Rural Populations and Early Periodic Screening, Diagnosis, and Treatment Services: Challenges and Opportunities for Local Public Health Departments

Abstract

Objectives. We examined geographic differences in Early Periodic Screening, Diagnosis, and Treatment (EPSDT) visits as the South Carolina Department of Health and Environmental Control (SCDHEC) transitioned from direct service provision (DSP) to assuring delivery within the larger health care system.

Methods. We examined infant cohorts with continuous Medicaid coverage and normal birth weights from 1995 to 2010. Outcome variables included any EPSDT visit and the ratio of observed to expected visits. Change in SCDHEC market share over time by residence was the primary variable of interest. We used growth curve models to examine changes in EPSDT visits by rural areas and levels of DSP over time.

Results. A small proportion of the study population (10%) resided in rural counties that were more dependent on SCDHEC for DSP. The trajectory of not having visits among counties with high DSPs was steeper in rural areas (0.208; P = .001) compared with urban areas (0.145; P = .002). In counties with high DSPs, the slope of the predicted ratio in rural areas (–0.033; P < .001) was steeper than that of urban areas (−0.013; P < .001).

Conclusions. Health departments operations continue to transition from DSP, which might decrease access to well-child care in rural communities. Health care reform provides opportunities for health departments to work with community partners to facilitate DSP from public to private sectors.

The release of the Institute of Medicine (IOM) report, The Future of Public Health, ushered in a new paradigm for the organization and delivery of public health services that is increasingly focused on population-based services.^1,2 Local health departments (LHDs) are under increasing pressure to reorient their service mix toward the core public health functions of assessment, assurance, and policy development while de-emphasizing the provision of clinical services for individuals.^2,3 A more recent IOM report further reflects the shifting focus, urging LHDs to work with other community-based providers to develop the capacity for clinical service delivery outside of the health department. In addition, the recommendations suggest a codification of the transition by reallocating funds from clinical services toward population-based services.^4,5

Despite the shifting paradigm, many state and LHDs remain engaged in direct service provision (DSPs), particularly in rural and underserved communities.^6–8 Once an established source of care for vulnerable populations, LHDs may find it difficult to discontinue DSP and transition to a population-based service delivery model.^9,10 One key, but largely unanswered question about this transition away from DSP, is what happens in communities when LHDs stop providing primary care and clinical services?

We examined this important policy issue in a single state whose health department transitioned away from DSP. In response to federal and state medical home initiatives and ongoing fiscal constraints, the South Carolina Department of Health and Environmental Control (SCDHEC) transitioned from directly providing Early Periodic Screening, Diagnosis, and Treatment (EPSDT) services in health departments, to assuring their delivery within the larger health care system. Precipitation of the transition varied by time and geography, with more deliberate transitions occurring in jurisdictions with adequate primary care supports. However, in communities with historically poor primary care infrastructures, the discontinuation of DSP occurred slowly, reflecting a greater reliance on LHDs for providing these services.

We examined how the level of EPSDT services changed in communities as LHDs transitioned from DSP to assuring their delivery in the larger health care system, and if the observed changes were consistent among rural and urban counties. We expected some level of service disruption during the early phase of SCDHEC’s transition away from DSP; however, we hypothesized that the impact of this transition would more negatively affect access to EPSDT services for infants living in rural counties. We informed public policy discussions on the organization and delivery of public health and primary care services in rural and underserved communities.

METHODS

We used birth cohorts of infants (0–12 months of age) enrolled in Medicaid from 1995 to 2010 for the analysis. We limited the study cohorts to infants with continuous Medicaid coverage during infancy, who were either income or categorically eligible for Medicaid; thus, we effectively included infants covered by Medicaid whose household income was less than 185% of the federal poverty level. To avoid potential bias from early life birth experiences influencing EPSDT service utilization, we further restricted the cohorts to infants with normal birth weights (> 2500 g). Infants who died or were otherwise disenrolled from Medicaid were also excluded.

Our data were derived from linked South Carolina Medicaid eligibility files, Medicaid billing claims, and birth certificates. We obtained linked de-identified files from the South Carolina Revenue and Fiscal Affairs Office. The Revenue and Fiscal Affairs Office serves as a repository for most health and human services data in South Carolina.

Dependent Variables

The American Academy of Pediatrics has an established periodicity schedule for childhood EPSDT visits.¹¹ During the first 12 months of life, infants are expected to receive 8 EPSDT visits (newborn, 3–5 days, 1, 2, 4, 6, 9, and 12 months of age). We tracked the EPSDT visits for each infant in the cohorts through Medicaid claims data using the following Common Procedural Terminology billing codes: 99381, 99382, 99391, and 99392.

