Abstract
This study examined whether access to and use of patient-centered medical home (PCMH) practices is associated with reduced nonurgent emergency department (ED) use, especially among the uninsured population. This retrospective study used 2010-2011 Medical Expenditure Panel Survey data. Difference-in-difference methods, multivariate logit model, marginal effect, and survey procedures were employed. A total of 1287 adults had any ED visit in 2011, which represented weighted 29 463 684 people in the population. Reductions in odds of nonurgent ED use between the full PCMH group and the “no regular provider” group was significantly larger for the uninsured group than publicly and privately insured groups (β = −1.70, P = .009, and β = −1.04, P = .040, respectively). Similar results were found between the partial PCMH group and the “no regular provider” group for the uninsured group compared to the public group (β = −1.67, P = .019). PCMH models demonstrate higher odds of reduced nonurgent ED use among uninsured individuals compared to public and private enrollees nationwide.
From 1997 through 2007, emergency department (ED) visits in the United States increased by 23% to a total of nearly 117 million visits per year. 1 Near half of these visits were for nonurgent medical care or were potentially preventable.2-4 These visits increase ED overcrowding, wait times, and financial strain on public health insurance programs.5,6 Stakeholders including health systems, physicians, and payers have devised various interventions to discourage nonurgent ED visits, such as patient education, financial disincentives for ED visits, encouragement of provision of primary care physician (PCP) services on evenings and weekends, and an increase in PCP supply. Despite these efforts, nonurgent ED visits have continued to rise, 7 which warrants an examination of the underlying reason and development of effective interventions.
Persistent nonurgent ED use may represent a deficiency in the primary care or ambulatory care system. A well-functioning primary care system can have the capacity to provide timely, adequate, and effective care for patients to avoid nonurgent ED use. In general, higher levels of primary care capacity are associated with lower rates of ED utilization. 8
Recently, the patient-centered medical home (PCMH) has emerged as an innovative health care delivery model that holds the conceptual promise to improve health care quality and patient satisfaction. Although there are variations in the definition and measurements of this model, the PCMH generally includes the following key principles: a personal physician, a physician-directed medical practice, a whole person orientation, coordination and/or integration of care, quality and safety as hallmarks, enhanced access to care, appropriate payment and incentives, and wide-ranging team-based care.9,10 Described as a “lifeline for primary care,” the PCMH has the potential to transform and increase the appeal and viability of primary care practice 11 and to increase patient satisfaction. 12 Given that the PCMH model can conceptually improve primary care quality and patient satisfaction, it may potentially reduce nonurgent ED use.
Furthermore, compared to insured people, uninsured individuals usually have less access to adequate health care, and are denied by some primary care doctors and specialists because of the underpayment concern, while the ED is always open and cannot refuse patients regardless of their ability to pay. Thus, uninsured individuals may be more likely to use ED services for nonurgent conditions. Given this context, access to PCMH models may be more important in reducing nonurgent ED use for uninsured people by improving their access to and use of primary care, because PCMH models have features such as coordination and integration of care, a whole person orientation, and enhanced access to care.
Although an important topic, there is little literature examining the effect of PCMH models on nonurgent ED use. Although several studies have found an association between PCMH and fewer ED visits,13-19 these studies examined overall ED use rather than nonurgent ED use. Quality primary care can only affect nonurgent ED use that can be addressed outside of the ED setting, but has little effect on ED care for urgent conditions such as accidents or injuries, which are legitimate uses of ED services. In such a context, differentiation of types of urgency of ED use would precisely reflect the role of primary care and PCMH models. In addition, these studies examined their research questions within regional health systems, in pilot demonstration programs, or among certain subpopulations, such as chronically ill patients or Medicare enrollees, thus making these results less comparable and generalizable.13-19 Thus far, only one study by Hochman et al examined the effect of a PCMH intervention on nonurgent ED use 20 in one safety net clinic in California, and the authors found insignificant results between PCMH and preventable ED use.
