John C. Fortney, PhD

 To inform the design of future informatics systems that support the chronic care model.

We describe the development and functionality of a decision support system for the chronic care model of depression treatment, known as collaborative care. Dissemination of evidence-based collaborative care models has been slow, and fidelity to the evidence base has been poor during implementation initiatives. Implementation could be facilitated by a decision support system for depression care managers, the cornerstone of the collaborative care model. The Net Decision Support System (https://www.netdss.net/) is a free Web-based system that was developed to support depression care manager activities and to facilitate the dissemination of collaborative care models that maintain high fidelity to the evidence base.

The NetDSS was based on intervention materials used for a randomized trial of depression care management that improved clinical outcomes compared with usual care. The NetDSS was developed jointly by a cross-functional design team of psychiatrists, depression care managers, information technology specialists, technical writers, and researchers.

The NetDSS has the following functional capabilities: patient registry, patient encounter scheduler, trial management, clinical decision support, progress note generator, and workload and outcomes report generator. The NetDSS guides the care manager through a self-documenting patient encounter using evidencebased scripts and self-scoring instruments. The NetDSS has been used to provide evidence-based depression care management to more than 1700 primary care patients.

Intervention protocols can be successfully converted to Web-based decision support systems that facilitate the implementation of evidence-based chronic care models into routine care with high fidelity.

The Net Decision Support System (NetDSS) is a free Web-based decision support system that was developed from the research protocol of a randomized effectiveness trial of collaborative care (chronic care model for depression). The NetDSS supports depression care manager activities and facilitates the dissemination of collaborative care models that maintain high fidelity to the evidence base. The NetDSS guides the care manager through a self-documenting patient encounter using evidence-based scripts and self-scoring instruments. The NetDSS has the following functional capabilities:

Panel management (patient registry and patient encounter scheduler)

Trial management (management of treatment trials and phases)

Progress note generator and workload and outcomes report generator

Depression care managers (DCMs) are the cornerstone of the depression chronic care model, known as collaborative care.

 Their activities include activating patients (ie, engaging the patient in care), promoting self-management, assessing symptoms and comorbidities, and monitoring adherence and treatment response. Depression care managers facilitate timely communication and coordination among patients, primary care providers (PCPs), and mental health specialists. In randomized trials of collaborative care, DCMs typically receive extensive training and use spreadsheets to monitor the clinical progress of patients.

 and in the patient-centered  medical home,

 clinical information systems are needed to increase fidelity to evidence-based protocols. In the Partners in Care study, fidelity to the DCM protocol was lower than anticipated, and a computerized tracking system was subsequently recommended.

 To address this need, we developed a Web-based clinical decision support system (Net Decision Support System [https://www.netdss.net/]) to support DCM activities. The NetDSS provides context-specific decision support in real time during patient encounters by guiding DCMs through a self-documenting and evidence-based encounter using self-scoring instruments, scripts, and clinical algorithms. The NetDSS has been used to provide evidence-based depression care management to more than 1700 primary care patients.

The objective of this article is to inform the design of future informatics systems that support the chronic care model. We describe the development of a decision support system for collaborative care and its functionality. We conclude with a discussion of future development plans.

The NetDSS was developed with federal research funds and is available to use free of charge. It is based on a highly structured intervention protocol used during a randomized trial of collaborative care conducted in the Department of Veterans Affairs (VA), which significantly improved medication adherence, treatment response, symptom remission, health-related quality of life, and satisfaction with care.

 Hence, the NetDSS is based on an evidencebased research protocol. Core values informing the design include (1) faithfulness to evidence-based protocols, (2) user friendliness and flexibility, and (3) parsimonious collection and dissemination of clinical data. The NetDSS was also designed to have the Desired features of a clinical information system as outlined by Kilbourne and colleagues

 that includes the following: (1) short training time, with uncluttered screens that are easy to navigate; (2) Web-based; (3) use of pull-down menus and logic checks; (4) compliance with Health Insurance Portability and Accounting Act of 1996 standards and conformance to standards of Internet security and encryption; (5) use of clinical reminders; (6) capability of creating individual care plans with self-management information and disease severity rating; (7) capacity to print out summary data on quality and other patient outcomes; and (8) capacity to be linked with, but not a substitute for, electronic medical records.

