that time pressures of an office visit prohibit meaningful discussions.

Given the infrequency of cost discussions, several experts have proposed deploying other members of healthcare teams to screen for financial stress and potentially intervene.

However, we are not aware of any studies that have tested the acceptability among patients of screening or interventions outside of traditional physician-patient face-to-face encounters.

As cost barriers become more prevalent and start to be recognized, it is important to understand how physicians and patients should incorporate financial considerations into medical decisions. The literature on patient preferences for their own role in medical decision making shows heterogeneous preferences.

 the majority of patients reported that they wanted to be asked their opinions on therapeutic options and be offered choices, yet almost half preferred to rely exclusively on their physician’s knowledge to acquire information, and just over half preferred to still leave final decisions to their physician. To our knowledge, studies have not examined patients’ preferences regarding physicians’ roles in navigating cost-efficacy trade-offs.

In this study, we had 2 main objectives. First, in a sample of patients with a chronic disease, we evaluated comfort with being screened for medication cost-related barriers by different types of healthcare providers and through different methods. Second, we examined patients’ preferences for physician and patient roles in making medical decisions involving cost-efficacy trade-offs.

We conducted a mail survey of a random sample of adults with a chronic disease who had applied for financial support to the HealthWell Foundation, a national nonprofit organization that assists patients with co-payments, co-insurance, and health insurance premiums (

); both individual and corporate donors contribute to the HealthWell Foundation. Embedded in the mailed survey was a randomized experiment to evaluate patient preferences regarding physician approaches to helping patients navigate cost-efficacy trade-offs in the prescribing of medications. The University of Pennsylvania Institutional Review Board approved this study, which was funded by the HealthWell Foundation.

We asked participants to rate their level of comfort with their own doctor and other members of the healthcare team screening them for medication cost—related barriers to care. In addition to their own doctor, we asked about nurses, professional counselors, and trained volunteers at their doctor’s office, as well as pharmacists where they fill prescriptions. We also asked participants to rate their comfort with a range of strategies outside the normal doctor-patient visit, to screen for cost-related nonadherence or cost-barriers. These strategies included “having someone from your doctor’s office review your records to see how often you fill your prescriptions”; “having your insurance company review your records to see how often you fill your prescriptions”; “having your pharmacy review your records to see how often you fill your prescriptions”; “filling out a form in your doctor’s office”; and “responding to an email message from your doctor’s office.” Participants rated their level of comfort with each item in this section on a 1-to-10 scale where 1 was very uncomfortable and 10, very comfortable.

The next section of the survey included an embedded experiment in which participants were randomized to 1 of 3 versions of a vignette describing a clinical decision with cost-efficacy trade-offs. Every survey included the following: “Lucy has high cholesterol. She can take a medicine to help prevent a heart attack. The options are Medicine A and Medicine B. Medicine A costs Lucy $20 a month. It works fairly well. 80 out of 100 people get better. Medicine B costs Lucy $100 a month. It works very well. 99 out of 100 people get better.” The experimental conditions varied as follows, with one-third of the survey sample allocated at random to each condition; in condition 1, the “cost-conscious” decision style, Dr Thomas is aware that Lucy is having difficulty paying for her medications and makes a unilateral decision to prescribe Medicine A—the less expensive, less effective medication. In condition 2, the “cost-indifferent” decision style, Dr Thomas knows that Lucy is having difficulty paying for her medicine but unilaterally decides to prescribe Medicine B because it is “best.” In condition 3, the “patient-directed” style, Dr Thomas knows Lucy is having trouble paying for her medicines. Before Dr Thomas prescribes a medicine, she discusses Medicines A and B with Lucy and lets Lucy decide. Participants rated their level of comfort with the decision style depicted in the condition to which they were randomized on a 1-to-10 scale, where 1 was very uncomfortable and 10, very comfortable.

