The associations between query‐based and directed health information exchange with potentially avoidable use of health care services

Abstract

Objective

To quantify the impact of two approaches (directed and query‐based) to health information exchange (HIE) on potentially avoidable use of health care services.

Data Sources/Study Setting

Data on ambulatory care providers’ adoption of HIE were merged with Medicare fee‐for‐service claims from 2008 to 2014. Providers were from 13 counties in New York served by the Rochester Regional Health Information Organization (RHIO).

Study Design

Linear regression models with provider and year fixed effects were used to estimate changes in the probability of utilization outcomes for Medicare beneficiaries attributed to providers adopting directed and/or query‐based HIE compared with beneficiaries attributed to providers who had not adopted HIE.

Data Collection

Providers’ HIE adoption status was determined through Rochester RHIO registration records. RHIO and claims data were linked via National Provider Identifiers.

Principal Findings

Query‐based HIE adoption was associated with a 0.2 percentage point reduction in the probability of an ambulatory care sensitive hospitalization and a 1.1 percentage point decrease in the likelihood of an unplanned readmission. Directed HIE adoption was not associated with any outcome.

Conclusions

The Centers for Medicare & Medicaid Services’ (CMS) EHR certification criteria includes requirements for directed HIE, but not query‐based HIE. Pending further research, certification criteria should place equal weight on facilitating query‐based and directed exchange.

1. INTRODUCTION

Better sharing of patient information underpins many efforts to improve quality, safety, and efficiency in the US health care system. Innovative strategies such as the Hospital Readmissions Reduction Program,1 Accountable Care Organizations,2 bundled payments,3 and Medicaid redesign4 depend on providers’ access to comprehensive and timely information. Health information exchange (HIE) is an intervention designed to meet such information needs by facilitating providers’ access to electronic patient information from multiple sources.5 Previous policy initiatives have identified HIE as a critical driver of health system improvement. For example, the Medicare Access and CHIP Reauthorization Act of 2015 “declares it a national objective to achieve widespread exchange of health information,” and because certified electronic health records (EHRs) must be able to exchange patient information, the Meaningful Use program can also be considered a national policy supportive of HIE.

Despite its importance to state and federal health policy, the evidence base for HIE as an effective intervention to improve quality while reducing utilization and costs has been criticized as insufficient.6, 7, 8 The most consistent evidence in support of HIE as a stand‐alone intervention to positively affect health care delivery comes largely from emergency department settings9, 10, 11, 12, 13 and from studies of specific use cases, such as provider use of repeat and/or appropriate imaging,14, 15, 16 medication reconciliation,17 or prescribing behavior.18, 19 In contrast, studies in ambulatory care settings have been less consistent, ranging from no impact,20 to mixed results,21, 22 to some observed improvements in outcomes for specific patient populations.23, 24 Absence of strong evidence of its effectiveness has been a barrier to broader HIE adoption.25

In addition, the existing evidence base of the impact of HIE does not adequately distinguish between different technological approaches to HIE. The actual health information technologies that enable the sharing of electronic patient information among different organizations can be variable in terms of when information exchange occurs or the types and amount of patient information accessible by HIE.26 Yet, most of the research treats HIE as a monolithic intervention without regard to the technological approach that actually makes exchange possible27 and fails to reflect potential differences or implications of changing technologies.8 As such, comparisons of different studies are challenging, synthesizing the literature difficult, the generalizability of the current research is suspect, and we have limited insights into the effectiveness of different technological approaches to HIE to improve the quality of care.28

This study sought to address the above shortcomings in the evidence base by focusing on the impact of HIE in ambulatory care settings with specific attention to the use of differing technological approaches. Specifically, the objectives were to quantify the impact of two different forms of HIE—query‐based and directed HIE—on three types of potentially avoidable services: ambulatory care sensitive hospitalizations (ACSH), nonemergent emergency department (ED) utilization, and unplanned 30‐day readmissions.

2. BACKGROUND

Fundamentally, HIE is an intervention intended to fill an information deficit about a patient.29 Information deficits arising from patient information that is unavailable or difficult to access are common30, 31 and are often the product of patient transitions across fragmented systems of care.32 Examples of basic patient information that is commonly difficult to access, or missing from the provider's own records, in the outpatient setting include prior encounters, past medical history, laboratory results, and clinical notes.31, 33, 34 Additionally, traditional paper‐based and fax sharing information between providers is not timely and is often incomplete. Both difficulty in accessing information and inefficient methods of information sharing contribute to the challenge of information deficits.35, 36 These information deficits are generally considered to have negative implications for provider decision making37 and increase the risk for poor patient outcomes.38, 39 Seeking additional information in order to eliminate or reduce uncertainty is a common response to information deficits.40, 41

Health information exchange improves provider access to patient information.42, 43 In the United States, health care organizations exchange information with each other through one of two general technological approaches: directed exchange and query‐based exchange.3 Both approaches facilitate providers’ access to patient information from other organizations, thereby meeting end users’ information needs, and both have the potential to support improvements in care delivery and to reduce costs.26 Moreover, both have been encouraged by federal health information technology (HIT) policy,44 though notable differences exist.

