Reducing Emergency Room Visits and In-Hospitalizations by Implementing Best Practice for Transitional Care Using Innovative Technology and Big Data

Abstract

Background:

Efforts to improve care transitions require coordination across the healthcare continuum and interventions that enhance communication between acute and community settings.

Aims:

To improve post-discharge utilization value using technology to identify high-risk individuals who might benefit from rapid nurse outreach to assess social and behavioral determinants of health with the goal of reducing inpatient and emergency department visits.

Methods:

The project employed a before and after comparison of the intervention site with similar primary care practice sites using population-level Medicaid claims data. The intervention targeted discharged persons with preexisting chronic disease and delivered a care transition alert to a nurse care coordinator for immediate telephonic outreach. The nurse assessed social determinants of health and incorporated problems into the EHR to share across settings. The project evaluated health outcomes and the value of nursing care on existing electronic claims data to compare utilization in the years before and during the intervention using negative binomial regression to account for rare events such as inpatient visits.

Results:

Avoiding readmissions and emergency visits, and increasing timely outpatient visits improved the individual’s experience of care and the work life of healthcare providers, while reducing per capita costs (Quadruple Aim). In the intervention practice, the nurse care coordinator demonstrated the value of nursing care by reducing inpatient (25%) and emergency (35%) visits, and increasing outpatient visits (27%). The estimated value of avoided encounters over the secular Medicaid trend was $664 per adult with chronic disease, generating $71,289 in revenue from additional outpatient visits.

Linking Evidence to Action:

Using health information exchange to deliver appropriate and timely evidence-based clinical decision support in the form of care transition alerts and assessment of social determinants of health, in conjunction with data science methods, demonstrates the value of nursing care and resulted in achieving the Quadruple Aim.

INTRODUCTION

Around the globe, healthcare systems are undergoing a technological revolution aided by unprecedented improvements in computational capacity, storage, and speed, making it possible to procure large datasets that may be used to improve patient outcomes through the identification of novel patterns, associations, or trends from data that is collected during routine patient care. These so-called “Big Data” are characterized by their variety, velocity, veracity, and value (Brennan & Bakken, 2015). Moreover, these technological advances have given birth to an exciting new field of data-driven scientific inquiry known as Data Science, an interdisciplinary approach that uses automated methods to extract actionable knowledge from large datasets that may not have been apparent using traditional methods of inquiry (Provost & Fawcett, 2013; Westra et al., 2015). Nurses have been using data science approaches for nearly a decade to discover knowledge, predict, and evaluate patient outcomes (Westra et al., 2016).

Knowledge discovery, the product of data science inquiries, strengthens the development of evidence-based care models and clinical decision support (CDS) tools. When applied in a practice setting, it becomes possible to streamline clinical processes, while generating the data-driven evidence needed to support and quantify the value of nursing care (Brennan & Bakken, 2015; Pruinelli, Delaney, Garcia, Caspers, & Westra, 2016; Westra et al., 2015). Data science innovations provide opportunities to analyze large datasets using modern methods that aim to improve healthcare quality, safety, and patient outcomes. When harmonized, care delivery methods become mutually beneficial to the patient, health system, and provider; thereby achieving the Quadruple Aim (Bodenheimer & Sinsky, 2014).

The Quadruple Aim expands upon the “Triple Aim” (to improve the health of the nation, experience, and per capita cost of care), which forms the basis of the conceptual framework for the Affordable Care Act of 2010 (Berwick, Nolan, & Whittington, 2008). The ACA brought profound changes to the US healthcare system and information systems became the central component for integrating healthcare delivery and improving population health. However, after observing “widespread burnout and dissatisfaction” (p. 573), Bodenheimer and Sinsky (2014) recommended that a fourth aim, the well-being of the healthcare provider, be added to modify the Triple Aim. Specifically, the Quadruple Aim focuses on the work environment and proposes that care settings: (a) implement team documentation, (b) use pre-visit planning, (c) expand the role of nurses and unlicensed assistive personnel, (d) standardize and synchronize workflows, and (e) colocate the team.