The current study advances the policy debates in that it is the first to empirically examine whether self-reported access to and use of PCMH practices was associated with reduced nonurgent ED use in a national sample. Furthermore, this study also examined whether the general effect of PCMHs on potentially reduced nonurgent ED use was more pronounced among the uninsured population than the insured population.
Methods
Data
Medical Expenditure Panel Survey (MEPS) data were used for this study. 21 MEPS is a nationally representative survey of the noninstitutionalized civilian US population and is designed to produce national estimates of health care use, costs, sources of payment, and insurance coverage.
Each new panel covers a series of 5 rounds of in-person interviews. 22 This design, which covers 2 full calendar years, allows the analysis to track individuals’ preferences and health care utilization over time. Like many other national survey designs, MEPS adopts a complex multistage, unequal probability, and cluster sampling study design. 23 Because Hispanics, African Americans, Asians, and indigent populations have been oversampled to increase statistical power and improve the precision of estimates for specific subgroups, sample weights have been provided to allow for calculation of population estimates.
Study Population
This study employed a retrospective design using the MEPS 2010-2011 panel. Individuals were included if they were ages 18 years and older, had any ED visit in 2011, and had data from all 5 survey rounds.
Outcome Variables
The outcome measure was urgency of ED use in 2011. The study by Sarver et al used MEPS data to measure nonurgent ED use. 24 Specifically, a visit was considered to be urgent if (1) it resulted in an admission; (2) the patient received an X-ray, magnetic resonance imaging, electrocardiography, electroencephalography, or any surgical procedure, and the patient reported the reason for the visit was an “accident or injury,” diagnosis, or treatment, and if it was an office or clinic visit, the visit was not the result of referral; or (3) the reason for the visit was an “accident or injury,” diagnosis, or treatment, and the visit was within 3 days of the “accident or injury” or onset of symptoms. The remaining visits were classified as nonurgent. The present study adopted the same approach to classify ED use types, but made a minor revision by deleting the component “if it was an office or clinic visit, the visit was not the result of referral” within the second criterion to improve the construct validity of nonurgent ED use. This revision was made because the present study focused on nonurgent or urgent health care utilization in the ED setting instead of an ambulatory care setting.
Independent Variables
Consistent with the key features of PCMH models, patient reports of practice characteristics and their experiences with care were used to develop robust indicators of PCMH practices. 25 The present study followed Beal et al’s approach in defining PCMHs 26 as MEPS respondents reporting (1) having a regular provider; (2) the provider’s role in the total care of the patient (ie, new health problems, preventive health care, ongoing health problems, referrals to other health professionals); (3) patient engagement in care (provider asks about medications and treatments prescribed by other doctors or asks respondent to help decide treatment); (4) care accessibility (able to contact providers during regular business hours, at night, or on weekends). The present study adopted the same approach to capture the main features of the PCMH, with the minor revision of deleting the criterion “Has no difficulty contacting regular source of care over the telephone during regular business hours” to improve the face validity of PCMH models.
Consistent with the Beal et al study, the present study further categorized respondents into 3 groups: (1) having a PCMH (those who said yes to all these 4 components); (2) having a regular source of care that is not a full PCMH (hereafter, “partial medical home” [eg, those with a regular source of care, but who said “no” to any of the other 3 components]); and (3) having no regular source of care (those who reported no regular provider). In addition, an unknown PCMH category also was added that included the values of “not ascertained,” “don’t know,” and “inapplicable” in any of the 4 components. PCMH status was measured in the baseline year 2010.
Control Variables
This analysis controlled for several factors in baseline year 2010 that could correlate to nonurgent ED use, including a person’s age, sex, race, ethnicity, rural/urban location, marital status, education, and insurance status. All predictors were treated as dummy variables because their effects on ED use may not change in a linear way. All these covariates were measured in baseline year 2010.
Statistical Analysis
There is a potential selection issue between nonurgent ED use and access to and use of PCMHs because patients with certain characteristics (eg, good health) may intentionally have low levels of nonurgent ED use and more interest in accessing PCMHs.