Directed exchange, also known as “push” HIE, was designed as a replacement for faxing or mailing patient information.26, 45, 46 In this approach, information from one provider in the form of structured documents, such as test results or clinical care documents (CCDs), is sent to other providers. The terms directed and push reflect that the sharing of patient information is initiated by the sender. The exchange process may be automated so that key events, like the posting of test results, or a hospital admission triggers the “push” of information to recipients’ EHRs, DIRECT Secure Messaging account inboxes, or even secured email. This process of exchange requires little effort from the end user, and the pushed documents should automatically be available through the aforementioned mechanisms at the time a provider sees a patient. In fact, the perception is that directed exchange is an easier process than alternative methods of information sharing due to this automatic delivery of information.47 In general, directed HIE is limited in the amount of information shared, with pushed documents reflecting the information available from a single organization or related to a single episode of care. Directed exchange has been offered by nearly all federally funded state HIE projects.45, 48 While not the only methods of directed exchange, the federally supported DIRECT Secure Messaging project is a prominent example of this approach as is the Indiana Network for Patient Care's DOCS4DOCS© system.49

We hypothesize that directed exchange will be associated with a lower probability of potentially avoidable use of health care services. In the case of readmissions, directed exchange could be an approach to overcome inaccessible and incomplete information when patient care spans more than one provider, since these events are often reflective of problematic transitions between the inpatient and ambulatory settings.50 Survey work suggests that providers perceive themselves to be better informed about patients’ conditions and needs postdischarge, and are more likely to follow up with patients, when directed HIE systems are in place.51 For other types of potentially avoidable utilization, directed HIE may better support the referral process by facilitating information sharing with and receiving information back from specialists, as primary care providers are frequently uninformed about the actual receipt or the content of specialty care to which they refer patients.52 Support for improved care coordination could be particularly relevant for patients with chronic conditions, who rely heavily on specialists, and are at risk for hospitalization if their conditions are inadequately managed. A standard use case for directed exchange is support for care transitions.53 Overall, directed exchange has been associated with reductions in avoidable utilization and higher quality of care in high‐risk ambulatory care24 and long‐term care populations.54

Query‐based exchange aggregates data from multiple organizations into a comprehensive longitudinal patient record that reflects a patient's care from across multiple providers.47 This approach is also referred to as “pull” HIE, because the acquisition of patient information is initiated by the user. Query‐based HIE users have access to a consolidated view of a combination of patients’ demographic information, prior diagnoses, medication history, radiology reports and images, laboratory results, and discharge summaries from multiple providers in one place. In general, the information available from query‐based HIE tends to be deeper, broader, and longitudinal (in comparison with directed exchange).42, 47, 55, 56 Like directed exchange, query‐based approaches can also fulfill information deficits during patient transitions. In addition, the more comprehensive, longitudinal information available through query‐based exchange may be a better tool for supporting ongoing patient relationships, allowing providers to review multiple types of information from multiple types of encounters in order to fill in gaps in the patient's record over time, and for maintaining a more comprehensive view of the patient's overall health status. Use of query‐based HIE can be through a stand‐alone web portal or access can be integrated into the end user's EHR.

Most of the studies that have assessed the impact of HIE have been on query‐based exchange, and the literature is generally supportive of the effectiveness of this approach to reduce utilization.7 Specifically, query‐based exchange has been associated with reductions in readmissions,57 hospitalization,11 and ED visits.18 Query‐based exchange may help avoid utilization through multiple mechanisms. As one example, query‐based exchange may lead to safer care by identifying discrepancies in medications.17 Alternatively, query‐based exchange can support better communication of information during transitions of care by giving ambulatory care providers access to diagnostic information from inpatient settings58 or by facilitating information sharing with specialists.59 In addition, and generally distinct from the directed approach, query‐based exchange's use of aggregate and longitudinal data naturally lends itself to population‐level analytics and identification of high‐risk patients for use by case managers and care coordinators.60, 61 Given the factors listed above, we hypothesize that query‐based exchange will be associated with a lower probability of potentially avoidable use of health care services.