According to Brennan and Bakken (2015), “The process of using big data begins with posing questions and recognizing opportunities” (p. 478). Therefore, relying on existing health information exchanges (HIE) to deliver tailored CDS alerts, combined with the use of administrative database analytics, are innovative technology and data analytic methods that help drive evidence-based protocols to achieve the Quadruple Aim. The purpose of this pilot study was to demonstrate the feasibility of implementing the Coordinating Transitions Intervention (CTI) in a primary care setting to reduce hospitalizations by delivering evidence-based CDS to the right person, in the right place, at the right time using HIE. The study consisted of a pre-post-intervention comparison design that integrated the use of CDS to guide care coordinator outreach during transitions of care, while using population-level Medicaid claims data to evaluate the intervention.

This report illustrates how linking evidence-based practice to innovative analytics on existing electronic databases and health information technology approaches helped nurses reduce unnecessary hospitalizations in a vulnerable Medicaid population by identifying and acting upon social determinants of health. The results of the study underscore how the deliberate use of existing data and information technology resources stimulated the primary care nurses and staff to add measurable value to the practice, thereby successfully achieving the Quadruple Aim. The results of the study suggest that it is possible to expand and replicate the CTI project in other community settings. Moreover, the study outcomes demonstrate that nurses and other healthcare providers may use HIE and data science methods for the development of creative solutions that have a positive impact on patient outcomes.

BACKGROUND

Poorly coordinated care transitions for persons with preexisting chronic conditions often result in unnecessary hospital readmission and emergency department (ED) use in the weeks following discharge, and high readmission rates are markers of poor care quality that result in reimbursement penalties (Centers for Medicare & Medicaid Services, 2012). The evidence demonstrates that interventions that focus on improving coordination between acute care and community settings, called bridging interventions, are effective in reducing readmissions and suggests that incorporating these best practices into care, supported by analytics on existing electronic databases and health information technology, improve patient safety and outcomes (Rennke, Nguyen et al., 2013; Rennke, Shoeb et al., 2013).

If coordinating care transitions requires a systems approach to improve communication and evaluate outcomes across settings, and the necessary technology and data analytics exist, why does the readmissions problem persist? Fragmentation of health care, poorly integrated electronic health records (EHR), and limited access to administrative databases are part of the problem; however, failure to implement a process to improve the adoption of evidence-based practice and use of technology certainly contributes to the problem.

There is a large body of evidence supporting approaches for implementing systematic, best practice protocols in single-site clinical settings such as the innovative Action-Oriented Framework proposed by Jungquist et al. (2014). However, many of these models rely on lean principles derived from the manufacturing industry that emphasize incremental improvements within an organizational culture that is committed to quality improvement (Harrison, 2017). Despite limited evidence of success, these approaches remain the gold standard, while conveying little value to community-based providers who are practicing and communicating between disparate settings, with varying resources, along a continuum of care that is often poorly integrated (Harrison, 2017).

In 2014, The Office of the National Coordinator for Health Information Technology (ONC) prioritized the issue and had begun to develop a 10-year strategic initiative fundamentally changing information exchange capabilities across the healthcare continuum (U.S. Department of Health and Human Services [DHHS], Office of the National Coordinator for Health Information Technology, 2014). Their first major accomplishment was the development of a roadmap designed to coalesce all stakeholders around a shared strategic approach for improving interoperability capabilities across the nation (DHHS, Office of the National Coordinator for Health Information Technology, 2017). However, it will be nearly a decade before the goals of the ONC’s initiative to create a better-connected healthcare system will be fully realized (DHHS, 2017). Therefore, many health professionals will continue to grapple with increasing documentation burdens and redundant “work-arounds” that contribute to gaps in care and increase provider frustration, leading to burnout and added costs of care that negatively affect patient outcomes (Bodenheimer & Sinsky, 2014).

This demonstration project shows the potential for the use of interoperable HIE to reduce IP and ED utilization among high-risk individuals and to improve the patient and clinician experience of care, in accordance with the ONC’s strategic initiative. The Quadruple Aim framework is used here to describe the implementation of the CTI study in a single primary care practice to show how reorienting quality improvement initiatives around a nucleus of technology-enabled, integrated approaches to care worked in a real-world setting. The CTI intervention sought to develop system solutions that bridged settings and facilitated automated communication between the hospital and primary care.