The NetDSS was developed jointly by a cross-functional design team of psychiatrists, DCMs, information technology specialists, technical writers, and researchers. The design process followed the concurrent engineering model, which includes end users on the design team and emphasizes early and iterative releases of prototypes for end-user testing.

 Version 1.0 was published after 2 years of rapid prototypetest-prototype product development cycles involving alpha testing with hypothetical patients. Later versions  were published over 3 years of beta testing during 3 collaborative care implementation research projects conducted in the VA and federally qualified health centers (clinicaltrials.gov identifiers NCT00304915, NCT00317018, and NCT00439452) involving 11 DCMs, 845 patients, and 21 primary care clinics. A formal usability test was also conducted with 5 DCMs involved in the national rollout of depression care management in the VA. This rollout, known as the primary care—mental health integration, was a centrally organized and funded initiative to promote adoption of collaborative care in VA primary care clinics across the country. In the usability test, the DCMs used the NetDSS to support clinical encounters with patient actors, and DCM feedback was incorporated into version 2.0. Outside of the context of research, version 2.0 has been used during the VA primary care–mentalhealth  integration initiative by 12 DCMs to care for 871 patients.

Rather than develop the NetDSS as a personal computer—based application, we chose a Web-based platform to facilitate dissemination and ease of updates. It is located at the

University of Arkansas for Medical Sciences, Little Rock. A mirror version is hosted behind the VA firewall (http://vaww.dcms.visn16.med.va.gov/), available by VA intranet only. The NetDSS was developed using the Microsoft .Net Framework version 2.0 (Microsoft Inc, Redmond, WA). The Web site uses JavaScript and Ajax to provide a rich user interface that supports asynchronous data manipulation. The application is hosted on a Microsoft Server 2008 by Microsoft Internet Information Services, which is secured using 128-bit SSL standard encryption. Passwords are encrypted using triple DES encryption. The Web application is distributed across a data and Web server. For data access, the system uses ADO. NET (Microsoft Inc) and connects to a Microsoft SQL 2008 server. The NetDSS User’s Manual provides instructions for using the application and can be downloaded from the home page (https://www.netdss.net/).

Depression care managers add patients to the registry and assign them to a clinic and PCP. The administrator (the first individual to register from an organization) can add or delete clinics and PCPs, as well as reassign patients from one DCM to another or from one PCP to another. Patient identification is displayed in a call list, which can be filtered for specific clinics or providers. The NetDSS tracks the schedule of follow-up encounter dates and automatically sorts (and color codes) the call list according to whether follow-ups are due, past due, or recently completed.

The encounter scheduler displays information for a unique patient, including contact information, encounter history, current trial type and phase, past trial history, progress notes, DCM comments, and DCM call notes (eg, “No answer at 10:00 am on Friday” or “Call back after 5:00 pm on Monday”). The encounter scheduler also displays target encounter dates, which can be modified by the DCM. Assessment modules are launched from the encounter scheduler.

Trials are evaluations of treatment plans for individual patients. Trials are divided into the following 4 categories: (1) watchful waiting (symptom monitoring and self-management, but no active treatment), (2) medications alone, (3) counseling alone, and (4) a combination of medication and counseling. Trial duration is defined by start and end dates (eg, an  8-week antidepressant trial), and outcomes are assessed at the end of trials. Trial type is determined by asking the patient about prescribed antidepressants (eg, prescription date, name, and dosage) and about referrals for counseling (eg, referral date and number of appointments). The NetDSS automatically updates the trial type whenever pharmacotherapy  or psychotherapy is initiated, whenever the treatment plan is modified, or whenever the patient discontinues treatment. To identify failed or successfully completed trials, the NetDSS uses hierarchical algorithms based on changes in depression symptom severity. Trial phases are classified as acute or as continuation.

For medication trials and for medication with counseling trials, encounters are conducted more frequently in the acute phase (every 2 weeks) than in the continuation phase (every 4 weeks) to closely monitor side effects and medication adherence. Trials in the acute phase are considered successful if patients respond to treatment (50% improvement in the score of the 9-item depression scale of the Patient Health Questionnaire [PHQ9] on or before the trial end date) or are in remission (2 consecutive encounters with PHQ9 scores <5).

Patients enter the continuation phase after they successfully complete the acute phase. During the continuation phase, the NetDSS uses algorithms to monitor depression symptoms for signs of relapse (2 consecutive PHQ9 scores >5 points higher than the last acute-phase PHQ9 score or a PHQ9 score >10).