We used 2 EPSDT outcome variables of interest to measure changes in visits over time. We coded any EPSDT visit during infancy as dichotomous (yes/no). In addition, we calculated the ratio of observed to the expected number of EPSDT visits⁸ during infancy; we used this ratio to gauge continuity of EPSDT visits. We treated this as a continuous variable, with values between 0.00 and 1.00 in the analysis. We added an additional month to allow for small variations in billing around the 12-month visit. Infants with observed visits beyond the expected value were treated as receiving the 8 recommended services.

Independent Variables of Interest

The interaction of changes in the proportional volume of EPSDT services provided by SCDHEC, or market share, and rural residence over time was our primary interest. SCDHEC market share is a county-level measure that was derived by dividing the total number of EPSDT visits provided by SCDHEC by the total number of EPSDT visits among all providers for each county and year of the cohort (1995–2010). We calculated the state average for SCDHEC market share at baseline (1995), and we used values 1 SD above and below the mean state value to create a 3-level categorical variable that indicated counties with low (≤ 19% of visits), average (20%–59% of visits), and high (≥ 60% of visits) SCDHEC market shares.

Rural residence was also a variable of interest. Theoretically, the subsequent influence of changes in SCDHEC market share on the selected outcomes over time could also vary significantly by residence. We used urban influence codes for the county of residence at delivery to identify infants who were residents of rural and urban counties. Urban influence codes 1 to 3, 5, and 8 included infants who were residents of urban and micropolitan (urban) counties, whereas urban influence codes 4, 6, 7, and 9 to 12 included small adjacent and small remote rural counties.¹²

Covariates

We used the Anderson’s Behavioral Model for Health Services Use as the conceptual framework for selecting fixed- and time-invariant covariates.¹³ The Anderson model conceptualizes health utilization as a function of predisposing characteristics, enabling factors, and person-level actual or perceived need for services. Predisposing time-invariant characteristics included the mother’s race (non-Hispanic White, non-Hispanic Black, other), mother’s ethnicity (Hispanic, non-Hispanic), mother’s age (< 20 years, 20–25 years, 26–34 years, > 34 years), mother’s marital status (yes/no), and mother’s education (high school or less, some college or technical school, college graduate or above).

We included enabling factors in the analysis to account for other potential health resources available within a given county. These variables were time-variant and included the market share for EPSDT by federally qualified health centers and rural health clinics. We calculated these measures as the total number of EPSDT visits provided by each divided by the total number of EPSDT visits, respectively. We treated these as a continuous variable in the analysis. Additional time-variant measures for Medicaid growth (defined as the percent increase in Medicaid enrollment by year) and Medicaid managed care market share (defined as the percent of infants enrolled in a Managed Care plan) were derived and treated as continuous variables. We included quartiles of the pediatrician to enrollee ratio as a measure of physician availability at the county level. In measures of individual need, we identified children with special health care needs to account for potential differences in underlying health status that potentially influenced receipt of EPSDT services.¹⁴

Analysis

We conducted a bivariate analysis that described categorical variables for the SCDHEC market share for EPSDT DSP and rural residence using THE χ² test for independence and an unadjusted multinomial logistic regression model. We estimated and interpreted growth curve models for the multivariable analysis. Growth curve models are a form of general linear mixed models and are an increasingly common vehicle for analyzing longitudinal data with repeated measures.^15,16

We fit 2 models for each outcome variable using fixed and random effects. We included an interaction term for time and SCDHEC market share in the model to allow for testing of the hypothesis that changes over time in the observed outcomes were influenced by changes in SCDHEC market share for EPSDT at the county level. We included a subsequent 3-way interaction term, including rural residence, in the model to test whether the observed changes also varied systematically by rural residence. We added county random effects to the models to account for potential unobserved heterogeneity at this level. We added a random coefficient for time to the model to allow for variation in the trajectory of the outcomes between counties. We specified random effects with an unstructured covariance matrix (that is, no specific relationship was assumed between the random effects and the overall variance component), and we evaluated a likelihood ratio test versus a model without random effects to assess model selection. Likelihood ratio tests suggested significance between county clustering and supported the use of more advanced models, including random effects.