METHODS

The project used a before and after comparison of the effectiveness of the intervention in reducing hospitalization and emergency visits. An urban primary care practice with a roster of 6,000 individuals with a third of cases insured by Medicaid served as the intervention site. Two comparison urban practices with similar disease prevalence and proportion of Medicaid cases provided usual care. In addition, the study compared utilization patterns in the remaining Medicaid population in the eight-county region. Clinical partners exchanged information using established rules for informed consent for sharing protected health information among participating providers, and the research team did not have access to any information exchanged between providers. The University at Buffalo, State University of New York, and Institutional Review Board approved the protocol for retrospective review of the EHR for individuals whose discharge triggered a care transitions alert by a research assistant. The outcome analysis relied on access to de-identified data extracted from the Medicaid Data Warehouse (MDW) and the review board determined that this activity did not constitute human subject research.

In addition to descriptive statistics, the analytic strategy used negative binomial regression to evaluate the significance of avoided events. Negative binomial regression is a type of generalized linear model where the outcome variable is a count (greater than o) of the number of times, an event occurs. This method is useful when attempting to understand rare events such as estimating population-level cancer risk, or in this case, counting avoided hospitalizations. The fidelity of the intervention was assured in two ways. First, a 2-day staff training session was conducted on the technology and clinical intervention aspects of the study. Second, research assistants from nursing and industrial systems engineering visited the site on a weekly and as-needed basis to provide support and ensure that the study protocol was followed.

The intervention consisted of three technology and big data innovations that relied on using existing de-identified electronic data in the Medicaid claims data warehouse for New York State and the interoperable exchange of EHR data across settings:

• automated care transition alerts delivered within 24 hours of discharge;

• social and behavioral determinants of health integrated into care planning; and

• health outcome analysis that demonstrates the value of the nursing intervention.

This report highlights how innovative technology (HIE and CDS) and data analytics can improve implementation of evidence-based transitional care and demonstrates the effectiveness of the CTI in improving post-discharge utilization value, cost reduction, and efficient use of technology.

Technology and Big Data Approach

Care transition alert: Health information exchange (HIE).

The specific aim of the coordinating transitions project was getting the right information (discharge of person with chronic condition) to the right person (nurse in role of care coordinator) in the right place (primary care) at the right time (within 24 hours of discharge) in the right way (secure electronic message). The technology for the interoperable exchange of health information existed at the regional health information organization, HEALTHeLINK. All regional hospitals shared clinical data including admission, discharge, and transfer (ADT) notification whenever individuals moved within the healthcare system. ADT were stored in a clinical data repository and shared with providers who had obtained consent. The intervention primary care practice routinely got consent from individuals on their roster, so it was possible for HEALTHeLINK to “push” clinical results into the primary care EHR. However, without filtering, the volume of ADT was so high that it rapidly overwhelmed a practice EHR, so the intervention needed to reduce notifications to actionable alerts only.

The first step was to limit the ADT to messages that signified a discharge from the acute hospital to the community setting. The practice already sent the list of their consenting adult patients (roster) to HEALTHeLINK. The evidence showed that care transition interventions were most effective on individuals with preexisting chronic conditions; therefore, a cohort table was created with a line for each individual on the roster that identified the common chronic conditions and classified the individual as either chronic or nonchronic based on their conditions. Analysts in the data repository joined information from the ADT, roster, and cohort table to create a care transition alert message that included information about the individual, the hospitalization, and their preexisting chronic conditions. The repository sent the alert as a secure email message to the nurse care coordinator in the intervention practice to signal the need for telephonic outreach within 48 hours. HEALTHeLINK automated the process within 3 months.

The ePCAM as clinical decision support (CDS).

