The Impact of Market Factors on Meaningful Use of Electronic Health Records (EHRs) among Primary Care Providers: Evidence from Florida using Resource Dependence Theory and Information Uncertainty Perspective

Abstract

Background:

Using federal funds from the 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act, the Centers for Medicare and Medicaid Services funded the 2011–2021 Medicaid EHR incentives programs throughout the country.

Objective:

Identify the market factors associated with Meaningful Use of EHRs after primary care providers (PCPs) enrolled in the Florida – EHR incentives program through Adopting, Improving, or Upgrading (AIU) an EHR technology.

Research design:

Retrospective cohort study using 2011–2018 program records for 8464 Medicaid providers.

Main Outcome:

Meaningful Use (MU) achievement after first-year incentives.

Independent Variables:

The resource dependence theory and the information uncertainty perspective were used to generate key independent variables, including county’s rurality, educational attainment, poverty, health maintenance organization (HMO) penetration, and number of PCPs per capita.

Analytical Approach:

All the county rates were converted into three dichotomous measures corresponding to high, medium, and low terciles. Descriptive and bivariate statistics were calculated. A generalized hierarchical linear model was used because Meaningful Use data were clustered at the county level (level 2) and measured at the practice level (level 1).

Results:

Overall, 41.9% of Florida Medicaid providers achieved Meaningful Use after receiving first-year incentives. Rurality was positively associated with Meaningful Use (P<.001). Significant differences in Meaningful Use achievements were obtained when we compared the ‘high’ terciles with the ‘low’ terciles for poverty rates (P=.002), HMO penetration rates (P=.02), and number of PCPs per capita (P=.01). These relationships were negative.

Conclusions:

Policy makers and healthcare managers should not ignore the contribution of market factors in EHR adoption.

INTRODUCTION

The use of Electronic Health Records (EHRs) among primary care providers (PCPs) in the United States has grown dramatically in the last two decades and particularly since the inception of the 2009 Health Information Technology for Economic and Clinical Health (HITECH) act, enacted as part of the American Recovery and Reinvestment Act (ARRA).¹ PCPs use a wide range of EHR systems. The basic EHR systems contain applications used to store and organize patient information while the more advanced EHR systems can be used to present medical history, e-prescribe drugs, support clinical decision, exchange health information, and improve clinical care coordination.² From 2011 to 2021, using federal funds available from the HITECH act, the Centers for Medicare and Medicaid Services (CMS) funded the Medicaid EHR Incentives programs, later renamed the Medicaid Promoting Interoperability (MPI) programs.³ When enrolled into the MPI programs, PCPs received an incentive payment of $21,250 and committed to “Adopt, Implement, or Upgrade” (AIU) a certified EHR system. In practice, “Adopt” is to acquire and install a certified EHR system. “Implement” is to begin using a certified EHR technology, for example, by providing proof of staff training or data entry of patient demographic information into a certified EHR system. “Upgrade” is to expand an existing EHR system to meet federal certification requirements. Each state individually administered its MPI program with 90% funding from CMS. All the 50 states and the District of Columbia voluntarily participated.³

Eligible providers included physicians, nurse practitioners, certified nurse-midwives, dentists, and physician assistants. These providers must meet a 30% Medicaid encounter volume threshold over a 90-day period, but pediatricians could qualify with a minimum of 20% Medicaid encounter volume. Additional requirements included being in active status, having no outstanding state or federal sanctions, and not being hospital based, that is, 90% or more encounters in a year did not occur in a hospital or emergency room setting. These providers received a second payment if they used the functionalities of their EHR systems to meet specific federal benchmarks that characterize “Meaningful Use” (MU) of EHRs.¹ A few providers with a history of using the EHR technology met these MU benchmark requirements at enrollment. For the purpose of this study, these providers were excluded from the study sample.