We included the previously mentioned additional time-invariant and time-variant covariates as fixed effects covariates in the models. We also added a quadratic term for time to account for nonlinear growth in observed outcomes. We completed the analysis using Stata.¹⁷ Mixed effect logistic models for having no visit (dichotomous) were fit using XTMELOGIT and mixed effect linear models for the ratio of observed to the expected visits (continuous) were fit using XTMIXED. We used the margins command to estimate marginal predicted probabilities (marginal means) for each selected outcome by year, SCDHEC market share, and rural residence.

RESULTS

Table 1 provides the distribution of EPSDT visits by infants at baseline (1995) by SCDHEC market share and rural residence. A relatively small proportion (approximately 10%) of the study population resided in a rural county. However, rural populations were heavily dependent on SCDHEC as DSPs. The proportion of EPSDT visits provided in counties that were heavily dependent on the SCDHEC was higher among infants in rural areas (27.93%) compared with those who resided in urban areas (16.93%; P < .001). In counties with an average SCDHEC market share, 64.09% of EPSDT visits were provided among rural residents compared with 58.91% in urban areas (P < .001). A much higher proportion of EPSDT visits in counties with low SCDHEC market share was provided in urban areas (24.16%) relative to 7.98% among rural residents (P < .001). As expected, results from the unadjusted multinomial logistic regression indicated that the likelihood of having an average SCDHEC market share was greater in rural counties compared with counties with a low SCDHEC market share (odds ratio = 1.18; 95% confidence interval = 1.02, 1.35). Rural counties were markedly more likely to have a high SCDHEC market share relative to those with a low SCDHEC market share (odds ratio = 1.59; 95% confidence interval = 1.41, 1.78).

TABLE 1—

EPSDT Visits Received at Baseline by Level of SCDHEC Market Share and Rural Residence: South Carolina, 1995–2010

SCDHEC Market Share	Total (n = 18 585), No.	Urban (n = 16 585), No. (%)	Rural (n = 1 980), No. (%)	Unadjusted OR (95% CI)
Low (Ref)	4 165	4 007 (24.16)	158 (7.98)	1.00
Average	11 040	9 771 (58.91)	1 269 (64.09)	1.18 (1.02, 1.35)
High	3 360	2 807 (16.93)	553 (27.93)	1.52 (1.41, 1.78)

Note. CI = confidence interval; EPSDT = Early Periodic Screening, Diagnosis, and Treatment; OR = odds ratio; SCDHEC = South Carolina Department of Health and Environmental Control. Low is defined as ≤ 19% of visits, average as 20%–59% of visits, and high as ≥ 60% of visits.

Initial Provider Contact

As noted in Table 2, the probability that an infant did not receive a visit increased over the duration of the study (0.212; P < .001); however, the trajectory of the initial increase slowed over time (−0.016; P < .001). Infants who resided in counties with an average SCDHEC market share had a lower probability of not having EPSDT visits (−0.599; P < .001) compared with infants in counties with a low SCDHEC market share at baseline. The probability of not having an EPSDT visit was much lower among infants who resided in counties with a high SCDHEC market share at baseline (−0.901; P < .001). Although not significant, infants who resided in rural areas also had a lower probability of not having an EPSDT visit at baseline (−0.326; P = .126).

TABLE 2—

Predicted Probabilities for Selected EPSDT Outcomes and Covariates of Interest: South Carolina, 1995–2010

Characteristics	No Visita	P	Ratio of Observed:Expected Visitsb	P
Fixed effectsc
Intercept	−3.00	< .001	0.493	< .001
Year	0.212	< .001	−0.009	< .001
Year x Year	−0.016	< .001	0.001	< .001
SCDHEC market shared
Low	Ref		Ref
Average	−0.599		0.082
High	−0.901	< .001	0.107	< .001
Residence
Urban	Ref		Ref
Rural	−0.326	.126	0.038	.057
Year x SCDHEC market share x rural
Low—urban	Ref		Ref
Low—rural	0.061	.006	−0.008	< .001
Average—urban	0.096	< .001	−0.018	< .001
Average—rural	0.141	< .001	−0.027	< .001
High—urban	0.145	.011	−0.013	< .001
High—rural	0.208	.002	−0.033	< .001
Random effectse
SD, y	0.061		0.007
SD (intercept)	0.561		0.057
Correlation, y (intercept)	−0.483		−0.554

Note. EPSDT = Early Periodic Screening, Diagnosis, and Treatment; SCDHEC = South Carolina Department of Health and Environmental Control.

^aModel derived using Stata XTMELOGIT.

^bModel derived using Stata XTMIXED.

^cAll models included additional fixed effects for race/ethnicity, mother’s age, education, marital status, special need status, federally qualified health centers or rural health center market share, private provider capacity, managed care market share, changes in Medicaid enrollment, and reimbursement for EPSDT.