A second technological innovation was the development of an evidence-based CDS tool to incorporate information about social and behavioral determinants of health into the EHR for exchange across healthcare settings. Researchers at the University of Minnesota Department of Family Medicine and Community Health developed the Patient Centered Assessment Method (PCAM; Maxwell, Hibberd, Pratt, Cameron, & Mercer, 2011; Pratt, Hibberd, Cameron, & Maxwell, 2015; Yoshida et al., 2017). The PCAM is a 12-item screening tool with a four-level rating scale of social and behavioral determinants of health. Research evidence supports the inclusion of assessment of social determinants as part of transitional care (Jones et al., 2015; O’Toole, Johnson, Aiello, Kane, & Pape, 2016).

The nurse care coordinator completed the PCAM during the outreach phone call to screen for individuals who might need additional support to prevent rehospitalization. The initial plan was to have the nurse enter the value for each item as discrete data for interoperable exchange. However, having numeric values for the 12 items did not help the nurse identify problems for care planning, so the research team developed a web-based version (called the ePCAM) that identified both problems and strengths based on the assessment. Results were textual statements describing the problem that the nurse copied and pasted into the outreach call documentation. The change reduced the number of steps in the process and facilitated development of a care plan that addressed social determinants of health.

The COMPLEXedex™ clinical algorithm: Big data analytics.

The project relied on analytics on existing electronic databases to identify the population with chronic disease using data extracted from the EHR for risk stratification and from the MDW for analysis of utilization outcomes. At the core was the COMPLEXedex™ clinical algorithm, which uses Clinical Classifications Software (H-CUP, 2017) data definitions to create registries for chronic conditions and classifies individuals into one of the 19 hierarchical disease categories based on their comorbidities. For claims data extracted from the MDW, the research team developed data definitions for encounters of care and the total number of inpatient (IP), emergency department (ED), and outpatient (OP) visits for the year for each individual were calculated. The process for data extraction, risk stratification, and enumerating encounters has been in place since 2012, and data definitions are validated with each data extraction cycle (Hewner, Wu, & Castner, 2016).

The first step in the outcomes analysis was to attribute patients to the intervention or comparison groups. The analyst reviewed the billing provider for OP visits over the last 3 years to assign individuals to a practice, considering both the most recent visits and total visits to a single site. For the baseline (2014) and intervention years (2015), the rate of utilization was calculated for the population classified as having a chronic disease (individuals identified as having chronic kidney disease, heart failure, coronary artery disease, diabetes, chronic obstructive pulmonary disease, mental health, substance abuse [excluding tobacco addiction], or asthma).

Because of the large number of cases with no IP events, it was necessary to model the data by evaluating the count, or number, of hospitalization events. Therefore, the analysis of the change in rate was completed using negative binomial regression. Negative binomial regression is a method that allows for an imbalance between the conditional mean and variance in the data. This imbalance becomes clear when the variance is either larger or smaller than the mean, which would result in bias when measuring the occurrence of rare events in large datasets using other methods such as Poisson regression.

The assumption is that without the intervention, rates would not change from the expected rate. The analyst calculated avoided (or excess) events based on the difference between 2014 and 2015 rates for each site. The analysis used the most recent cost for events based on Healthcare Cost and Utilization Project data to determine the potential for savings and to demonstrate the value of the nursing care coordination. The value of nursing care was determined as the difference in the estimated expense between the practice site and the secular Medicaid trends.

Achieving aim 4 of the Quadruple Aim.

This study used big data and HIE to integrate care while addressing the practical steps to address the fourth aim: Improving the work life of healthcare providers, clinicians, and staff (Bodenheimer & Sinsky, 2014). Table 1 demonstrates the implementation tasks that align with each recommended step.

Table 1.

Comparison of the CTI Intervention Within the Context of the Quadruple Aim

Steps to achieve fourth Quadruple Aim	CTI intervention implementation tasks
1. Implement team documentation	Nursing documented assessment and problems and adjusted the care plan in the EHR.
2. Use pre-visit planning	Secure alerts notified nurse care coordinator to initiate an outreach phone call to high-risk cases for readmission.
3. Expand the role of the nurse	Nurse received alerts, made outreach calls, identified problems, engaged appropriate team members and updated interprofessional care plan. Workshops and weekly feedback to nurse during implementation supported and reinforced benefit of expanded role.
4. Standardize and synchronize workflows	Identification of vulnerable population, linked to discharge notification, enabled care coordinator to engage high-risk patients proactively.
5. Colocate the team	The primary care team was currently working in a single practice setting.