The present study investigated the associations between local market factors and MU of EHRs after providers enrolled through AIU into the Florida - Medicaid Promoting Interoperability (FL-MPI) program. ⁴ It contributes to the literature on EHR adoption in several ways. First, due to the paucity of available data, empirical work on the use of EHR functionalities by PCPs is overall lacking.² Second, while PCPs’ participation and practice characteristics in the MPI programs have been assessed,^5,6 the extent to which local market factors contributed to attesting to MU of EHRs has not been empirically examined. Third, we used the key constructs of two theoretical perspectives, the resource dependence theory (RDT) and the information uncertainty perspective (IUP), that offered organizational concepts to identify relevant market constructs for this study objective. Fourth, in addition to receiving reimbursement rates that lag Medicare and private insurance, practices serving the Medicaid populations typically have a higher proportion of minority, low-income, and underserved individuals with multiple comorbidities.⁷ It is thus critical to assess how market factors impact MU achievements among Medicaid providers and whether federal funding designed to promote MU of EHRs among these providers is effective.

METHODS

Conceptual framework

Several frameworks in organizational theory have offered relevant concepts to understand the primary care system by characterizing the importance of local market factors.^8,9 This study drew upon two of these theoretical perspectives: the resource dependence theory (RDT) and the information uncertainty perspective (IUP). These two frameworks inform the relationship between local market conditions and MU attestations by PCPs that participated in the FL-MPI program. They are used to operationalize the market environments in which these PCPs conducted businesses. RDT conceptualizes the healthcare market as a source of resources upon which providers depend to operate and survive ^10,11 RDT dictates that providers will do what they can to secure resources from their markets through a variety of actions or exchanges. ¹⁰ IUP adds that, due to lack of perfect information about the healthcare market, uncertainty will arise as providers compete with each other to secure necessary resources.¹² Hence, uncertainty is given important consideration as organizations are making strategic decisions on the use of market resources.¹¹ Using counties to approximate the healthcare markets for providers, an increasing number of studies have used these two perspectives by focusing on one or more of the following market dimensions: munificence, dynamism, and complexity. ^8,13,14

Market Munificence

Market munificence is concerned with the availability of environmental resources to an organization.¹⁵ The fact that practices do not have exclusive control over resources in their markets produces uncertainty for managers that must strategize to secure resources to operate and survive.¹⁵ In less munificent markets, providers face greater uncertainty in their efforts to secure environmental resources.¹⁵ Market munificence was operationalized by the county’s rurality, educational attainment, and poverty. Rurality is marked by the lack of technological and workforce resources, which has been a significant barrier to EHR adoption among medical practices.¹⁶ Educational attainment is a proxy for human capital and skills available in a market. It reduces adoption costs of technology, and thereby raise the probability of EHR adoption.¹⁷ The last munificent measure, poverty, was found associated with lower patient portal utilization, often because healthcare providers failed to offer patient portal access to poor patients.¹⁸ Moreover, providers in wealthier counties might find it more beneficial to use the advanced functionalities of their EHRs to appeal to potential wealthy clients, who might be more selective in their health care utilizations.^8,14

Market Dynamism

The rate of change in the healthcare markets reflects the unpredictability of competitors and clients.¹⁹ Fluctuations in the healthcare markets would only exacerbate the level of uncertainty for the Medicaid providers, which already face several barriers. Dynamism was operationalized by the county’s health maintenance organization (HMO) penetration. Growth in HMO in a county might change the day-to-day operations of the providers, healthcare costs, and treatment patterns.²⁰ While a few studies found no associations between managed care penetration and the adoption of health information technology (health IT),^13,21 a recent study conducted among Florida physicians found that HMO penetration was negatively associated with the use of computerized provider order entry.²²

Market Complexity

Market complexity reflects the number of various actors in the healthcare markets that should be taken into consideration.²³ If a relatively large number of providers exist in the market, they must compete for the scarce resources and secure their share of patients. Complexity was operationalized by the county’s number of nonfederal PCPs per capita. Competition in health care markets can help contain costs, improve quality, and encourage innovation.²⁴ In the face of competition, providers can adopt strategic health information systems, become more innovative, and create competitive advantage.²¹ Providers might thus find it beneficial to use the advanced functionalities of their EHRs as they need to appeal to the patient population.²⁴

Data source and sample construction

The main database, the Provider Participation Database, supplied data on each participant’s 10-digit national provider identification (NPI), enrollment year, payment dates and amounts, zip codes, and AIU phases.⁴ We used the NPI to link the main database to several provider’s characteristics, including the provider’s specialty and practice type as well the vendor from which the providers bought and/or got technical assistance for their EHR systems. We then matched the providers’ zip codes with the 67 Florida counties, which were then used to link state data to the main database. The Florida rural counties map provided information on the rurality of the counties.²⁵ The Florida Office of Insurance Regulation (FOIR) provided information on managed care enrollments for each county.²⁶ The rest of the county data came from the Area Health Resource Files (AHRF), a national database of health resources measured that provide data for over 6,000 indicators collected from more than 50 sources.²⁷ The data source years included typically covers the previous 10 years. The AHRF file contained county codes that enabled it to be linked to the main database.