^dLow is defined as ≤ 19% of visits, average as 20%–59% of visits, and high as ≥ 60% of visits.

^eAll models included random effects for county and year.

Results from the 3-way interaction suggested that the impact of SCDHEC’s transition was not equitable by residence, even in counties with similar levels of SCDHEC market share. We noted differences by rural residence (0.061; P = .006) in counties with a low SCDHEC market share. However, the overall probability was relatively low compared with what was observed among infants in counties with an average and a high SCDHEC market share. Compared with the reference population, the slope of not having EPSDT visits was greater in infants who resided in rural counties with an average SCDHEC market share (0.141; P < .001) than what was observed in urban counties that also had an average SCDHEC market share (0.096; P < .001). In counties with a high SCDHEC market share, the slope of not having a visit in rural counties (0.208; P = .001) was markedly higher than what was observed in urban counties that were also heavily dependent on SCDHEC as a service provider (0.145; P = .002). Figure 1 shows the predicted probabilities (marginal means) from the 3-way interaction. The correlation between the intercept and slope (−0.483) in the random part of the model suggested that the infants in counties with a higher probability of not having an EPSDT visit at baseline tended to have larger negative slopes than other counties over time.

FIGURE 1—

Predicted probability of not having an EPSDT visit by residence and (a) low SCDHEC market share, (b) average SCDHEC market share, and (c) high SCDHEC market share: South Carolina, 1995–2010.

Note. EPSDT = Early Periodic Screening, Diagnosis, and Treatment; SCDHEC = South Carolina Department of Health and Environmental Control.

Comprehensiveness of Visits

Because it was related to the comprehensiveness of EPSDT visits, the impact of SCDHEC transitioning from DSP of EPSDT services appeared to be marginal. As noted in Table 2, the predicted ratio of observed to expected visits decreased over time (−0.009; P < .001); however, the trajectory of the decline slowed significantly over time (0.001; P < .001). The predicted ratio at baseline was higher among infants who resided in counties with an average SCDHEC market share (0.082; P < .001) and infants who resided in counties that were heavily dependent on SCDHEC (0.107; P < .001) relative to their peers who resided in counties with a low SCDHEC market share at baseline. On average, infants who resided in rural counties had a higher predicted ratio relative to their urban peers at baseline; however, this coefficient was marginally nonsignificant (0.038; P = .057).

Results from the 3-way interaction suggested the impact of SCDHEC’s transition from DSP was not equitable by residence. In counties with low SCDHEC market share, differences in the negative slope were significant, but the magnitude of the association was marginal (−0.007; P < .001). However, substantial deterioration in the predicted ratio of observed to expected EPSDT visits among infants in counties with an average and a high SCDHEC market share was noted. Among counties with an average SCDHEC market share, the negative slope of the predicted ratio in rural counties (−0.027; P < .001) was greater than what was observed in urban counties (−0.018; P < .001). Considering counties with a high SCDHEC market share, the negative slope of the predicted ratio in rural counties (−0.033; P < .001) was markedly steeper than what was observed in urban counties with high SCDHEC market share (−0.013; P < .001). Figure 2 shows the predicted probabilities (marginal means) from the 3-way interaction. The majority of the improvement observed statewide occurred among infants who resided in counties with a low SCDHEC market share. The correlation between the intercept and slope (−0.554) in the random part of the model suggested that infants in counties with a higher ratio of observed to expected visits at baseline tended to have larger negative slopes than other counties over time.

FIGURE 2—

Predicted ratio of observed to expected EPSDT visits by residence and (a) low SCDHEC market share, (b) average SCDHEC market share, and (c) high SCDHEC market share: South Carolina, 1995–2010.

Note. EPSDT = Early Periodic Screening, Diagnosis, and Treatment; SCDHEC = South Carolina Department of Health and Environmental Control.

DISCUSSION

We examined the extent to which changes in EPSDT services provided by health departments varied by rural residence as SCDHEC transitioned from DSP to assuring their delivery within the larger health care system. In urban counties, the transition resulted in improved EPSDT visits during the last 5 years of the study, which was encouraging for public health leaders and policymakers who argued that DSP by health departments was an inefficient way of improving well-child care services. Unfortunately, in rural counties that were more reliant on SCDHEC for clinical service provisions, receipt of any EPSDT service deteriorated over time. The level of EPSDT services provided in rural areas did not return to pretransition levels, particularly in counties that were heavily dependent on SCDHEC.