RESULTS

The impact of the intervention on informational and clinical workflows has been reported elsewhere (Hewner et al., 2017). The project goal was to reduce low-value utilization (IP and ED) in the population with preexisting chronic conditions, and to increase OP follow-up, especially after discharge. Table 2 presents the utilization rates per 1,000 adult Medicaid cases with preexisting chronic disease for the intervention practice, two comparison sites, and the regional Medicaid population in 2014 (baseline) and 2015. In general, IP and ED rates decreased between 2014 and 2015 while OP rates increased for the intervention and comparison populations (except IP rates in comparison B), which is not surprising given the national emphasis on avoiding readmissions in 2015.

Table 2.

Utilization Rates for Adult Population With Chronic Conditions in 2014 and 2015 for Intervention and Comparison Groups

Group (population in 2015)	Event type	2014 rate per 1,000	2015 rate per 1,000
Intervention site (n = 419)	IP	338	255
	ED	2,038	1,327
	OP	6,996	8,907
Comparison site A (n = 963)	IP	306	282
	ED	1,608	1,178
	OP	7,741	9,033
Comparison site B (n = 2,085)	IP	279	287
	ED	1,711	1,551
	OP	8,328	9,285
Other regional Medicaid (n = 38,612)	IP	358	297
	ED	1,754	1,434
	OP	7,925	8,253

Note. IP = Inpatient, ED = Emergency department, OP = Outpatient utilization.

Figure 1 shows the growth rate for intervention and comparison sites for the three types of utilization. The decline in IP and ED is greatest in the intervention population in addition to the increase in OP visits among this group. Overall, this pattern represents improved value utilization. Although the regional Medicaid population also shows significant decreases in IP utilization, there is very little improvement in OP rates. Comparison site A showed better value utilization than comparison site B, but not as much improvement as the intervention practice. Notably, the intervention practice was able to meet New York State delivery system reform goals for 2019 during the intervention year (2015).

Figure 1.

Comparison of IP, ED, OP growth rates among practice sites (2014 baseline and 2015).

Note. IP = Inpatient, ED = Emergency department, OP = Outpatient utilization.

Statistical analysis is difficult when trying to understand rare events, such as IP hospitalization in primary care settings. Using negative binomial regression, the study found in the large other Medicaid population (N = 38,612) growth rate changes in IP, ED, and OP rates between 2014 and 2015 were statistically significant with p-values of 4.06 × 10⁻³¹, 2.58 × 10⁻⁶⁶, and 1.31 × 10⁻¹⁰, respectively. Although the study practice had the smallest population (n = 419), its IP rate reduction was closest to statistical significance with p = .09, and the change in ED and OP were statistically significant among all groups (see Table 3).

Table 3.

Study and Comparison Practices of Growth Rate in Inpatient, Emergency, and Outpatient Utilization (2014 Baseline and 2015)

Practice	Change in IP rate (p)	Change in ED rate (p)	Change in OP rate (p)
Study practice (N = 419)	.09+	1.15 × 10−4**	2.01 × 10−4**
Comparison A (N = 963)	.48	8.01 × 10−5**	2.38 × 10−5**
Comparison B (N = 2,085)	.68	.04*	1.19 × 10−5**

Note.

^*indicates p < .05

^**indicates p < .001, and

⁺indicates p < .10.

It is possible to estimate the economic impact of these changes (see Table 4). The difference between the expected utilization rate (the rate at baseline) and the actual rate in 2015 was used to calculate avoided events. Avoided events were computed by multiplying the difference in rates by the population size in 2015. The reduction in the rate of IP and ED encounters resulted in $1,669 less per adult Medicaid recipient with chronic disease in the intervention practice than expected based on 2014 rates. The secular trend can be seen in the avoided events in the regional Medicaid. Thus, the avoided expense related to the intervention is $664, and control practices cost more than the trend. Furthermore, additional OP visits (n = 801), generated $71,289 of new revenue for the intervention practice, which could be used to support the expanded role of the nurse care coordinator.