Critically, the technological approaches to HIE are not mutually exclusive. Providers may adopt either or both approaches to meet their information needs, and the Office of the National Coordinator for HIT, the federal entity responsible for coordinating US HIT policy, formally endorses both approaches for sharing information.3 However, no study to date has contrasted the specific effects of directed and query‐based exchange of information or examined the impact of information exchange jointly through both approaches and, a priori, it is not clear that one approach is likely to be more useful than the other. This study sought to provide this lacking evidence by determining the association between query‐based and directed HIE on potentially avoidable use of health care services.

3. METHODS

We modeled the association between HIE adoption and subsequent utilization of inpatient and ED services among Medicare fee‐for‐service (FFS) beneficiaries in a longitudinal panel of providers serving Western New York State.

3.1. Setting and sample

The study is based on a unique dataset combing the health technology adoption by providers, health care organizations, and utilization by Medicare FFS beneficiaries in a 13‐county region served by the Rochester Regional Health Information Organization (RHIO). The Rochester RHIO provides HIE services for primary care, specialty care, emergency, public health, inpatient, and long‐term care sites. Twenty‐one hospitals and more than two‐thirds of the region's physicians participate in the Rochester RHIO, which facilitates exchange for more than 1.4 million patients.62

Established in 2006, the Rochester RHIO offers both query‐based (beginning in spring 2008) and directed (beginning in spring 2009) HIE services to participating providers. Query‐based exchange occurs via a secure, stand‐alone web‐based portal. The portal provides authorized users access to longitudinal individual‐level patient information aggregated and summarized from the EHRs and clinical information systems of the RHIO's participating providers. The data available through the portal include the following: laboratory results, radiology reports and imaging, clinical documents (eg, discharge summaries, histories, operative notes), encounter‐based records, contact information, and CCDs. Query‐based exchange data are only accessible for patients who have consented to information exchange. However, consent rates are higher than 97 percent.62 In terms of directed exchange, all participating ambulatory providers are able to receive clinical documents (eg, discharge summaries) and laboratory and imaging results from area hospitals and laboratories. The way in which clinical documents and results are accessed by providers differs according to their specific EHR vendor. Some providers are able to access the information from within their EHR, but others leverage a RHIO‐supplied DIRECT Secure Messaging web application. Patient consent is not required for information sharing via directed exchange.47

3.2. Data

The Rochester RHIO supplied registration listings of all users of directed and query‐based HIE services. The registration lists included adoption dates (eg, dates of first HIE usage) for all practices, individual providers, and individual staff members for both types of HIE. These data were merged with a 100 percent sample of inpatient and outpatient claims for continuously enrolled Medicare FFS beneficiaries from the 13 counties served by the Rochester RHIO from 2008 to 2014. Beneficiaries were attributed to the provider with the plurality of their outpatient evaluation and management claims in the calendar year.63 In instances of ties, beneficiaries were assigned to providers in the following sequence: to primary care providers (see below), to the provider with the most recent claim, and lastly to a randomly selected provider (who had a claim). Providers’ practices were identified according to common tax identification numbers as reported in Medicare Data on Provider Practice and Specialty (MD‐PPAS). The panel included 9986 providers in 3182 practices over the 7‐year period. Beneficiaries were aggregated into quarter‐level observations and merged with HIE adoption files based on the attributed provider's National Provider Identifier (NPI) to examine ACSHs, nonemergent ED visits, and unplanned 30‐day readmissions.

3.3. Determinants of interest: directed and query‐based health information exchange adoption

The primary independent variables of interest were the provider's directed and/or query‐based HIE adoption status (see Appendix). Each provider's approach to HIE was measured separately at the quarter level as a binary variable. At baseline, none of the providers in the sample had adopted either approach to HIE. We excluded the first quarter of a provider's HIE adoption as a washout period to account for (a) potential productivity loss from adopting new technology,64, 65 (b) the potential learning curve associated with fitting HIE into clinical workflows,66, 67 and (c) the potential for measurement error due to the timing of HIE adoption relative to patient events (eg, events could have occurred before HIE adoption).

3.4. Outcome measures

We examined three distinct types of potentially preventable health care utilization. For each quarter, we determined whether a beneficiary had an ACSH as defined by the Agency for Healthcare Research & Quality's68 prevention quality indicators. ACSHs are generally considered potentially preventable with appropriate primary care.69 We also determined whether the beneficiary had a nonemergent ED visit during the quarter according to a widely used algorithm.70 ACSHs and nonemergent ED visits were observed regardless of whether they occurred at a hospital that participated in the Rochester RHIO or not. Third, the portion of hospital admissions that resulted in an unplanned 30‐day readmission during the quarter was identified using the Center for Medicare & Medicaid Services’ algorithm.71 Readmissions were only measured among beneficiaries with admissions to hospitals participating in the Rochester RHIO during the quarter (as no information from either type of HIE would be available to providers if the beneficiary had been admitted to nonparticipating hospitals).