We also asked respondents about their experiences with cost-related nonadherence or difficulty paying for their medications in the last year, as well as whether they would like to discuss costs with their doctor. Self-rated health was measured using the SF-1 questionnaire, a predictor of all-cause mortality.

Physician trust was measured using a 5-item instrument.

The HealthWell Foundation provided demographic information for participants, including income, age, insurance status, and marital status.

We administered the survey between August 2012 and October 2012. We mailed surveys to 1400 individuals randomly selected from the HealthWell Foundation database. We sent each individual an introductory letter, followed approximately 5 days later by the mailed survey and consent form as well as a $5 participation incentive. We followed this mailing with a postcard reminder 1 week later, and a fourth and final mailing containing a second survey approximately 2 weeks after that. Among the people invited to participate, 107 were unreachable because of incorrect addresses; and 6 people were deceased according to a family member. A total of 842 participants responded for a final survey response rate of 60.4%.

Our main outcomes were how comfortable the respondent was with the method or decision described in the survey instrument. We dichotomized our outcome into high comfort (8 to 10) versus low to medium comfort (1 to 7). We tested bivariate associations between demographic characteristics, self-reported health status, self-reported cost difficulty, and self-reported cost-related nonadherence with our main outcomes using χ2 statistics. We also tested for differences between our 3 experimental groups using χ2 statistics. We adjusted for characteristics that differed (

<.10) between experimental groups using logistic regression. We performed all analyses using the Stata software program, version 12.1 (StataCorpLP, College Station, Texas).

In this sample of adults with a chronic disease who faced financial challenges associated with their healthcare, the majority were women and almost half were 65 years or older (

). Although 70% of the sample earned less than $30,000 per year, most respondents had some form of health insurance—either through a private insurer or through Medicare. More than half of the respondents rated their health as fair or poor. The majority of participants noted that they had experienced difficulty paying for prescriptions and/or cost-related medication nonadherence in the past year. Most agreed that, “I want my doctor to think about cost when choosing a medicine for me” (86%). Among the 1400 people mailed a survey, responders were somewhat older than nonresponders (

Financial Screening by Healthcare Team Members

Participants were asked how comfortable they would be with different members of the healthcare team inquiring about their own financial challenges (

). Responses suggest that patients view members of the healthcare team differently. Participants most often reported being highly comfortable being questioned by physicians (81.1%), followed closely by pharmacists (74.8%). Nurses (69.4%) and professional counselors (68.3%) ranked next in terms of participants’ comfort, with trained volunteers (50.5%) least often endorsed. These ratings were similar across all sociodemographic groups. Participants who reported a high degree of physician trust reported higher levels of comfort with all members of the healthcare team.

Participants were prompted that “there are many ways your doctor can learn that you are having trouble paying for your medicines” and were asked about their comfort with a range of screening strategies to identify patients’ financial challenges (

). The majority of participants reported that they were highly comfortable with record reviews by doctors’ offices (58.8%), insurance companies (53.3%), and pharmacies (62.1%), as well as with self-reported screening paper work to be completed in doctors’ offices (62.2%). However, participants were significantly less likely to report that they were highly comfortable with the use of email from the doctor’s office as a screening tool (48.1%).

The proportion of people reporting a high level of comfort with each screening strategy did not vary by sex, age, marital status, income, insurance status, or health status. However, there were significant differences between participants who did and did not report difficulty paying for their prescriptions (Table 3). Participants who reported difficulty paying (vs no difficulty) were open to a range of strategies. They were significantly more likely to be highly comfortable with screening by doctor’s office or pharmacy record review, in-office paper-based screening, and screening via e-mail. Higher trust in physicians was also significantly and positively correlated with endorsement of every proposed approach to screening for financial difficulty (Table 3).