The main dataset did not contain the dates of MU attestations. It only contained the dates providers received payments for having attested to MU. Hence, the sample dataset was constructed as a cross-section with multiple years of data.

Outcome Measure

Providers applied for the first-year payment (Payment Year 1) and committed to one of the AIU phases during enrollment. Providers received a second payment (Payment Year 2) after they have attested to MU of EHRs as specified by CMS requirements. The outcome measure (MU) was equal to 1 if providers received a second payment, and 0 otherwise.

County Characteristics

Counties were categorized as rural and urban counties. We estimated educational attainment as the percentage of the population aged 25 years and older that had a bachelor’s degree. Poverty rate was calculated as the percentage of the county’s population that lived in poverty based on standard federal definitions. The HMO penetration rate was measured as the percentage of the county’s population covered by an HMO. The number of PCPs per 10,000 populations included only non-federal PCPs practicing in the county. Except for the HMO enrollment data, all the county rates were estimated using data from the AHRF.

Control Variables

We controlled for provider’s specialty (physician, nurse practitioner, certified nurse-midwife, dentist, and physician assistant), AIU phases, and practice type indicating whether the provider was conducting businesses at an independent practice or at a group practice.^5,6 We also controlled for EHR vendor types. Approximately 160 vendors with varying market share sold EHR units and/or provided technical support to providers in the FL-MPI program.²⁸ The vendors’ market share ranged from selling to only a few providers up to 1300 providers. Vendors were thus operationalized as small vendors if they sold EHR units to fewer than 200 providers and large vendors if they sold EHR units to at least 200 providers.²⁸

Analytical Approach

We received 8748 records of providers that participated in the FL-MPI program. A total of 8464 providers (96.8%) enrolled through AIU phases. Our sample included 2641, 2048, 1574, 779, 719, and 703 providers that enrolled each year, from 2011 to 2016, respectively. This allowed for a minimum of two years of observations through 2018. The unit of analysis was thus provider-years. All data accessed complied with data protection and privacy regulation.

To facilitate comparisons of MU attestations across county characteristics, all the county rates were converted into three dichotomous measures corresponding to terciles (e.g., Q₁-tercile group with the lowest values was referred to as ‘low’ group; Q₃-tercile group with the highest values was the ‘high’ group; and Q₂-tercile group was the ‘medium’ group). We then calculated descriptive statistics to characterize our sample. We used chi-square statistics from one-way ANOVA tests to examine the bivariate relationships between MU attestation rates and county characteristics. We then used a generalized hierarchical linear model (GHLM) or mixed-effects model given that the MU attestation data were clustered at the county level (level 2) and measured at the practice level (level 1).²⁹ GHLM adjusts for biases that might exist because of possible clustering of observations (physician practices within counties), which would violate the assumption of independence of observations.²⁹

We estimated two models. Model 1 was an unadjusted or null random intercept model while Model 2 was a full or adjusted random intercept model for MU attestations. To estimate Model 2, fixed effect covariates for the county explanatory variables were added to Model 1while controlling for provider and practice characteristics, AIU phases, and vendor types. To examine the associations between county characteristics and MU attestations, the above models were estimated by pooled logistic regressions using generalized estimated equations (GEE) with the binomial family and the logit link function.