3.5. Other measures

For each provider, we obtained their gender, age, and specialty from MD‐PPAS. Each provider was placed into one of five categories: primary care (eg, family medicine/practice, internal medicine, geriatrics, or cardiology),63 specialists, hospital‐based, psychiatrists, and all nonphysician providers (eg, nurse practitioners). For each provider, we also determined the cumulative number of quarters that he or she was enrolled in HIE services during the study period. As a proxy for panel size, MD‐PPAS was used to measure the total number of unique beneficiaries for whom the provider billed during the year, which we then divided into quartiles based on the distribution. To approximate practice size, we counted the number of unique NPIs within each tax identification number (categorized as: solo, 2‐5, 6‐10, 11‐25, 26‐50, and ≥50). Additionally, for each practice we counted the number of unique individuals with directed or query‐based HIE accounts during each quarter. This number of additional HIE users within the practice accounts for the possibility that other staff members were accessing the HIE system on behalf of physicians (ie, proxy usage).72 In order to ensure that these measures were simply not a proxy for practice size, we divided each count of additional HIE users in the quarter by the total number of NPIs associated with each tax identification number. At the beneficiary level, we abstracted demographics, dual eligibility status, Elixhauser comorbidity scores for each year, and diagnoses of high prevalence chronic conditions.

3.6. Analysis

In unadjusted comparisons, we described the study sample of providers and attributed beneficiaries by demographic characteristics stratified by the type of HIE participation. The unit of analysis was the beneficiary‐quarter. Linear regression models with provider and year fixed effects were used to estimate changes in the probability of potentially avoidable utilization for a Medicare beneficiary attributed to a provider using HIE compared with beneficiaries attributed to providers who were not using HIE. The interaction term of the query‐based HIE indicator and the directed exchange indicator identified providers that had adopted both forms of HIE. Provider fixed effects controlled for any unobserved differences among providers that did not change over time and the year fixed effects controlled for temporal trends by comparing beneficiaries within the same time periods. All models included robust standard errors clustered at the practice level. Adjusted estimates were obtained from the best‐fitting models identified through a backward elimination approach based on improvements in information criterion measures.73 To facilitate interpretation, we expressed the regression coefficients of our main HIE variables as marginal effects and estimated marginal means for each outcome by the type of HIE approach used by the provider.74

The above analyses test our hypotheses that directed and query‐based exchange will be associated with a lower probability of potentially avoidable use of health care services. However, we also directly compared these approaches in two secondary analyses structured to facilitate such an explicit comparison. First, for each of our adjusted models we conducted post hoc tests for differences between the estimated coefficients for query‐based and directed exchange, that is, among providers who adopted any HIE approach, is there a difference in terms of the effect on avoidable utilization? Second, we restricted our sample to providers who had adopted directed or query HIE, but not both types, and then replicated our models with a binary indicator of the type of HIE adopted. Because this was a sample limited to providers post‐HIE adoption, the coefficient of interest reflects the effect of an already user of directed exchange adopting query‐based exchange, the effect of an already query‐based exchange user adopting directed exchange, or the effect in instances of switching between the two approaches. Full results from this secondary analysis are available in the Appendix.

4. RESULTS

Primary care (27 percent) and nonphysicians (27 percent) accounted for more than half of the study panel's providers (Table 1). One out of four providers were solo practitioners (22 percent), and an additional 22 percent were in small groups of 2‐5 providers. The mean number of beneficiaries attributed to each provider was 280.2 (SD = 419.8). In the study panel, the mean percentage of beneficiaries who had an ACSH during a given quarter was 1.6 percent, 5.3 percent had a nonemergent ED visit, and 10.2 percent of hospitalized patients had an unplanned readmission within 30 days of discharge (5.6 percent had at least one admission to a Rochester RHIO participating hospital).

Table 1.

Characteristics of providers and attributed beneficiariesa by health information exchange (HIE) adoption status