 Trials in the continuation phase are considered successful if the patient

remits or does not relapse for 6 months. The NetDSS notifies the DCM whenever patients enter a new phase of treatment or fail to proceed to the next phase after an adequate trial (eg, an 8-week antidepressant trial).

To guide the DCM through an evidence-based encounter, the NetDSS presents a series of multiple-choice questions (with point-and-click menus of predetermined responses), scores patient responses in real time, and branches to the next clinically appropriate question or script. For example, the 

shows a series of questions and script links for the medication adherence module. In this example, the causes of nonadherence questions and scripts are triggered because the patient reported taking his or her medication 9 of the last 14 days. Questions and scripts are divided into the following clinical modules, which are summarized in the 

: education and activation, barriers, self-management, symptom monitoring, comorbidity, counseling adherence, medication adherence, and side effects. Additional decision support is provided in a companion 

, which can be downloaded from the NetDSS home page (https://www.netdss.net/).

The NetDSS is self-documenting. All clinical information collected during an encounter is summarized in an automatically generated progress note. Although highly structured, the progress notes are unformatted to facilitate copying and pasting into an electronic medical record. In addition to documenting the encounter date and trial type or phase, the progress note summarizes barriers, depression severity, suicide risk, comorbidities, adherence, side effects, and self-management goals or progress. The progress note also documents that specific scripts were administered (eg, the DCM addressed medication side effects with the patient). The subject line of the progress note is determined by a hierarchical algorithm that identifies the most clinically important information. When applicable, the title of the progress note states whether the patient has successfully completed the trial or if the trial has failed. Failed trials in the acute phase are defined as less than 50% improvement in PHQ9 score after the trial end date. The NetDSS also alerts the DCM of the following: (1) if there are 2 consecutive encounters with an increase in PHQ9 score that is 5 points higher than baseline, (2) if there are 2 consecutive encounters with side adverse effects, or (3) if there is less than 80% adherence (eg, in the number of days taking medication or attending counseling sessions).

In addition to producing progress notes that summarize each encounter, the NetDSS generates workload and outcomes that present data aggregated by DCMs, clinics, or PCPs. The current date report provides information about all patients in a DCM’s panel and can be used to measure the current workload level. Information in this report includes the total number of patients; the percentages in first, second, and third trials; the mean depression severity; the percentage with suicide risk; the percentage who are nonadherent; the percentage failing the acute-phase trial; and the percentage who are relapsing in the continuation phase. The date range report provides information about all patients in a DCM’s panel within a specified date range and can be used to assess historical workload productivity (eg, the total number of encounters and the mean number of days between encounters), fidelity to the evidence base (eg, the percentage assessed for side effects and the percentage with self-management goals), and overall patient outcomes (eg, the percentage responding to treatment in the acute phase and the percentage completing the continuation phase without relapse).

The NetDSS has the potential to facilitate the implementation of highly structured evidence-based research protocols into routine care. A cross-functional team designed the NetDSS to be user friendly and intuitive. Routine clinical use of the NetDSS led to the identification of unanticipated clinical issues, and continuous refinement has resulted in a robust and flexible decision support system. The NetDSS provides context-specific decision support in real time during encounters with patients by using scripts, self-scoring instruments, and clinical algorithms to identify new trials, treatment phases, and outcome milestones such as nonadherence, treatment response, remission, and relapse.

Other clinical information systems have been developed to support depression care management and have features similar to those of the NetDSS. The IMPACT system

includes a Web-based patient registry that tracks encounter history, current trial type and phase, and past trial history. It cues care manager activities according to a built-in schedule. The system calculates trial end dates and percentage change in PHQ9 scores. Progress notes are automatically generated and, although highly structured, can be exported as simple text to facilitate copying and pasting into an electronic medical record. The system generates treatment summaries for the patient and the PCP and various reports, including ones similar to the NetDSS workload outcomes report and current date report. Behavioral Health Laboratory software