The logit link function can be represented as follows:

$$\mathrm{P}\mathrm{r}\mathrm{o}\mathrm{b}\left({\mathrm{M}\mathrm{U}}_{\mathrm{i}\mathrm{j}}=1\mid {\beta }_{0\mathrm{j}}\right)={\mathrm{u}}_{\mathrm{i}\mathrm{j}}\left(\text{Level}1\text{Model}\right)$$ (1)

$$\mathrm{l}\mathrm{o}\mathrm{g}\left[{\mathrm{u}}_{\mathrm{i}\mathrm{j}}/\left(1-{\mathrm{u}}_{\mathrm{i}\mathrm{j}}\right)\right]={\xi }_{\mathrm{i}\mathrm{j}}$$ (2)

$${\xi }_{\mathrm{i}\mathrm{j}}={\beta }_{0\mathrm{j}}$$

$${\beta }_{0\mathrm{j}}={\gamma }_{00}+{\epsilon }_{0\mathrm{j}}\left(\text{Level}2\text{Model}\right)$$ (3)

Where $\mathrm{l}\mathrm{o}\mathrm{g}\left[u_{\mathrm{i}\mathrm{j}}/\left(1-{\mathrm{u}}_{\mathrm{i}\mathrm{j}}\right)\right]={\xi }_{\mathrm{i}\mathrm{j}}$ is the probability of MU attestation for provider $i$ in county $j.{\beta }_{0\mathrm{j}}$ is the intercept for county $j$, which represents the likelihood of MU attestation by any individual provider in that county. The intercept ${\gamma }_{00}$ represents the mean county MU attestation rate without any provider characteristics. The county-level random effect is represented by ${\epsilon }_{0\mathrm{j}}$, which can be interpreted as the deviation of the intercept for county $j$ from the population mean intercept. The errors terms, ${\mathrm{u}}_{\mathrm{i}\mathrm{j}}$ and ${\epsilon }_{0\mathrm{j}}$, are normally distributed with mean 0 and variances ${\sigma }_{u0}^2$ and ${\sigma }_{\epsilon 0}^2$. The intraclass correlation is given $\rho ={\sigma }_{u0}^2/\left({\sigma }_{u0}^2+{\sigma }_{\epsilon 0}^2\right)$.

The use of GEE provides consistent estimates of effects similar to those obtained with the use of time-dependent Cox analyses.³⁰ Model 1 yielded the likelihood of attesting MU by any individual provider in a given county and some random effects that might results from external and/or internal factors.³⁰ Model 2 assumed that the average MU attestation rates across providers (level 1) may be explained by county characteristics (level 2) while allowing for random variation in intercepts across providers. Odds ratios and standard errors were derived for all the regression models using Stata.³¹

RESULTS

Descriptive analyses

Table 1 displays the descriptive statistics of the analysis sample. Among the providers, physicians represented the largest group (52.1%), followed by nurse practitioners (23.6), pediatricians (13.8%), dentists (8.1%), certified nurse-midwives (1.9%) and physician assistants (0.5%). Group practices represented 63.4 % while solo practices represented 36.6%. Approximately, 36.6% of the providers in the sample adopted a certified EHR system; 28.5% implemented a certified system; and 34.9% upgraded to a certified EHR system. Finally, we found that 41.8% of providers bought their EHR systems from large vendors while the rest bought their systems from small vendors.

Table 1.

Characteristics and enrollment data for primary care providers in the Florida Medicaid Promoting Interoperability program, 2011–2016 (N=8464)

Characteristics	N (%)
Specialty
Physician	4410(52.1)
Nurse Practitioner	1997(23.6)
Physician Assistant	42(0.5)
Certified Nurse-Midwife	161(1.9)
Pediatrician	1168(13.8)
Dentist	686(8.1)
Practice Type
Solo	3098(36.6)
Group	5366(63.4)
AIU Phases*
Adopt	3098(36.6)
Implement	2412(28.5)
Upgrade	2954(34.9)
EHR vendors^
Small vendors	4926(58.2)
Large vendors	3538(41.8)
Enrollment
2011	2641(31.2)
2012	2048(24.2)
2013	1574(18.6)
2014	779(9.2)
2015	719(8.5)
2016	703(8.3)

^*To enroll in the program, eligible providers had the option of Adopting, Implementing, or Upgrading (AIU) an EHR system.

^{^}Providers bought EHR systems and received technical assistance from small vendors (sold to fewer than 200 providers) or large vendors (sold to at least 200 providers)

Table 2 presents the unadjusted MU attestation rates by county characteristics. On average, 41.9 % of the providers attested to MU after receiving the first-year incentive. The results indicate that rural counties, counties with lower poverty rates, counties with lower HMO penetration rates, and counties with lower number of PCPs per capita had higher MU attestation rates.