	Total (%)	Never adopted HIE (%)	Adopted directed HIEb (%)	Adopted query‐based HIEb (%)	Adopted both types of HIEb (%)	P
Providers	n = 9986	n = 8693	n = 715	n = 427	n = 151
Female gender	38.0	37.0	46.0	44.7	43.7	<0.001
Age (mean, SD)	47.6 (11.5)	48.0 (11.6)	45.7 (10.3)	43.3 (10.5)	44.3 (9.9)	<0.001
Specialty
Primary care	27.4	27.2	32.4	19.9	37.1	<0.001
Medical/Surgical specialtyc	45.6	45.8	39.0	54.8	36.5
Nonphysician	26.9	26.9	28.5	25.3	26.5
Practice size
Solo	21.8	23.8	9.7	5.6	10.6	<0.001
2‐5	22.4	22.3	26.6	11.7	35.1
6‐10	13.6	13.2	21.5	9.4	13.9
11‐25	13.7	13.3	18.5	15.0	13.9
26‐50	8.3	8.8	3.1	8.9	4.0
≥50	20.1	18.6	20.7	49.4	22.5
Panel size quartile
Q1 (≤ 133)	28.2	27.0	27.9	32.4	40.2	<0.001
Q2 (134‐238)	24.9	24.5	23.8	25.8	34.6
Q3 (239‐446)	24.9	24.2	29.1	24.2	12.9
Q4 (≥ 446)	22.1	24.3	19.2	17.6	12.3
Beneficiaries	n = 167 983	n = 108 685	n = 40 608	n = 11 005	n = 7685
Female gender	55.4	54.7	57.2	54.8	57.3
Age (mean, SD)	67.2 (15.6)	67.0 (15.7)	68.2 (15.1)	65.8 (16.1)	66.8 (15.7)	<0.001
Race/ethnicity
White non‐Hispanic	87.6	87.3	91.5	78.9	84.7	<0.001
African American	8.6	8.9	5.4	15.2	11.3
Hispanic	1.2	1.2	0.6	2.6	1.3
Other	2.7	2.6	2.5	3.4	2.7
Medicaid dual eligible	32.7	33.4	29.4	38.5	33.2	<0.001
Elixhauser score (mean, SD)	3.4 (2.8)	3.3 (2.8)	3.4 (2.7)	3.5 (2.8)	3.4 (2.8)	<0.001
Condition (select)
Congestive heart failure	13.0	12.7	13.6	13.8	12.9	<0.001
Hypertension	65.0	63.3	68.6	66.0	68.0	<0.001
Chronic obstructive pulmonary disease	23.5	23.5	23.3	25.1	22.9	<0.001
Diabetes	29.6	29.0	30.3	30.9	31.7	<0.001

^aAt first observation.

^bEventually adopted during the study period per Rochester RHIO.

^cHospital‐based and psychiatry reported with Medical/Surgical Specialty to conform to CMS reporting requirements.

Over the course of the study period, 12.9 percent of providers adopted HIE. Of those adopting HIE, 55 percent adopted directed HIE, 33 percent adopted query‐based HIE, and 12 percent adopted both. The characteristics of providers and attributed beneficiaries differed significantly by the type of HIE adoption (Table 1). Compared with nonadopters, providers adopting either HIE approach tended to be younger, to be employed by larger practices, to have smaller panels of Medicare FFS beneficiaries, and cared for beneficiaries with higher Elixhauser comorbidity scores. In addition, the providers adopting query‐based exchange tended to have a more racially/ethnically diverse set of attributed beneficiaries and a higher percentage of dual eligible beneficiaries.

4.1. Association between health information exchange adoption and admissions for ambulatory care sensitive conditions

In analyses adjusted only with provider and time fixed effects, adoption of query‐based HIE (ß = −0.0023; P = 0.025) was associated with a lower probability of a beneficiary experiencing an ACSH (Table 2 Column A). Directed HIE adoption was not associated with changes in ACSHs. Increasing beneficiary age, female gender, dual eligibility, higher comorbidity scores, and presence of chronic conditions were all associated with an increased likelihood of experiencing an ACSH. Beneficiaries were also more likely to experience an ACSH if they were attributed to providers in larger practices. After controlling for patient gender, dual eligibility status, comorbidity score, and diagnoses as well as the size of the practice, query‐based HIE adoption was negatively associated with the likelihood of an ACSH (ß = −0.0025; P = 0.037). In the adjusted model, directed HIE was not associated with ACSH. The marginal effect of query‐based HIE adoption was a 0.22 percentage point (P = 0.037) reduction in the probability of a beneficiary experiencing an ACSH. Based on the marginal means, this represented an estimated decrease from a 1.6 percent to a 1.3 percent likelihood of an ACSH (or a 15 percent relative decrease) in a given quarter (Figure 1). Adoption of both directed and query‐based exchange was not associated with a lower probability of an ACSH.

Table 2.