Similar results were observed when considering the comprehensiveness of EPSDT visits. The impact of the SCDHEC transition in urban counties was marginal, regardless of the level of SCDHEC market share. However, the impact on infants in rural counties with average and high dependence on SCDHEC as a service provider was substantial. The comprehensiveness of EPSDT services provided to infants in these counties deteriorated over time and did not demonstrate signs of improvement.

It was probable that deficiencies in the primary care infrastructure in rural counties contributed significantly to the observed results. LHDs that operated in a more urban setting presumably had more local primary care infrastructure to absorb the increased demand stemming from the loss of SCDHEC as a service provider—even when SCDHEC was relatively active early in the study period. Rural and underserved communities were historically more dependent on SCDHEC as a provider of direct services. Our findings suggested that the existing primary care infrastructure in these communities was not sufficient to offset the increased demand from the loss of SCDHEC as a DSP.

Our study was not without weaknesses. Although the methods we used to analyze the data were among the strongest available, the data used for this research were derived from administrative claims. These data were not collected for research purposes. Therefore, the strength of conclusions drawn from the study was contingent on the quality of the data, how well the models fitted the data, and how well the specified models were able to explain observed changes in unmet need over time. Also, we were limited to only variables available from administrative claims and secondary sources. It was that possible additional variables relevant to our theoretical framework were not included in the analysis. We also did not analyze the quality of the transition (which was not germane to our findings)—specifically how well SCDHEC negotiated with local primary care providers to increase their pediatric patient populations enrolled in Medicaid.

Despite its limitations, our study made demonstrable policy and programmatic contributions. We exploited the opportunities made available by historic changes in the organization and delivery of public health services in South Carolina. This was a cost-effective means of examining complex issues regarding the organization and delivery of public health services. Moreover, we used advanced longitudinal data analysis methods, which improved the level of causal interference that could be drawn from our findings. Our approach might provide a blueprint for states to evaluate the impact of LHDs transition away from DSPs.

Policy Implications

Although our study was conducted in a single state, our findings were directly relevant for informing larger discussions on the organization and delivery of public health services in rural and underserved areas. Our findings were particularly relevant in the context of health care reform, because implementation provides both challenges and opportunities for LHDs wishing to discontinue clinical services.

Infants included in our study were restricted to only those continuously enrolled in Medicaid during the first year of life. The marked differences in the use of EPSDT services in rural communities observed in our study were not a function of insurance; they were a function of access. Although health care reforms are instrumental for expanding access to health care insurance, the effectiveness of these reforms may be constrained by the capacity of existing community-based primary care.

Previous research clearly noted LHDs operating in rural jurisdictions were often an integral component of the existing community-based primary care system.⁸ According to a recent survey by the National Association of County and City Health Officials (NACCHO), 36% of LHDs provided EPSDT services directly, 32% provided well-child visits, 54% provided family planning services, 27% provided prenatal care, 90% provided childhood immunizations, and 11% provided comprehensive primary care services.¹⁸ As the demand for preventive and primary care services increases, the extent to which existing providers have the capacity to absorb the increased demand remains a priority area for policy development. Faced with increasing demand for direct services, LHDs operating in rural and underserved areas might find it difficult to undertake this transition away from DSP. Furthermore, our results suggested that the impact of LHDs discontinuing DSP in these communities might be detrimental unless steps are taken to address the existing constraints on community-based primary care capacity.

Conclusions

As states redefine the roles of their public health agencies relative to DSP, strategic planning that accounts for the vulnerabilities of early childhood systems of care in rural areas should be considered carefully. Disruption of systems could be minimized if key policy partners are engaged, such as State Offices of Rural Health, State Offices of Primary Care, and state primary care, rural health, and hospital associations. These policymaking and advocacy organizations have unique insights into the vulnerabilities of rural health care systems. They represent organizations that are incentivized by the Centers for Medicare and Medicaid Services to care for vulnerable and underserved populations. Their involvement, as well as exploiting opportunities available through health care reform, such as expansion of federally qualified health centers and patient-centered medical home designations,¹⁹ could enable public health’s transition from DSP to systems facilitation. A statewide strategy that includes localized systems planning could minimize the disparities in access to EPSDT experienced by rural communities that are notorious for having sparse primary care capacities.

Although our study’s results might not be generalizable to every state, they were innovative in scientific contributions to a policy issue that has not been explored. We hope our findings inform policy discussions that lead to the fulfillment of IOM’s call to action that public health serve as a facilitator of systems development for all people, including those with vulnerabilities, so that all children have the opportunity to achieve their potential.