This can lead to biased estimates of the impact of the PCMHs on the outcomes of interest. The authors attempted to attenuate this problem by including controls for health status, as well as using the lagged time effect in the regression model. The authors measured health status using the physical component score derived from the SF-12 in the Medical Outcomes Study to reflect health-related quality of life. 27 The authors used the approach of Harman et al, 28 who used top tertile and bottom tertile scores in actual data distribution as cutoff points to divide their study sample into 3 condition severity groups. The bottom tertile was composed of sick individuals, and the top tertile represented healthy individuals and the reference group. The lagged time effect also was used to account for the selection issue, because later ED use urgency in 2011 could not affect access to and use of PCMHs in the earlier period of 2010.
The authors estimated multivariable logit models to analyze urgent and nonurgent ED use. The authors further estimated the marginal effect of significant independent variables to indicate the magnitude of change in predicted probability, in addition to the odds ratio between different categories of PCMH variable, holding all other covariates at the mean level.
The analysis essentially takes a difference-in-difference (DID) approach, comparing the difference in odds of nonurgent ED use between the full PCMH group and the “no regular provider” group for uninsured individuals versus insured individuals with either public or private coverage. Specifically, the authors created a set of interaction terms between PCMH dummy variables and insurance dummy variables.
To account for clustering in the sampling design and obtain correct standard errors, the survey procedures in STATA version 13 (StataCorp LP, College Station, Texas) were used to conduct the analysis. They account for the complex weight and variance in the sampling design and yield nationally representative results.
Results
Study Sample Characteristics
Demographic and socioeconomic characteristics of the sample used in the analysis are presented in Table 1. The final study sample consisted of 1287 adults with at least one ED visit in 2011, which represented a weighted 29 463 684 people in the total population. Among them, 390 individuals (30.3%) had nonurgent ED use, representing a weighted 8.3 million people in the total population.
Study Sample Characteristics (N = 1287).
Abbreviation: PCMH, patient-centered medical home.
Among the study sample, there were 498 individuals in the full PCMH group, 228 individuals in the partial PCMH group, 257 individuals in the unknown PCMH group, and 304 individuals in the “no regular provider” group. They represented the weighted population size of 12 million, 5.0 million, 6.7 million, and 5.9 million individuals, respectively, and a total of 29 million people in the United States.
Demographic and socioeconomic characteristics of the PCMH groups are presented in Table 2. PCMH group status was statistically associated with insurance types (χ2 = 140.3, P < .001; population inference F = 15.1, P < .001). For example, in the full PCMH group, the insurance type distribution between the uninsured group and the insured group that included both public and private patients (11.2% for the uninsured vs 88.8% for the insured group) was different from that in the “no regular provider” group (39.8% vs 60.2%).
Characteristics of Patient-Centered Medical Home (PCMH) Groups a (N = 1287).
The frequency and percentage refer to column frequency and percentage. For example, the frequency and percentage of each insurance type within the full PCMH group.
Multivariate Analysis Results
Table 3 reports the multivariate analysis results without the interaction between PCMH group status and insurance types, and Table 4 reports results with the interaction.
Multivariate Logit Model for Nonurgent ED Use Without the Interaction Between PCMH Groups and Insurance Types.
Abbreviations: ED, emergency department; PCMH, patient-centered medical home.
Multivariate Logit Model for Nonurgent ED Use With the Interaction Between PCMH Groups and Insurance Types.
Abbreviations: ED, emergency department; PCMH, patient-centered medical home.
In Table 3, compared to the group without a regular source of care, none of the PCMH groups was associated with reduced nonurgent ED use (P = .973 for the full PCMH and P = .445 for the partial PCMH group) for the overall study sample that included both uninsured and insured patients.