Association between health information exchange (HIE) adoption and potentially preventable utilization outcomes

	Ambulatory care sensitive hospitalization (A) n = 2 611 530		Nonemergent emergency department visit (B) n = 2 611 530		Unplanned 30‐d readmission rate (C) n = 149 513
	Bivariate ß	Adjusted ß	Bivariate ß	Adjusted ß	Bivariate ß	Adjusted ß
Health information exchange adoption
None	Reference	Reference	Reference	Reference	Reference	Reference
Directed	−0.0007	−0.0003	−0.0001	0.0008	−0.0022	−0.0012
Query‐based	−0.0023*	−0.0025*	0.0018	0.0009	−0.0122*	−0.0138*
Directed * Query‐based	0.0020	0.0021	−0.0029	−0.0019	0.0090	0.0123
Provider characteristics
Practice size
Solo	Reference	Reference	Reference		Reference
2‐5	0.0010	0.0001	0.0036**		0.0071
6‐10	0.0019*	0.0006	0.0030		0.0083
11‐25	0.0012	0.0003	−0.0011		0.0105
26‐50	0.0035**	0.0028*	0.0035		0.0090
>50	0.0022*	0.0015	0.0044*		0.0238**
Panel size quartile
Q1 (≤ 133)	Reference		Reference	Reference	Reference
Q2 (134‐238)	0.0002		0.0043***	0.0034***	−0.0049
Q3 (239‐446)	0.0008		0.0052**	0.0044**	−0.0001
Q4 (≥ 446)	0.0017		0.0052*	0.0039*	0.0060
Additional HIE users in practice
Directed users	−0.0001		−0.0001		0.0003*
Query‐based users	−0.0001***		0.0001		−0.0001
Time since first HIE adoption	−0.0001		−0.0001		−0.0001
Beneficiary‐level characteristicsa
Age	0.0003***		−0.0009***	−0.0011***	−0.0005***	−0.0008***
Female gender	0.0016***	−0.0004	0.0108***	0.0096***	−0.0044**	−0.0048**
White non‐Hispanic	−0.0017*		−0.0218***	−0.0114***	−0.0139***
Medicaid dual eligible	0.0076***	0.0007*	0.0427***	0.0174***	0.0304***	0.0055***
Elixhauser score	0.0098***	0.0089***	0.0137***	0.0162***	0.0193***	0.0216***
Conditions
Congestive heart failure	0.0573***	0.0215***	0.0460***	−0.0109***	0.0725***	0.0069***
Hypertension	0.0143***	−0.0097***	0.0162***	−0.0076***	0.0266***	−0.0272***
Chronic obstructive pulmonary disease	0.0340***	0.0092***	0.0470***	0.0083***	0.0494***	−0.0018
Diabetes	0.0149***	−0.0070***	0.0173***	−0.0181***	0.0319***	−0.0259***

***P < 0.001; **P < 0.01; *P < 0.05 Year dummies omitted for readability.

Figure 1.

Marginal means of outcomes by type of health information exchange (HIE) approach. *P < 0.05

In the first of our secondary analyses, the parameter estimates from our adjusted models for query‐based and directed HIE were statistically different (P = 0.033), suggesting that query‐based HIE has a greater effect. When the adjusted model was replicated in the sample conditional on prior HIE adoption, the magnitude of the estimate associated with query‐based exchange adoption increased, but it was not statistically significant, indicating that its impact on the likelihood of a patient experiencing an ACSH was not different from directed exchange (ß = 0.0038; P = 0.069).

4.2. Association between health information exchange adoption and nonemergent emergency department visits

Neither adoption of query‐based or directed exchange had a statistically significant association with the likelihood of experiencing a nonemergent ED visit (Table 2 Column B) after adjusting for physician and time fixed effects, nor after additional covariates were included in the regression model. In post hoc comparisons, neither parameter estimate was statistically significant (P = 0.969). Higher Elixhauser scores, dual eligibility, and the presence of chronic conditions were associated with an increased probability of a beneficiary experiencing a nonemergent ED visit. In the sample conditional on HIE adoption, again the magnitude of the estimate associated with query‐based exchange adoption increased, but it was not statistically significant (ß = 0.0054; P = 0.191).

4.3. Association between health information exchange adoption and unplanned 30‐day readmissions

Adoption of query‐based HIE was associated with lower rates (ß = −0.0138; P = 0.012) of unplanned 30‐day readmission (Table 2 Column C). After controlling for potential confounders, adoption of query‐based HIE was associated with a 1.2 percentage point decrease in the likelihood of readmission (marginal effect = −0.0123; P = 0.009) for a hospitalized beneficiary. Based on marginal means, this was equivalent to a decrease from an unplanned readmission rate of 10.3 percent without any HIE to a rate of 8.9 percent with query‐based HIE adoption (Figure 1). Directed HIE adoption had a negative, but not statistically significant association with the likelihood of readmission, which was equivalent to a decrease from an unplanned readmission rate of 10.3 percent without any HIE to a rate of 10.1 percent with HIE. The parameter estimates for directed and query‐based HIE were statistically different (P = 0.035). Again, adoption of both directed and query‐based HIE (as indicated by the interaction term) was not associated with change in the outcome. In the conditional sample, query‐based exchange did not have a not statistically different effect on readmissions compared with directed exchange (ß = −0.0011; P = 0.941).