Table 2.

Unadjusted rates of meaningful use of EHRs by county characteristics for the Florida Medicaid Promoting Interoperability program, 2011–2018.

County characteristics	MU rates (%)	p-value
Mean MU rate	41.9
Munificence
Rurality		<0.001
Rural	63.2
Urban	44.7
Persons 25 years and older with a bachelor’s degree		0.56
Low (8.8 – 26.2%)	45.8
Medium (26.3– 29.9%)	45.8
High (30.0 – 43.1%)	44.4
Poverty		<0.001
Low (9.4 – 16.2%)	51.8
Medium (16.3 – 20.9%)	47.4
High (21.0 – 31.0%)	32.8
Uncertainty
HMO penetration		<0.001
Low (5.5 – 15.4%)	55.6
Medium (15.5 – 25.5%)	40.7
High (25.6 – 54.1%)	38.6
Complexity
Number of primary care providers per 10,000 population		0.01
Low (1.7 – 7.2)	53.2
Medium (7.3 – 7.8)	46.0
High (7.9 – 8.1)	36.3

Hierarchical models

The results of the hierarchical models are presented in Table 3. The unadjusted random intercept model (Model 1) indicates that there is variability in MU attestation rates between counties. The intraclass correlation indicates that 8 percent of the total variance in MU attestation rates was explained by differences at the county level (p<.001).

Table 3.

Odd Ratios for random intercept models estimating meaningful use of EHRs among primary care providers in the FL-Medicaid Promoting Interoperability program

	GHLM Model 1: Unadjusted Model β Coefficient (SE)	GHLM Model 2: Full Model β Coefficient (SE)
Fixed part
Intercept	0.83(0.006)	0.45(0.10)
Rurality
Urban		Reference
Rural		2.45(0.53)
Persons 25 years and older with a bachelor’s degree
Low		Reference
Medium		0.95(0.18)
High		1.44(0.28)
Poverty rates
Low		Reference
Medium		0.98(0.13)
High		0.58(0.10)
HMO penetration
Low		Reference
Medium		0.82(0.14)
High		0.60(0.13)
Primary care providers per 10,000 population
Low		Reference
Medium		1.07(0.16)
High		0.61(0.09)
Level 1 Variables
Specialty
Physician		Reference
Nurse Practitioner		0.82(0.06)
Physician Assistant		0.54(0.22)
Certified Nurse-Midwife		0.94(0.18)
Pediatrician		2.64(0.21)
Dentist		0.16(0.03)
Practice type
Solo		Reference
Group		1.59(0.10)
AIU Phases
Adopt		Reference
Implement		1.92(0.14)
Upgrade		1.44(0.10)
EHR procurement
Small vendors		Reference
Large vendors		1.17(0.07)
Random Part
Intraclass correlation (ρ)	0.08

Note: Boldface values are statistically significant at the p=0.05 level.

The results of the adjusted model (Model 2) indicate that providers located in rural counties were 145% more likely to attest to MU compared to providers located in urban counties (P<.001). Using the categorizations of tercile groups as discussed earlier, we found that providers located in counties with ‘high’ poverty rates were 42 percent less likely to attest to MU compared to providers located in counties with ‘low’ poverty rates (P=.002). Providers located in counties with ‘high’ HMO penetration rates were 40 percent less likely to attest to MU compared to providers located in counties with ‘low’ HMO penetration rates (P=.02). Providers located in counties with a ‘high’ number of PCPs per capita were 39 percent less likely to attest to MU compared to providers located in counties with a ‘low’ number of PCPs per capita (P=0.01).

Most of the control variables were significant in the adjusted regression model. More specifically, compared to physicians, pediatricians were more likely to attest to MU (P<.001) while dentists (P<.001) and nurse practitioners (P=.003) were less likely to attest to MU. Group practices were more likely to attest to MU compared to solo practices (P<.001). Compared to providers that adopted EHR systems, those who implemented EHRs (P<.001) or upgraded EHRs (P<.001) were more likely to attest to MU. Practices that bought their EHR systems from large vendors were more likely to attest to MU compared to those that bought their systems from small vendors (P=.001).