 is installed and run on a central server. The software is designed to address multiple behavioral health issues, including depression, anxiety, posttraumatic stress disorder, alcohol misuse, and cognitive impairment. It includes self-scoring instruments, decision support about diagnosis and severity, and treatment recommendations. The software facilitates triage to specialty mental healthcare and provides decision support for care management. It also facilitates tracking of symptoms and adherence, generates progress notes, and produces reports similar to those of the NetDSS. The TIDES informatics package (http://vaww.portal.gla.med.va.gov/sites/Research/HSRD/ClinicalPart/GTIDES%20Informatics/Forms/AllItems.aspx) is embedded within an electronic medical record (the VA’s computerized patient record system), which has advantages compared with other stand-alone systems. For example, queries can be run to identify important sentinel events such as missed appointments or inpatient admissions. It uses reminder dialogue templates, with self-scoring instruments, check boxes, and free-text fields, to generate progress notes that are saved directly into the electronic medical record, thereby maximizing data security. Outcomes can also be stored as patient data objects, known as health factors, so that data can be aggregated across patients to generate clinic and provider outcome reports. Compared with these other systems, the strengths of the NetDSS are the highly structured clinical algorithms and scripts that provide real-time decision support to DCMs.

The NetDSS lacks one important feature recommended by Kilbourne et al,

 namely, the ability to support multiple chronic illnesses. The NetDSS supports care management for depression only. Data from other depression quality improvement trials suggest that PCPs prefer to implement best practices that offer broad-based support for a range of chronic conditions rather than piecemeal support for a single chronic illness.

 This limitation of the NetDSS results from a lack of research about how to design effective chronic care models for multiple disorders. Because many patients have comorbid psychiatric disorders, the challenge will be to design future clinical information systems around the needs of patients rather than around treatments for specific illnesses. We are developing version 3.0 of the NetDSS from the ground up using Microsoft .Net Framework version 3.5. Version 3.0 offers decision support for multiple mental health disorders that are commonly managed in primary care, including depression, posttraumatic stress disorder, generalized anxiety disorder, panic disorder, and alcohol abuse.

From HSR&D Center for Mental Health and Outcomes Research (JCF, JMP, JSW, AKL, WNR, BAA, JLS), Central Arkansas Veterans Healthcare System, North Little Rock, AR; the Department of Psychiatry (JCF, JMP, CAS, RGH, LOD, JLS), University of Arkansas for Medical Sciences, Little Rock, AR; and Northwest Center for Outcomes Research in Older Adults (LMB), VA Puget Sound Healthcare System, Seattle, WA.

This study was funded by grants IMV 04-360, IIR 00- 078-3, MNT 05-152, MHS-03-218, and MNH 98-000 from the Department of Veterans Affairs; by grant R01 MH076908-01 from the National Institute of Mental Health; and by a pilot grant from the Department of Psychiatry, University of Arkansas for Medical Sciences.

The authors (JCF, JMP, CAS, JSW, RGH, AKL, WNR, BAA, LMB,JLS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (JCF, JMP, CAS, JSW, RGH, AKL, WNR, BAA, JLS); acquisition of data (JCF, WNR); analysis and interpretation of data (JCF); drafting of the manuscript (JCF); critical revision of the manuscript for important intellectual content (JCF, JMP, CAS, JSW, RGH, AKL, WNR, BAA, JLS); obtaining funding (JCF); and administrative, technical, or logistic support (JMP, CAS, JSW, RGH, AKL, WNR, BAA,JLS).

John C. Fortney, PhD, Department of Psychiatry, University of Arkansas for Medical Sciences, 4301 W Markham St, Little Rock, AR 72205. E-mail: 

1. Williams JW Jr, Gerrity M, Holsinger T, Dobscha S, Gaynes B, Dietrich A. Systematic review of multifaceted interventions to improve depression care. Gen Hosp Psychiatry. 2007;29(2):91-116.

2. Wells KB, Sherbourne C, Schoenbaum M, et al. Impact of disseminating quality improvement programs for depression in managed primary care: a randomized controlled trial [published correction appears in JAMA. 2000;283(24):3204]. JAMA. 2000;283(2):212-220.

3. Rost K, Nutting P, Smith JL, Elliott CE, Dickinson M. Managing depression as a chronic disease: a randomised trial of ongoing treatment in primary care. BMJ. 2002;325(7370):e934-e939.

4. Simon GE, Ludman EJ, Tutty S, Operskalski B, Von Korff M. Telephone psychotherapy and telephone care management for primary care patients starting antidepressant treatment: a randomized controlled trial. JAMA. 2004;292(8):935-942.

Articles

This article describes the development and capabilities of a Webbased decision support system for care managers working in the context of the chronic care model.