5. DISCUSSION

Provider adoption of query‐based HIE was associated with small but statistically significant reductions in both ACSHs and unplanned 30‐day readmissions among Medicare FFS beneficiaries, but adoption of directed HIE was not associated with reductions. Prior studies of the association between query‐based HIE and readmissions and/or ACSHs have been largely cross‐sectional and with mixed results.57, 75, 76, 77 This study presents stronger evidence of ambulatory care providers’ adoption of query‐based HIE reducing avoidable utilization. Contrary to expectations based on prior research,23, 54 directed HIE adoption was not associated with reductions in potentially avoidable utilization. This study was the first to contrast the impact of query‐based and directed HIE in the ambulatory care setting on potentially avoidable utilization, contributes to the growing and more rigorous evidence base of HIE, and highlights the current limitations of US HIE policy.

The moderate negative association between query‐based exchange adoption and ACSHs and unplanned readmissions is consistent with positive benefits from HIE reported in the literature.27 This negative association may be a product of query‐based exchange's broad and longitudinal information sources, which might be more supportive of ongoing patient care and resource utilization decisions through more precise querying of available information. (However, directed exchange examples that draw on information from multiple organizations, like Indiana's DOCS4DOCS© system, do exist.) To explore the role of the breadth and depth of the information available as a potential explanation would require patient‐level utilization data linked with detailed end‐user behaviors; this information was not available. Alternatively, a provider's adoption of query‐based HIE usage may reflect a general tendency toward more purposeful information seeking behavior in response to specific patient issues or problems. Because we measured adoption and not usage in this study, and measured effects within providers over time, our findings should not be subject to this type of provider selection.

The absence of a relationship between directed exchange and readmissions in this study was unexpected. Prior studies based on strong research designs have identified specific directed exchange tools as a beneficial intervention to reduce readmissions in select populations.23, 54 This discrepancy with previous research may be attributable to our study's focus on unplanned readmissions, ACSHs, and nonemergent ED visits. Directed exchange may “inundate” providers with information, making it difficult to find the relevant details. Therefore, it may be less effective at avoiding preventable utilization.72 In contrast, query‐based exchange may allow the provider to more efficiently search for specific information relevant to care. Alternatively, the fact be attributable to the prior studies’ much higher rates of utilization than in our sample population. Also, our study differs substantially from the prior literature in three key ways: Prior studies (a) considered only one directed exchange technology (we considered provider adoption of all available directed exchange approaches as our focus was push vs pull); (b) focused on a dense urban area (we had a much larger and varied geographic region); or (c) differed in terms of setting (long‐term care compared with our use of primary care providers).

Contrary to expectations, the adoption of both approaches to HIE did not result in any additive effects. This was evidenced both in our full sample analysis (in the significant interaction terms) and in our secondary analysis conditional on prior HIE adoption. One potential explanation may be that directed exchange and query‐based exchange share a common underlying causal mechanism: Those providers who have more complete information about patients, including information from outside their own provider organization, can avoid unnecessary utilization and provide better care for their patients.29 Because either approach can fulfill this role, there may be little incremental gains by adopting the other, in this specific context and with the three outcomes we examined. For example, in the ED setting where care is unscheduled, larger gains in information accessibility may be possible through query‐based exchange.78 Similarly, directed exchange may be better positioned to support other outcomes. However, the lack of a combined effect could also suggest that our findings may be spurious and driven by unmeasured confounders not captured in our time and provider fixed effects. Notably, the decision to actually adopt any form of HIE is always driven by provider and practice choices.

The differential associations for query‐based and directed exchange identified in this study conflict with current policies. Overall, federal HIT policy has strongly and repeatedly encouraged the movement of patient information between providers using HIE. Additionally, federal strategic plans and reports cite both query‐based and directed HIE as important approaches,3, 79 often as equivalent approaches,80 to exchanging patient information. Notwithstanding, the equivalent support for each has been less consistent in practice. Directed exchange has enjoyed more robust policy support (even though only query‐based exchange was found to beneficially impact quality measures in this study). Most notably, the $34 billion Meaningful Use program only requires that EHRs have directed exchange capabilities in order to be certified and, therefore, eligible for incentive payments. No requirement exists for EHRs to make the web portals that enable query‐based exchange easily or efficiently accessible. Independently, many organizations offering query‐based exchange have managed better integration with EHRs, but that was accomplished in the absence of a federal requirement. Concurrent with the Meaningful Use program, the Office of the National Coordinator initiated the Direct Project to develop standards and services to support directed exchange in the form of DIRECT Secure Messages, but no such approach was initiated for query‐based exchange. Also, the federally funded state HIEs prioritized the implementation of directed exchange services even though query‐based HIE was viewed as the preferable approach.81 At the same time, there has been a trend of less government support for RHIOs, which have predominately offered query‐based HIE.82 The findings of this study indicate that, at a minimum, any certification criteria for EHRs should place equal weight on facilitating query‐based HIE and directed HIE. This would help ensure that richer longitudinal patient information available through query‐based HIE is accessible when needed which may increase the currently low HIE adoption rates in ambulatory care settings.83 Moreover, increased support for query‐based exchange at the individual end‐user level would be in alignment with the Office of the National Coordinator for HIT's Trusted Exchange Framework and Common Agreement, which sets forth principles and guidelines for queries between health information organizations.84