Table 4.

Avoided events based on the difference between expected IP and ED utilization rates based on 2104, and actual rates in 2015 for the population size in 2015

Population Size in 2015	Avoided IP Events @ $10,855*	Avoided ED Events @ $1,077*	Total expense related to Avoided Events (IP + ED)	Cost avoided per person	Difference from regional Medicaid
Intervention (N = 419)	−35	−298	−$699,117	−$1,669	−$664
Comparison A (N = 963)	−22	−414	−$689,616	−$716	$289
Comparison B (N = 2,085)	18	−333	$165,050	$79	$925
Regional Medicaid (N = 38,612)	−2,341	−12,354	−$38,792,171	−$1,005	$0

Notes: IP = Inpatient, ED = Emergency Department.

^*rates for IP, ED from national average cost data in 2014 from HCUP.net.

DISCUSSION

The expected outcome is better-coordinated transitions between healthcare settings for individuals with chronic health conditions. Front-line staff were engaged in planning and implementing the change and a business case was developed to support the change. However, if the long-term goal is sustaining and expanding the scope of the CTI, then technology and data analytics are critical to creating an environment where high value post-discharge care is the norm, rather than the exception. From its inception, the coordinating transitions project relied on health information technology to improve care for chronically ill individuals and on existing electronic data to evaluate outcomes.

The findings of the CTI project demonstrate that avoiding readmissions and emergency visits and increasing timely visits to the PCP to improve the individual’s experience of care, work life of healthcare providers, clinicians, and staff, while reducing per capita costs (i.e., Quadruple Aim), is possible using data-driven evidence-based practice and technology. The results of the study highlight how using data science and HIE to harmonize person-centered care with evidence-based transitional care practices adds value to the practice and can be used to quantify the value of nursing care (see Table 4). Thus, technology has the potential to support quality improvement processes along the continuum of care by delivering relevant and actionable evidence-based information to the right person, place, time, and way when integrated into existing clinical workflows (Hewner et al., 2017).

Limitations and Implications for Future Research

The CTI intervention was implemented in a single primary care practice and selected comparison practices based on their technology capabilities as well as similarities in disease prevalence and proportion of Medicaid-insured individuals. There were multiple concurrent interventions in the region during implementation that could have affected the results and we used the regional Medicaid population to approximate the secular trends in utilization. Those changes are accounted for by focusing on avoided events that we can attribute to the intervention by only counting events that are greater than the secular trend. Finally, the HIE and CDS interventions were not part of a unified package and required additional steps on the part of the care coordinator. Therefore, our recommendations for future research include bundling the CTI interventions into a seamless package that leads to the development of a comprehensive shared care plan (Sullivan, Mistretta, Casucci, & Hewner, 2017) and deployment regionally as a pragmatic clinical trial.

CONCLUSIONS

High quality, high value transitional care requires that quality improvement initiatives move beyond the single setting where providers are merely empowered to improve healthcare quality, toward settings where they are inspired to develop novel solutions to improve the healthcare system. Integrating traditional quality improvement processes into redesigned, technology-enabled systems that authentically involve all members of the care team (including primary and acute care settings, and community resources to support these efforts), can achieve this goal. According to the practice manager at the CTI intervention site, the change has been dramatic and has transformed care delivery. The practice is using the same resources more effectively and efficiently without increasing the workload burdens of the providers, while reducing unnecessary hospitalizations and emergency visits, by providing supportive person-centered care during care transitions. WVN

LINKING EVIDENCE TO ACTION.

• Nurses play a critical role as front-line providers to pose questions and identify opportunities to use big data to support evidence-based practice and solve healthcare problems.

• Nurses MUST establish methods to demonstrate the tangible and intangible value of nursing care to influence healthcare policy and research.

• Efforts to coordinate care need to transcend single settings and provide a bridge for information relevant to the person’s health such as social factors.

• Moving beyond the Triple Aim to include clinician’s work-life quality requires attention to workflow, meaningful alerts, and reducing workload by targeting the appropriate populations and has management, practice, and education implications.