Preferred Patient-Physician Roles in Cost-Based Decisions

Experimental vignettes were used to model a clinical scenario in which a patient could be treated with a higher- or lower-priced cholesterol-lowering medicine; in each scenario, the higher-priced medicine was more effective. Participants were randomly assigned to 1 of 3 conditions—in all 3 conditions, the doctor is aware that the patient was having financial difficulty. In the first condition, the “cost-conscious” doctor unilaterally chose a lower-priced, marginally less effective medication. In the second condition, the “cost-indifferent” doctor chose the higher-priced medication because it was the most effective. In the third scenario, the doctor and patient discussed the cost-efficacy trade-offs and the ultimate decision was “patient-directed.” This last scenario was significantly favored. In unadjusted analyses (

), the majority (57.6%) of participants assigned to the patient-directed condition reported that they would be highly comfortable (>7 on 1-10 scale) with a clinical decision style like this in which the patient ultimately chose which medication to use. A lower proportion of patients reported that they would be highly comfortable with the other clinical decision styles. Only 32.5% of participants reported that they were highly comfortable with the cost-indifferent style in which the physician unilaterally chose the more expensive, marginally more effective medication. Even fewer participants (25.2%) were highly comfortable with the cost-conscious decision, in which the physician unilaterally decided to prescribe the less expensive, marginally less effective medication.

In multivariate analyses (Table 4) that adjusted for respondent characteristics that varied (

<.10) between experimental groups (health status and marital status), participants’ ratings of comfort differed significantly across experimentally assigned clinical decision styles. Compared with participants who were randomized to the cost-conscious decision style, in which the physician unilaterally chose the lower-cost medication, participants randomized to the patient-directed condition were substantially more likely to report a high level of comfort with the clinical decision style (odds ratio [OR], 4.64; 95% CI, 3.14-6.84; 

<.001). A high level of comfort was more common (OR, 1.53; 95% CI, 1.04-2.24; 

= .03) among participants assigned to the cost-indifferent condition (physician chose the higher-price medicine because it was “better”) than among participants assigned to the cost-conscious condition (physician chose the lower-price, marginally less effective medication). Results were similar when also adjusting for age.

Following the rating of the vignette, all participants were asked to rate, on the same 10-point Likert scale, their comfort with the same vignette modified to depict a shared-decision style, in which the doctor and patient collaboratively chose the medication. Most (85%) participants were highly comfortable with that approach. Later in the survey, we asked participants to explicitly choose a preferred physician role if the respondent was “having trouble paying for my medicines.” Respondents favored a shared decision (77%; “my doctor gives me options, tells me their prices and how well they work, and together we decide”) over a patient-directed decision (14%; “my doctor gives me options, tells me their prices and how well they work, and lets me decide”), cost-indifferent decision (6%; “my doctor prescribes the best medicine for me, even if it is more expensive”), or cost-conscious decision (3%; “my doctor chooses a cheaper medicine, even if it is not the best”).

Participants in our study articulated their preferences around screening and negotiation of cost barriers in their medical care. The most important finding from this work is that participants were open to having their physicians engage them in discussions regarding medical decision making when faced with difficulty paying for prescription medications. Participants were also open to such discussions with pharmacists, nurses, and other members of the healthcare team—though less so with lay health workers. The second important finding is that patients want a participatory decision-making process to navigate cost-efficacy trade-offs and do not want to defer those decisions to their doctor. This research has several important implications for clinical practice and future research.

First, as more patients enroll in health plans that require high levels of cost sharing, it will be important for providers to monitor for, and pre-empt when possible, cost-related nonadherence. Prior studies show that physicians often do not recognize when their patients are experiencing cost barriers. To identify cost barriers to medication adherence, participants in this study were asked if they were comfortable with a range of screening strategies using administrative records or queries in the doctor’s office. Our study points to alternative personnel and approaches that medical offices could utilize, with likely broad patient support, to better identify patients experiencing cost barriers.