A Web-Based Clinical Decision Support System for Depression Care Management

To examine the impact of a medication copayment increase on adherence to diabetes, hypertension, and hyperlipidemic medications.

Retrospective pre—post observational study.

This study compared medication adherence at 4 Veterans Affairs medical centers between veterans who were exempt from copayments and propensity-matched veterans who were not exempt. The diabetes sample included 1069 exempt veterans and 1069 nonexempt veterans, the hypertension sample included 3545 exempt veterans and 3545 nonexempt veterans, and the sample of veterans taking statins included 2029 exempt veterans and 2029 nonexempt veterans. The main outcome measure was medication adherence 12 months before and 23 months after the copayment increase. Adherence differences were assessed in a difference-in-difference approach by using generalized estimating equations that controlled for time, copayment exemption, an interaction between time and copayment exemption, and patient demographics, site, and other factors.

Adherence to all medications increased in the short term for all veterans, but then declined in the longer term (February-December 2003). The change in adherence between the preperiod and the postperiod was significantly different for exempt and nonexempt veterans in all 3 cohorts, and nonadherence increased over time for veterans required to pay copayments. The impact of the copayment increase was particularly adverse for veterans with diabetes who were required to pay copayments.

A $5 copayment increase (from $2 to $7) adversely impacted medication adherence for veterans subject to copayments taking oral hypoglycemic agents, antihypertensive medications, or statins.

Adherence to diabetes, hypertension, and hyperlipidemic medications among veterans exempt from copayments and propensity-matched nonexempt veterans was examined at 4 Veterans Affairs medical centers.

A $5 copayment increase (from $2 to $7) adversely impacted medication adherence for veterans subject to copayments.

Copayment increases need to be considered carefully by the Department of Veterans Affairs to ensure that veterans who have greater comorbidity and lower incomes than the general US population do not forgo needed medications.

Diabetes and hypertension are among the most common chronic conditions in the United States, with prevalence rates of 9.6% and 29.7% for individuals age 20 years and over in 1999-2001.

 Between 20% and 40% of patients with these conditions receive no medication management.

 Pharmacologic therapy is a mainstay in the management of both conditions, but only 50% to 70% of patients receiving medication management are adherent to medications.

Improving medication adherence for individuals with diabetes or hypertension has been challenging as health plans and employers increase medication copayments,

 lower limits on the number or total reimbursement of covered medications,1

Between 2000 and 2005, average copayments for commercially insured individuals increased from $7 to $10 for generic medication, $13 to $22 for preferred medications, and $17 to $35 for nonpreferred medications.

 Cost-related nonadherence has increased with higher copayments.

The Department of Veterans Affairs (VA) copayment policy mirrored these market trends by increasing medication copayments from $2 to $7 for a 30-day fill on February 4, 2002. In 1999-2000, 19.6% of veterans had diagnosed diabetes

 Lipid-lowering medications are indicated for nearly all of these veterans. This study examined the impact of the VA medication copayment increase on adherence to diabetes, hypertension, and hyperlipidemic medications by veterans with diabetes or hypertension during a 35-month period (February 2001-December 2003).

This analysis contributes to the extensive literature on the effect of copayments on medication adherence by assessing several therapeutic classes across several conditions, allowing us to examine whether copayment increases have a differential impact across conditions. In addition, we assessed a copayment increase in a population with higher comorbidity and lower income than those in many prior studies, which is critical to clarify whether copayment impacts vary by income.

A recent review article of interventions to impact medication adherence found that many evaluations of formulary and cost-sharing changes lacked control groups or pre—post comparisons, which limited their internal validity.

 We included a colocated control group that controlled for site effects not included in other studies,

 compared adherence before and after the copayment increase,

 and reduced potential bias from the nonequivalent control group with propensity score matching and covariate adjustment that could have confounded copayment effects in prior studies.

 This longitudinal comparison provides information on additional decrements in adherence that might occur with further copayment increases in the VA or private insurers. Identifying patients with chronic conditions who might be especially adversely impacted by copayment increases also could suggest targets for interventions to offset the adherence impacts of increased copayments.

The systemwide increase in the VA medication copayment

 from $2 to $7 in February 2002 created a natural experiment to examine changes in medication adherence for veterans with diabetes or hypertension. Prior to this medication copayment increase, the VA implemented $15  copayments for primary care, $50 copayments for specialty care, and $10 per diem copayments for inpatient care effective December 6, 2001.

