Improving Medication Adherence in a Regional Healthcare Information Exchange using a Scalable, Claims-Driven, and Service-Oriented Approach

Abstract

Evidence-based pharmacotherapy is a central aspect of optimal patient care for many chronic conditions. However, medication non-adherence frequently inhibits the attainment of optimal pharmacotherapy regimens. In this study, we designed, developed, and implemented a multifaceted clinical decision support (CDS) intervention that supports evidence-based pharmacotherapy and enhanced medication adherence through the use of a scalable, claims-driven, and service-oriented approach. The intervention includes a medication management report and a low adherence alert based on thirteen evidence-based pharmacotherapy rules for seven chronic conditions. Reports and alerts are delivered to primary care clinics and care managers that participate in a healthcare information exchange in North Carolina. The resulting system architecture may enable this CDS intervention to be widely disseminated to healthcare networks through an open-source model.

Introduction

The IOM Quality Chasm report identified 20 priority areas for which enhancement in care delivery would result in substantial overall improvement in the quality of healthcare.¹ For many of these priority areas (e.g., diabetes, heart failure, hypertension), the use of evidence-based pharmacotherapy is a central component of overall patient care. However, low medication adherence frequently inhibits the attainment of optimal pharmacotherapy.²

Poor adherence to medications is a significant cause of morbidity and mortality, and has been estimated to cost $100 billion a year in the United States.² Lack of adherence to evidence-based pharmacotherapy is the result of two broad factors: (i) clinicians not following evidence-based recommendations; and (ii) patients not taking their prescribed medications as instructed.³

CDS interventions are a promising strategy for dealing with medication management issues, including low adherence.⁴ However, existing CDS-based medication management interventions have important limitations, such as the focus on a single condition and implementation approaches that cannot be widely disseminated across healthcare networks.

In this study, we aimed to address these limitations by designing, developing, and implementing a CDS system that (i) supports multiple evidence-based pharmacotherapies for several IOM priority conditions; and (ii) can be disseminated to various healthcare systems. A randomized controlled trial to assess the impact of the intervention on medication adherence is currently underway, involving 2,652 Medicaid beneficiaries who are enrolled in a care management program and are assigned to one of 15 primary care clinics located in a six-county region in North Carolina.

System Description and Development Methods

Intervention: We developed a Web service that delivers two types of CDS interventions: (i) a point of care medication management report (Figure 1); and (ii) a care manager alert for low medication adherence. The format and content of both interventions were informed by provider and care manager focus groups and iteratively refined by the study research team.

Figure 1.

A medication management report for a patient who has three IOM priority conditions and is eligible to receive anti-hypertensive, anti-diabetic, and lipid-lowering pharmacotherapy for these conditions.

The medication management report consists of the following sections: (i) patient demographics and appointment data; (ii) list of the patient’s IOM priority conditions; (iii) 12-month adherence to target evidence-based pharmacotherapy in numeric and graphical format; (iv) other non-target dispensed medications; and (v) evidence-based suggestions to consider prescribing recommended drugs in the absence of dispensation events for these drugs.

Reports are generated by the Web service and faxed or e-mailed to sites one business day prior to the patient’s appointment. Reports are then placed into the patient’s chart for clinician review during the encounter so low adherence issues can be discussed with the patient and evidence-based suggestions can be considered.

Low adherence alerts are generated when a patient meets the following criteria: (i) average adherence rate lower than 50%; (ii) no outpatient encounters in the last six months; and (iii) no scheduled primary care visits. The alerts indicate that the patient meets the above criteria and recommend facilitating a visit with the patient’s primary care provider.

Evidence-based pharmacotherapy rules: Candidate evidence-based pharmacotherapies for IOM priority conditions were identified and reviewed for potential inclusion in this study based on clinical practice guidelines and primary literature. Each rule was reviewed by study investigators and experts within the relevant therapeutic area to confirm accuracy and relevance in clinical practice. Following this process, we selected and developed 13 pharmacotherapy rules for seven IOM priority conditions: persistent asthma, diabetes mellitus, heart failure, hypertension, ischemic heart disease, stroke, and hyperlipidemia (Table 1).