In Table 4, compared to the group without a regular source of care, although the partial and unknown PCMH groups were not associated with nonurgent ED use (P = .162 and P = .127, respectively), lower odds of nonurgent ED use were observed for the full medical home group among the uninsured group: odds ratio (OR) = 0.37 (P = .021). The full PCMH group had a marginal effect (at means) of 22.3% lower predicted probability of nonurgent ED use. Compared to uninsured individuals, those with public and private insurance had lower odds of nonurgent ED use among the group without a regular source of care (OR = 0.35, P = .018, and OR = 0.46, P = .034, respectively).
Holding covariates constant, difference in odds of nonurgent ED use between the full PCMH group and the “no regular provider” group was significantly different between the uninsured and insured individuals (β = 1.70, OR = 5.46, P = .009 for public insurance, and β = 1.04, OR = 2.84, P = .040 for private insurance). Specifically, reductions in odds of nonurgent ED use between the full PCMH group and the “no regular provider” group was significantly larger for the uninsured group than the insured groups (β = −1.70, OR = 0.18, P = .009, compared to the public group, and β = −1.04, OR = 0.35, P = .040, compared to the private group).
Similarly, difference in odds of nonurgent ED use between the partial PCMH group and the “no regular provider” group was significantly different between uninsured and publicly insured individuals (β = 1.67, OR = 5.30, P = .019). Specifically, reductions in odds of nonurgent ED use between the partial PCMH group and the “no regular provider” group was significantly larger for the uninsured group than the public group (β = −1.67, OR = 0.19, P = .019).
Discussion
In summary, although the PCMH model may not be associated with reduced nonurgent ED use for the overall population that includes both insured and uninsured patients, lower levels of nonurgent ED use were observed among uninsured patients in the full medical home group compared with the “no regular source of care” group. The PCMH model still holds the promise of reducing nonurgent ED use nationwide, especially among the uninsured population. Compared to people with any insurance coverage, uninsured individuals had higher odds of nonurgent ED use. This is observed among the overall population in Table 3 when uninsured individuals were compared to private enrollees, and in the “no regular provider” group in Table 4 when uninsured individuals were compared to both private and public enrollees. In reality, some primary care doctors and specialists do not accept uninsured patients because they do not receive adequate payment, while ED services are always open and cannot refuse patients regardless of their ability to pay. Findings based on the DID method indicate that compared with any insurance group, the PCMH’s effect on reducing nonurgent ED use is the most pronounced among the uninsured group. This reduction in magnitude of nonurgent ED use may be attenuated among either public or private enrollees.
The insignificant results of the partial PCMH (having a regular source of care alone) may reflect its limited role in reducing nonurgent ED use. Compared to the full PCMH model, the regular source of care only represents a structure measure, and a primary care practice with a structure in place alone may not lend itself to reduced nonurgent ED use. This structure measure does not reflect the health care quality of a regular provider and the capacity of its supporting primary care system, such as care coordination and integration, which is indicated in PCMH features. Thus, individuals in the partial PCMH group with a regular source of care may not necessarily have lower levels of nonurgent ED use than those without.
These findings have important policy implications. They shed light on the effectiveness of innovative health care delivery models in reducing nonurgent ED use. The PCMH model may not be effective when applied to the overall population, but only work well among the target subpopulation. Thus, discretion may be exercised for PCMH practices to achieve health care efficiency with respect to the patient population.
Compared with any insurance group, because PCMH’s effect on reducing nonurgent ED use is the most pronounced among the uninsured group, the uninsured population would be the target population for PCMH implementation. Priority should be given to providing them with access to PCMH practices and encouraging them to use PCMH services to strengthen and amplify PCMH’s effects on reduced nonurgent ED use.
Also, unsurprisingly, this study’s findings indicate an association between lack of insurance and nonurgent ED use, especially among the “no regular source of care” group. Thus, policies such as Medicaid expansion and health insurance exchange programs that expand insurance may help reduce nonurgent ED use.