 On January 1, 2006, the medication copayment was increased again to $8 for a 30-day fill.

We used a retrospective, pre—post cohort design with a nonequivalent, colocated control group at 4 large tertiary Veterans Affairs medical centers (VAMCs). We identified 60,017 veterans with diabetes (n = 23,182) or hypertension (n = 51,503) who were diagnosed and prescribed a medication for either of these conditions in 2000. Veterans were included in the analysis if they (1) were alive during the entire study period, (2) had a majority of their primary care visits at 1 of the 4 VAMCs, (3) had complete information on level of military service–connected disability to determine exemption from drug copayments, (4) were not hospitalized when the copayment increase went into effect or for more than 1 year during the study period, (5) had at least 1 fill in a relevant drug class during the quarter prior to the copayment change, and (6) had at least 1 fill during the second, third, and fourth quarters prior to the copayment change.19-23 We excluded subjects on non-NPH insulin therapy that would preclude taking oral hypoglycemic agents. We did not exclude patients on NPH, because NPH insulin may be added to oral regimens in a stepped approach. The application of these criteria resulted in analytic samples of 7852 veterans with hypertension, 4407 veterans with diabetes, and 4217 veterans with diabetes or hypertension who were taking statins.

A veteran’s obligation to pay medication (and healthcare) copayments is determined by priority group assignment, based on military service-connected disability for each diagnosed condition, and on income. In 2002, veterans were exempt from medication copayments if (1) their annual income was less than $9556 if single and $12,516 if married; (2) their diabetes, hypertension, or hyperlipidemia was a service-connected disability; or (3) their diabetes, hypertension, or hyperlipidemia was not a service-connected disability, but they exceeded the $840 copayment cap in a given year. Priority group 1 veterans are exempt from all healthcare and medication copayments because the VA has determined that 50% or more of their overall disability is due to their military service, whereas priority group 7 and 8 veterans are required to pay all healthcare and medication copayments because they have no military service—related disability and have income and/or net worth above the VA national income threshold.We excluded veterans in priority groups 2 to 6 from the study because we were unable to determine whether they were required to pay medication copayments.

Unadjusted comparisons of exempt and nonexempt veterans demonstrated significant differences in every observed characteristic. To reduce potential bias from imbalance in observed covariates between exempt and nonexempt veterans and to improve equivalence of the control groups, we conducted 1-to-1 nearest-neighbor propensity score matching with replacement.

 After running 3 logistic regressions to generate propensity scores and matching exempt and nonexempt veterans, 762 exempt veterans with hypertension, 317 exempt veterans with diabetes, and 159 exempt veterans taking statins were excluded because there were no nonexempt veterans with similar propensity scores.

Our final hypertension matched sample included 3545 exempt veterans and 3545 nonexempt veterans. Our final diabetes matched sample included 1069 exempt veterans and 1069 nonexempt veterans. Our final matched sample of veterans with diabetes or hyperlipidemia taking statins included 2029 exempt veterans and 2029 nonexempt veterans. The unit of analysis was person-month with each veteran having up to 35 repeated measures. Human Subjects committees for all coinvestigators’ facilities (Ann Arbor, MI, Durham, NC, Hines, IL, Little Rock, AR, and Seattle, WA, VAMCs) reviewed and approved this study.

We used 4 VA datasets for 2001-2003. All medications dispensed from the VA are recorded in the national Pharmacy Benefits Management database; data elements include drug name, date dispensed, number of days of medication supplied, and dosage.

 The VA inpatient and outpatient care files provided information on veteran demographic characteristics and diagnoses for every inpatient hospitalization and outpatient visit in the national VA system. Benefit Identification and Record Locator System death record data identified which veterans died during the study period. Finally, the Diagnostic Cost Group Hierarchical Cost Category (DCG/HCC) version 6.0 score was used to adjust for overall comorbidity; this measure has been shown to predict veterans’ total costs

Prescription Drug Use and Assessment of Medication Adherence

We calculated monthly medication adherence using the validated ReComp algorithm, a modification of a widely used method that is correlated with a variety of clinical outcomes.

 This algorithm estimates the proportion of days covered for a given measurement interval using the date dispensed and the number of days supplied with each fill. Subjects were considered adherent if they had medications available for at least 80% of each month, which is a conventional threshold that was used to maintain congruence with prior studies.