Table 1.

Summary of the evidence-based pharmacotherapy rules.

Pharmacotherapy	Eligibility logic
ACEI	age >= 18 AND Coronary artery disease AND (Diabetes mellitus OR Chronic kidney disease)
	age >= 18 AND Hypertensive disorder AND Coronary artery disease
	age >= 18 AND systolic heart failure
Anti-diabetic	Diabetes mellitus
Anti-hypertensive	age >= 18 AND Hypertensive disorder
ARB	age >= 18 AND Hypertensive disorder AND Coronary artery disease
	age >= 18 AND systolic heart failure
Beta-blocker	age >= 18 AND (Heart failure OR Coronary artery disease)
Hydralazine-nitrate	age >= 18 AND systolic heart failure AND African-American
Lipid-lowering drug	age >= 18 AND Coronary artery disease AND Hyperlipidemia AND NOT (Coronary artery disease OR Cerebrovascular disease OR Diabetes mellitus)
	age >= 18 AND <= 39 AND Diabetes type II AND Hyperlipidemia AND NOT Cerebrovascular disease
	age <= 17 AND Diabetes mellitus type I AND Hyperlipidemia
Statin	age >= 40 AND (Diabetes mellitus OR Coronary artery disease)
Inhaled corticosteroid	age >= 3 AND Persistent asthma
Montelukast	age >= 3 AND Persistent asthma
Short-acting beta-agonist	age >= 2 AND Persistent asthma
Theophylline	age >= 12 AND Persistent asthma
Warfarin	age >= 18 AND age <= 39 AND (Cerebrovascular disease) AND (Atrial fibrillation OR Valvular heart disease OR Prosthetic mechanical valve)

To confirm the significance and implementability of the 13 selected rules, we performed a preliminary 12-month adherence analysis in a population of 4,173 patients with one or more IOM priority conditions. On average, patients in this population were only 44% adherent to applicable pharmacotherapy rules.

The CDS logic consists of four reusable processes: (i) detection of pharmacotherapy eligibility; (ii) retrieval of relevant medication dispensation events; (iii) adherence calculation; and (iv) production of CDS output. Pharmacotherapy eligibility is determined by applying the criteria described in Table 1 and considering potential contraindications. Patient conditions are inferred from billing record encounter diagnoses according to the Healthcare Effectiveness Data and Information Set (HEDIS) diagnostic criteria.⁵ The output of this process is a time interval during which the patient fully met the criteria to receive a particular pharmacotherapy. Since our intervention is specifically focused on chronic conditions, a condition is assumed to be present from the condition’s first occurrence in a billing record to the rule execution date.

If patient eligibility for a given pharmacotherapy is confirmed, a 12-month medication dispensation history is retrieved from pharmacy claims. These claims contain attributes such as dispensation date, medication supply days, and National Drug Code (NDC). For each dispensation event, the following steps are executed: (i) the NDC code is converted to a generic drug code using First Data Bank’s (FDB) Drug Information Framework;⁶ (ii) the generic drug membership in a pharmacotherapy class of interest (e.g., statin) is verified using FDB’s drug classification system; and (iii) a drug coverage time interval is calculated for each pertinent dispensation event. A patient is considered to be covered by a drug for the length of time indicated by the “days of supply” field in the pharmacy claims data. Since the proposed intervention focuses on medications indicated to manage chronic conditions, the rules assume that medications have been prescribed for use on a regular basis and not “as needed.” If the patient is still covered by a drug when a second dispensation event occurs for a drug with the same ingredients, dose, and dose form, the overlap in coverage is added to the time interval of the second dispensation event. The final output of this process is a set of non-overlapping time intervals during which the patient had supplies for a particular recommended pharmacotherapy.