5.1. Limitations

First, the generalizability of these findings may be limited given that the study includes only one health information organization provider. Limited generalizability is a longstanding challenge in HIE research due to the selection issues associated with the significant organizational adoption costs, differences in technology, and varying levels of effort maturity and participation.8, 27 For example, specific technical features in this study may be very different than in other settings. In the case of directed HIE, some approaches include receiving structured documents, but other settings may be sharing nonmachine readable files (like images or PDFs). Additionally, New York has invested more than any other state in HIT infrastructure.85 Moreover, our ambulatory care focused study may not be generalizable to other settings of care. However, our findings fill a gap in the growing evidence base as the bulk of studies supporting the effectiveness of HIE as an intervention comes from ED settings.27, 86 Regardless of generalizability limitations, this study provides the first evidence comparing the effect of directed and query‐based HIE. Second, potentially avoidable utilization among Medicare FFS beneficiaries was measured with three outcomes, but other forms of avoidable utilization may have been reduced and other use cases may be more relevant to the different approaches to exchange. Third, we were only able to measure HIE adoption and not actual usage. It is possible that individual providers who adopted HIE never actually applied the information from either directed or query‐based HIE. However, studies of technology impact with usage as the determinant of interest have found statistically significant results more often than studies measuring only adoption.87 Adoption, therefore, is a much more conservative measure of HIE since measurement error introduced by this approach would bias estimates toward no effect. Fourth, we were limited to provider and practice characteristics available from secondary sources and did not have information on clinic workflows or the level of integration between HIE and EHRs, which are known to affect provider adoption and perceptions of HIE value.51, 66 Our data sources precluded us from linking specific types of patient information accessed by providers (eg, laboratory, imaging, and discharge summaries) to specific patient encounters. Quantifying the anticipated value or impact of specific types of data would provide a more nuanced view of how HIE impacts utilization.88

While this study provides evidence that ambulatory care providers’ adoption of query‐based HIE may reduce avoidable utilization, these findings raise additional questions for research. Critically, we postulated mechanisms by which HIE could change utilization, which have been explicitly argued for, or are suggested from empirical findings in the literature. We did not test any specific mechanism, which is a clear avenue for additional work. Likewise, future investigations could increase specificity around the actual interventions to provide better insights. For example, identifying the key data elements that influence provider decision making or further stratifying the various types (and formats) of clinical documents received through directed HIE would help explicate the causal mechanisms.

6. CONCLUSION

Ambulatory care providers’ adoption of query‐based HIE was associated with a lower likelihood of experiencing an ACSH and lower rates of unplanned readmissions for Medicare FFS beneficiaries. At a minimum, any certification criteria for EHRs should place equal weight on facilitating query‐based HIE and directed HIE.

Supporting information

 

APPENDIX 1. Secondary analyses conditional on health information exchange adoption.

1.1.

We restricted our sample to providers who had adopted directed or query HIE, but not both types, and then replicated our models with a binary indicator of the type of HIE adopted. Because this was a sample limited to providers post‐HIE adoption, the coefficient of interest (Table A1) reflects the effect of an already user of directed exchange adopting query‐based exchange, the effect of an already query‐based exchange user adopting directed exchange, or the effect in instances of switching between the two approaches. The results follow the same modeling strategy as the full analysis. Other variables omitted for readability.

Table A1.

Association between HIE and outcomes conditional on HIE adoption

	Ambulatory care sensitive hospitalization (A) n = 395 909	Nonemergent emergency department visit (B) n = 395 909	Unplanned 30‐d readmission rate (C) n = 22 708
	Adjusted ß	Adjusted ß	Adjusted ß
HIE adoption	0.0038069 (P = 0.069)	0.0054199 (P = 0.191)	−0.001082 (P = 0.941)

Figure A1.

Provider's health information exchange adoption over time

Vest JR, Unruh MA, Shapiro JS, Casalino LP. The associations between query‐based and directed health information exchange with potentially avoidable use of health care services. Health Serv Res. 2019;54:981‐993. 10.1111/1475-6773.13169