For adherence to oral hypoglycemic agents (OHAs) (sulfonylureas, metformin, thiazolidinediones) and antihypertensive medications (angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, calcium channel blockers, beta-blockers, and alpha-1 antagonists), we calculated refill adherence separately for each class of medications. We averaged scores to produce a monthly composite OHA adherence score among the diabetes cohort and a monthly composite antihypertensive adherence score among the hypertension cohort. Adherence in veterans with diabetes and/or hypertension taking HMG-CoA reductase inhibitors (statins) was calculated by combining all statin drugs in a single adherence measurement.

There were 3 explanatory variables of interest in the medication adherence analysis using the propensity-matched cohorts: (1) an indicator of whether a veteran was required to make copayments; (2) time indicators for the 12-month preperiod before the copayment increase (February 2001-January 2002), the 12 months (short-term postperiod) just after the copayment increase (February 2002-January 2003), and the subsequent 11-month longer-term postperiod (February 2003-December 2003); and (3) an interaction of the copayment exemption and time indicators to enable a difference-in-difference analysis. The postperiod was subdivided to examine whether adherence differed in the short term and longer term.

All models also were adjusted for age, sex, race, marital status, patient comorbidity measured as DCG/HCC score, whether a veteran was hospitalized in prior or current months, presence of a depression diagnosis at baseline, presence of comorbid diabetes (if hypertension cohort) or hypertension (if diabetes cohort), and the number of all other medications that the patient was prescribed during the preperiod. To adjust for the impact of the December 2001 outpatient visit copayment increases, we adjusted for the number of primary care, specialty care, and mental health visits 90 to 180 days prior to the current month. It is possible that the increase in healthcare copayments would decrease outpatient visits and decrease prescription renewals; the lagged visit counts attempted to control for these cross-price effects.50 Models that included a pre—post indicator for initiation of the healthcare copayment generated similar results (results not presented).

Logistic regressions were estimated to identify whether a veteran was exempt or nonexempt from copayments and to generate predicted probabilities of being nonexempt, which served as the propensity scores for matching exempt and nonexempt veterans in the diabetes, hypertension, and statin cohorts. The propensity score model for the diabetes cohort included 19 main effects related to demographics, comorbidity, and medication burden and 18 interactions, because this specification reduced covariate imbalance within propensity score quintiles better than other specifications. Similar iterative specification tests were conducted for the logistic regressions on the hypertension cohort with a final specification including 11 main effects and 18 interactions, and the statin cohort with a final specification including 20 main effects and 28 interaction terms. Detailed results are given in 

Bivariate statistics (t tests, χ2 test) were estimated to compare patient characteristics between matched exempt and nonexempt veterans, and to compare medication adherence at baseline (February 2001) and the last month of the study period (December 2003). To examine pre—post differences between propensity-matched exempt and nonexempt veterans in medication refill adherence over 35 months (12 preperiod and 23 postperiod months), we used generalized estimating equations assuming a binomial distribution, logit link, and independent working covariance structure with person-month as the unit of analysis. The working covariance structure was specified to obtain unbiased parameter estimates for the timevarying covariates.

To estimate first differences, predictions from these generalized estimating equations were obtained to identify the proportion of exempt and nonexempt veterans in each cohort who were adherent in the preperiod, the immediate postperiod, and the longer-term postperiod. Confidence intervals (CIs) for these proportions were estimated through 1000 bootstrap iterations. All analyses used Stata software (StataCorp, College Station, TX). All reported P values are 2-sided, and significance was lowered to P = .01 to account for multiple comparisons.

Propensity matching eliminated imbalance in several covariates (proportion of veterans with diabetes, hypertension, or both; marital status, sex, race, baseline hospitalization rates, and number of diabetes medications) across all 3 cohorts, but did not eliminate imbalance in site composition (

). Nonexempt veterans with diabetes were slightly older than exempt veterans with diabetes (

 <.001) and were taking more hypertension medications (

 <.001) but fewer medications of all kinds (

 <.0001). Nonexempt veterans with diabetes had slightly fewer primary care and specialty care visits on average than exempt veterans (both 

Nonexempt veterans with hypertension were older than exempt veterans with hypertension (