Finally, the medication adherence rate is calculated as the proportion of days covered by medication within a 12-month period in relation to the number of eligible days. In calculating this adherence rate, the number of days in the eligibility time interval is used as the denominator and the total number of days in the medication coverage time intervals is used as the numerator. This method is consistent with established claims-based adherence calculation methods, such as the Medication Refill Adherence (MRA).⁷ A description of the development and validation of our proposed method is provided elsewhere.⁸

If the patient is considered eligible to receive a particular pharmacotherapy, content for the medication management report is generated as rule output. Finally, a master rule gathers the output of each pharmacotherapy rule and aggregates these outputs to produce the CDS Web service response. Rules were thoroughly tested

High-level system architecture and flow: The system architecture leverages infrastructure components that were developed within the Duke Division of Clinical Informatics as part of previous research projects. These components include: (i) a regional healthcare information exchange (HIE) known as COACH;⁹ (ii) multiple data interfaces, including encounter diagnoses and procedures, pharmacy claims, and clinic schedules; (iii) a CDS Web service known as SEBASTIAN;¹⁰ (iv) a population health management system (PHMS);¹¹ and (v) a document generation service.

COACH serves a community-wide network of academic, private, and public healthcare facilities that provide care for Medicaid beneficiaries. The system facilitates communication between multidisciplinary team members who collaborate in the care of patients. Multiple sources of data are connected to COACH through data interfaces, including the North Carolina State Medicaid Office and healthcare facilities that constitute the community network. In this project, COACH provides the data required for the execution of pharmacotherapy rules.

SEBASTIAN is a software component that encapsulates core functional capabilities required for CDS into reusable, system-agnostic Web services.¹⁰ In SEBASTIAN, patient data (inputs) are provided through Web service requests over the Internet and CDS recommendations (outputs) are sent back to client applications. Both inputs and outputs are encoded in extensible markup language (XML). SEBASTIAN currently serves the needs of multiple CDS applications in various institutions.^9–12 In addition, this architecture served as the basis of the HL7 International and OMG Decision Support Service (DSS) standard.¹³ The pharmacotherapy rules were represented in SEBASTIAN’s knowledge base. Rules execute in a decision engine that can be accessed through a Web services layer.

The PHMS controls the overall CDS execution and delivery workflow as an asynchronous process. First, at a predefined schedule, the PHMS defines the set of patients that need to be evaluated by the pharmacotherapy rules. For the medication management report, this set is restricted to study patients who have an upcoming appointment within the next three days. Second, for each patient, Web service requests are submitted to SEBASTIAN including as input all required data for the pharmacotherapy rules execution. Third, the PHMS receives conclusions (output) from SEBASTIAN. Fourth, the system submits the XML output to the document generation service for rendering into PDF format (Figure 1). Last, the system identifies the clinic and care manager recipients for each report and alert. Reports are then faxed or emailed securely to clinics and alerts are posted to care managers via a PHMS module called the “message manager.”

System deployment and preliminary evaluation:

In September 2009, the medication management report was deployed in 3 pilot sites. The remaining 12 study sites were added in the following 3 months. A total of 4,221 reports were generated between the initial release date and the end of June 2010. In addition, the PHMS generated 1,097 low adherence alerts to care managers. The randomized controlled trial is scheduled to complete in November 2010. In the interim, 22 providers from 11 sites were briefly interviewed regarding the usefulness of the reports. Overall, 73% of these providers rated the calculated adherence percentages as helpful/very helpful.¹⁴

Discussion

We describe here a large-scale, population-based CDS intervention that aims at improving adherence to a selection of evidence-based pharmacotherapies for a set of high-priority conditions. The intervention and its underlying architecture have several strengths, challenges, and limitations as discussed below.

Strengths: The medication management report and low adherence alert provide timely information that is not routinely available to clinicians. Although most EHR systems offer information on prescribed medications, they often lack medication dispensation information, which can be used as a surrogate for adherence estimates. In addition, EHR medication lists typically are restricted to medications that are prescribed by providers who share the same system, therefore potentially providing only a partial picture of the prescribed medication regimen. In contrast, the medication management report discussed in this study includes a comprehensive view of medications that are dispensed throughout the communities covered by the HIE.