Second, pharmacists are a potential untapped resource and partner for medical practices to engage for cost discussions with their patients. Of all the providers we asked participants about, pharmacists were the only professional group not typically located within medical practices. This presents both barriers and opportunities. On the other hand, communication between medical offices and pharmacists can be a greater challenge given the separation of electronic communication systems. On the other hand, at least for medications, pharmacists have ready access to cost information, and likely have frequent contact with patients given the need for medication refills, and may be in closer geographic proximity to a patient’s home and have more flexible access. Further research is needed to determine how pharmacists can be effectively deployed to lead cost discussions with patients and how these discussions compare with those between patients and physicians.

Third, patients want to engage with their physicians in discussions and decisions regarding trade-offs between the costs and efficacy of their medications. The finding that patients want to engage in shared decision making is not new.

Our study adds to the literature by showing that patients also value shared decision making when facing cost-efficacy trade-offs due to personal financial barriers. Physicians engaging in shared decision making when confronting cost-efficacy trade-offs, face a number of challenges, including time and knowledge of costs. These challenges point to the need to develop and deploy new decision-making tools (eg, tools within electronic medical records), as well as training within medical education and practices, to help patients navigate these decisions.

Our study has several limitations. First, our participants were all applicants to a patient assistance foundation and tended to be older, insured, and female. These participants, given their applications to a patient assistance foundation, might have different attitudes than other individuals about seeking help or discussing personal financial matters with healthcare providers. In addition, they may have greater levels of patient activation or may just have different strategies for dealing with personal adversity than other members of the population. We also did not have measures of race and ethnicity. Thus, our study may overestimate patient willingness to discuss cost barriers with healthcare providers or patient desire to engage in participatory decision making around cost-efficacy trade-offs. However, by recruiting participants from this program, we were able to readily identify individuals with a chronic condition facing cost barriers.

Second, we measured attitudes toward different physician decision roles using experimental vignettes. Patients’ expressed preferences in these hypothetical scenarios may differ from their behaviors in actual clinical encounters. However, the experimental design increases our confidence that the differences we observed across experimental groups are likely to reflect between-group differences that would manifest in clinical encounters. Third, we anticipate that patients would view physician roles in cost-efficacy trade-offs differently depending on the clinical conditions or types of treatment under consideration. This question was beyond the scope of our study.

Fourth, as with all survey research, it’s possible that responders and nonresponders have differing attitudes. However, while it does not eliminate the concern for nonresponse bias, our relatively high response rate of 60.4% further mitigates it.

Given the underlying trends in costs and insurance design, more patients are likely to confront cost barriers. Patients desire more direct communication with healthcare providers regarding cost-efficacy trade-offs; they want to be included in the decision-making process. Interventions will have to address 2 major barriers to effective communication regarding cost-efficacy trade-offs. First, healthcare providers receive little to no formal education in discussing patients’ financial challenges and implications for health, health behaviors, and treatment compliance. Second, the costs of healthcare, including anticipated out-of-pocket costs for patients, are not at all transparent to providers or patients. Corrective action is needed on both fronts in order to deploy screening and intervention strategies in clinics, hospitals, and pharmacies.

The authors thank the HealthWell Foundation for support of this research. This work was presented as an oral presentation at the Society of General Internal Medicine Annual Meeting on April 26, 2013, in Denver, Colorado.

 Division of General Internal Medicine (DG) and Department of Family Medicine and Community Health (CC), Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA; Leonard Davis Institute of Health Economics (DG, CC), University of Pennsylvania (MT), Philadelphia, PA; Department of Medicine, University of California (ML), San Francisco, CA.

Patients are receptive to diverse strategies to screen for cost barriers but want participatory decision making to address cost-efficacy tradeoffs.

David Grande, MD, MPA

Articles

Access to Primary and Dental Care Among Adults Newly Enrolled in Medicaid

Twitter Accounts Followed by Congressional Health Staff

Addressing Cost Barriers to Medications: A Survey of Patients Requesting Financial Assistance