DISCUSSIONS

The results indicate that providers in rural counties were more likely to attest to MU than providers located in urban counties. A plausible economic explanation is that the incentive payments received by the smaller providers in rural counties represent a significantly higher marginal benefit compared to the larger practices in urban counties. Another plausible explanation is the proliferation of regional extension centers (RECs) in rural areas during the study period. The REC program was established by the HITECH Act and administered through the Office of National Coordinator (ONC) for Health Information Technology. This program included 62 grantee organizations throughout the U.S. that promoted EHR adoption and the optimal use of EHR technology through outreach and the provision of technical assistance.³² The REC program prioritized smaller and rural providers and was very successful in engaging these providers.^3,33,34 In addition to the REC program, several health IT workforce training programs existed in rural areas. For example, in 2013 the Health Services and Resource Administration (HRSA) granted $5.3 million to fund educational and training programs to expand the rural health IT workforce.³⁵ Other government programs developed resources, checklists, and toolkits to help practices in these areas.³² These findings are encouraging and run contrary to those of most previous studies^36,37 although more recent studies investigating practices that participated in the Medicaid EHR incentives programs found similar results.³³ This suggests that initiatives such as the REC program and other health IT assistance programs in rural areas can help address the long-standing urban-rural divide in EHR adoption.³³

The results found significantly negative associations between MU and poverty rates, penetration rates, and the number of PCPs per capita only when the ‘high’ terciles were compared to the ‘low’ terciles. These results are consistent with the EHR adoption literature. Indeed, previous studies found negative associations between high poverty rates and EHR adoption suggesting that PCPs in counties with high poverty face greater challenges to EHR adoption, which might be related to tighter operating budgets and limited capacity for integrating health IT training into the workflow.^36,38 But others found mixed results^39,40. Also, while studies found mixed results for the associations between HMO penetration and the adoption of a healthcare technology,^20,41 a recent study conducted in Florida found a negative association between HMO penetration and the adoption of computerized provider order entry (CPOE) systems.⁴² Finally, the finding that the high number of PCPs per 10,000 population was negatively associated with MU attestations can be similarly explained by the presence of RECs and other programs that were implemented to help small practices in low-resource urban areas as discussed above.^32,33,43

This study might provide useful information to current initiatives to increase health IT use in underserved communities. For example, HRSA is presently supporting several initiatives of EHR adoption in community health centers and rural health clinics, which are important safety net providers of care for traditionally underserved populations.⁴⁴ In 2021, the American Rescue Plan awarded over $47.2 million through the Rural Public Health Workforce Training Network (RPHWTN) program that targets four key critical training areas, including rural health IT workforce and telehealth technical support.⁴⁵ The results from this study can provide insights to these initiatives and the like across otherwise similar areas. They can also can provide insights into the larger field of health informatics regarding factors that can influence an unintended “digital divide” for providers in medically underserved communities.⁴⁵

This study has an important limitation that is worth noted. It used data from the state of Florida and, as a result, the generalizability to other states should be done with caution. Nonetheless, Florida is a relevant state for this study for several reasons. First, Florida was among the states that received the highest number of RECs.^3,32 Second, Florida ranks among the states with the highest number of Medicaid enrollments (over 6.7 million recipients as of 2023). Third, it is the third-largest state by population size (22.2 million in 2022) in the U.S.⁴⁶

Previous research found negative associations between EHR adoption and the proportion of Medicaid patients treated at a facility.^13,47 It is thus critical to assess whether Federal funding designed to promote MU of EHRs among physician practices that care for Medicaid patients is effective rather than inadvertently exacerbates health IT disparities.^48,49 Policies regarding increasing EHR adoption have typically focused on providing financial support to physician practices. While this is an important factor for many providers, particularly the Medicaid providers, the findings for this study suggest that policy makers should not ignore the contribution of healthcare market conditions in EHR adoption. Healthcare managers should also assess the effects of their specific markets when making decision on the use of their EHRs.⁵⁰ Further research is needed to further investigate geographic characteristics in which EHR adoption initiatives can be leveraged to optimize technology use among underserved populations.