This multifaceted intervention offers different mechanisms to address low adherence issues. While providers can discuss low adherence with their patients during clinic visits, care managers can prioritize their attention to the patients with the lowest adherence and no recent encounters with their providers. Moreover, the medication management report offers evidence-based suggestions in the absence of dispensation events for particular recommended drug classes. These suggestions give the provider an opportunity to review her therapeutic strategy considering the evidence-based advice.

The CDS intervention has features that enable its dissemination to multiple healthcare systems. First, the application-independent, system-agnostic architecture can be integrated into other HIEs or EHR systems via Web services provided that the required data are available within these systems. Previous integration attempts using SEBASTIAN successfully demonstrated the feasibility of such an approach and are currently in production use.^9–12 In addition, several other large scale CDS initiatives are employing a similar service-oriented approach to enable scalable CDS.^15,16 Second, the reliance on ubiquitous claims data, represented with nationally adopted standard terminologies such as ICD-9, NDC, and CPT, maximizes the replicability of the proposed intervention. Third, the CDS intervention imposes fairly low technical requisites on the recipients of the medication management interventions, such as access to scheduling data and fax or e-mail for report delivery. Last, the CDS intervention addresses multiple pharmacotherapy rules that are recommended for the management of common, high-priority conditions. Hence, it is potentially relevant to a large number of patients in typical primary care settings.

Challenges and limitations: While the capacity of the proposed approach to integrate into low technology sites is a clear strength, the desire to support such practices also imposes some challenges. For example, scheduling systems at the study sites often did not offer a standard method for exchanging appointment data nor information technology staff that could help to create an interface. As a result, in most sites, we had to develop these interfaces ourselves using ad-hoc, non-replicable techniques.

Another technical limitation was the reliance on fax technology for the delivery of reports. Unlike graphical user interfaces, faxes are limited to a noninteractive, black-and-white presentation. In addition, faxes generate workflow overhead, since clinic staff members need to receive and properly manage the received reports. An alternative that eliminates dependencies on scheduling data and faxing is the integration between clinic EHR systems and SEBASTIAN through Web services so that providers can request medication management reports on-demand. Such an approach has been successfully implemented within the Duke University Health System for disease management CDS interventions.¹² However, most of the EHR systems available at the study sites are based on closed architectures that preclude integration via Web services. Hence, an optimal interim solution may be a mixed approach, where low technology sites are served through the method described in this study, while sites with more sophisticated technology would be served on-demand through EHR integration via Web-services.

While the reliance on ubiquitous data also contributes to the replicability of the proposed intervention, billing data have several limitations. For example, billing data is not very accurate at the detection of certain diagnoses such as persistent asthma, thereby restricting the range of conditions that can be covered using such data. Furthermore, the post-hoc nature of billing data affects the currency of the report content. For example, our HIE receives pharmacy claims from Medicaid 4 to 6 weeks after a medication dispensation event occurs. In addition, pharmacy claims may have gaps due to factors such as over-the-counter medications; drug benefit coverage by health insurance providers that are not connected with the HIE; and purchase of low-cost generics without using an insurance benefit.

Finally, our HIE is limited to the Medicaid population. Additional challenges arise when attempting to replicate the proposed approach to other insurance populations. For example, access to pharmacy claims data would need to be obtained from various alternate sources, such as private prescription benefit providers and state-wide HIEs.

Future directions: This research has several promising future implications. The results of the large-scale randomized controlled trial will inform the impact of the proposed intervention, perhaps supporting a wider dissemination of the medication management CDS service, such as the integration with North Carolina’s statewide HIE. To contribute to a large-scale dissemination, we are currently developing a next generation of the SEBASTIAN Web service that is based on an open-source model and that is fully compliant with the HL7 International and OMG DSS standard.

Conclusion

We successfully developed and deployed a medication management CDS service based on a service-oriented, system-agnostic architecture. If the ongoing randomized controlled trial demonstrates benefits of the CDS intervention to patient care, particularly improving adherence to evidence-based pharmacotherapies, the intervention has a strong potential to be widely disseminated to other healthcare systems through an open-source model.