This article uses patient-reported experience to measure medical home models rather than national guidelines. Currently, patient experience is widely used as a measure of health care quality, 29 and the PCMH model has gained ascendency by highlighting its core features from the patient’s perspective: a whole person orientation by meeting all patient needs, encouraging patient participation, and enhancing patient access to care. The PCMH model seeks to improve the continuum of personalized care from primary care or preventive care to treatment of chronic and acute illnesses. Although health care organization and delivery relate to care effectiveness, patients’ values, beliefs, and circumstances also influence their expectations of, needs for, and use of services. 30 If the patient does not perceive that he or she is receiving PCMH care, then even if a PCMH practice is meeting national standards of PCMH, it may not be effective in reducing nonurgent ED use. Thus, patients’ buy-in to PCMH models likely would affect the extent to which they interact with ED services, regardless of improvements in the organization and delivery of primary care practices. On the other hand, if a practice is not a PCMH per se, but patients could perceive its care qualities as being those expected from a PCMH, then their positive perceptions could serve as a beneficial factor that contributes to their satisfaction with primary care quality and lead to a reduction in nonurgent ED use, even if the practice is not a fully functioning PCMH.
The research question is whether a PCMH is associated with reduced nonurgent ED use. Thus, for an overall population that includes a majority people who do not have any ED use, PCMH may not reduce nonurgent ED use. PCMH may have a potential effect only for those with any ED use, which was supported by results of the following sensitivity analysis.
The authors further examined the research question among 2 overall study populations, including (1) those with nonurgent ED use and no ED use and (2) those with overall ED use (both nonurgent and urgent ED use) and no ED use. The authors found that, basically, ED visits in the general population that includes a majority of people without ED use were not affected by PCMH, either in main effects or in DID estimations.
Limitations
This study has some limitations. First, the measurement of PCMH is based on patient perception. The authors acknowledge that the national dialogue is evolving to one in which the PCMH is not considered a “one size fits all.” Even though each PCMH model has unique features, it does not necessarily mean that common components within unique PCMH models that vary widely across health systems and locations cannot be captured or summarized. MEPS data are still a good source to reflect the main common features of the PCMH model, especially patients’ perspectives. PCMH measures based on these common features can still contribute to estimating this model’s effect on reduced nonurgent ED use nationwide. Second, this study has a short observation window of only a one-year follow-up period, which is constrained by the 2-year panel design in the MEPS data. A longer follow-up period may yield more pronounced effects. Third, the selection issue may not be thoroughly accounted for although the authors have controlled a set of covariates and used the time lag effect to attenuate it. Fourth, another covariate is not available in the data; the length of time of PCMH implementation, which cannot be reflected by the PCMH measure in the one-year baseline period. As is known, PCMH models are a new concept with a history of just a few years since the PCMH joint principles were released in 2007, 31 and most of the large demonstrations to transform practices did not begin until 2009 or later. Additionally, it takes years of difficult transitioning before a practice can be transformed into a PCMH. Thus, it is expected that there are not many practices with a long history of PCMH implementation in the 2010-2011 cohort. Thus, these results, conservatively, have demonstrated consistent patterns of reduced nonurgent ED use in the PCMH model among the uninsured population. That is, even if the current full PCMH group contains large numbers of practices with a short history of PCMH implementation, these results are already significant. Having more primary care practices with a longer history of PCMH implementation in the coming years will further strengthen the significant results of lower levels of nonurgent ED use in the PCMH model. Future research should have better measurement of the PCMH concept, such as features reflecting integrated care and health care quality and safety, and measurement of the duration of PCMH implementation, have a longer follow-up period, and further account for the selection issue.
Conclusions
In conclusion, the empirical evidence suggests that, at a national level, although the PCMH model may not be associated with lower levels of nonurgent ED use for the overall population, it demonstrates higher odds of reduced nonurgent ED use among the uninsured group, among whom are proxies for incorporating patient perspectives. These findings inform policies of and shed light on the effectiveness of innovative health care delivery models in reducing nonurgent ED use. The PCMH model may not be effective when applied among the overall population, but only work well among the target subpopulation to achieve health care efficiency.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.