 <.0001), had slightly higher DCG/HCC risk scores (

 <.01), and fewer specialty care visits (

 <.0001). Nonexempt veterans with diabetes and/or hypertension taking statins were older than exempt veterans taking statins (

 <.0001), were taking fewer medications overall (

 <.001), had slightly fewer primary care, specialty care, and mental health visits on average (all 

 <.0001), and slightly fewer mental health visits (

The unadjusted proportion of nonexempt and exempt veterans with diabetes who were adherent to their OHAs at baseline (February 2001) was similar, but the proportion adherent in the last month of the study period (December 2003) was significantly lower among nonexempt veterans with diabetes (60% vs 69%; 

 <.0001). A greater proportion of nonexempt veterans with hypertension were adherent to their antihypertensive medication at baseline (80% vs 76%; 

 <.0001), but there were no differences in adherence in December 2003. Unadjusted adherence rates were similar between nonexempt and exempt veterans taking statins.

After covariate adjustment, OHA adherence among exempt veterans with diabetes increased 4.1 percentage points in the year after the copayment increase (February 2002-January 2003) compared with the preperiod, while adherence remained constant for nonexempt veterans subject to copayments (first difference −3.8%; 95% CI = −3.7%, −3.9%). In the longer-term postperiod (February-December 2003), OHA adherence declined for both diabetes cohorts compared with the preperiod but significantly more so for nonexempt veterans (−10.3% vs −0.9%; 

 <.001) (first difference −9.6%; 95% CI = −9.5%, −9.8%). See 

Adherence to antihypertensive medications increased for exempt and nonexempt veterans in the year after the copayment increase (4.1% vs 5.9%; first difference −1.8%; 95% CI = −1.8%, −1.9%) compared with the preperiod, but decreased thereafter for both groups compared with the preperiod. The decline in adherence to antihypertensive medications was greater for nonexempt veterans (−5.4% vs −2.3%; first difference −3.2%; 95% CI = −3.1%, −3.3%). See 

Adherence to statins increased for exempt and nonexempt veterans in the year after the copayment increase compared with the preperiod, but more so for exempt veterans (3.5% vs 6.6%; first difference −3.0%; 95% CI = −2.9%, −3.1%). Statin adherence continued to increase (1.2%) for exempt veterans in the longer-term postperiod (February-December 2003) compared with the preperiod, but decreased for nonexempt veterans (−1.9%; first difference −3.1%; 95% CI = −3.0%, −3.2%). See 

Adherence to OHAs, antihypertensive medications, and statins by veterans with diabetes or hypertension increased in the year after a $5 medication copayment increase (from $2 to $7), but subsequently declined. The change in adherence between the preperiod and postperiod (as indicated by the first differences) was significantly different for exempt and nonexempt veterans in all 3 cohorts, which indicates that the medication copayment increase had adverse effects on medication adherence, as has been found in similar studies. The longer-term impact of this policy change was particularly adverse for veterans with diabetes who were required to make copayments, because their adherence to OHAs 13 to 23 months after the copayment increase was 10.3% lower than their preperiod adherence and 9% lower than the adherence of comparable veterans who were exempt.

The initial increase in adherence may have been due to VA physicians encouraging all of their patients (regardless of copay status) to stockpile medications in anticipation of the copayment increase, because VA physicians tend not to be aware of a veteran’s copayment exemption status and to treat all veterans in their panel similarly. This initial increase and subsequent decline also could have been due to a lagged effect of the copayment increase on adherence, because veterans may initially refill medications until their budgets get stretched and cost-related nonadherence ensues.

Cost-related nonadherence may have been reinforced by the cross-price effect

 of the increased primary care copayment (from $0 to $15) and specialty care copayment (from $15 to $50) that occurred 2 months prior to the medication copayment increase. Increased healthcare copayments could impact adherence rates in 2 ways. The out-of-pocket costs of outpatient visits could have reduced visit rates at which prescriptions would be initiated, modified, or renewed. (In the postperiod, we observed significantly lower rates of primary care visits for veterans in all 3 cohorts [

 <.0001].) In addition, veterans with fixed incomes who were keeping their outpatient appointments might sacrifice their medications to pay for these visits.

 It also is possible that the initial adherence increase and subsequent decline could have been an artifact of our inclusion criteria. In a sensitivity analysis, we included veterans with 2 or more fills in the 12 months prior to the copayment increase (instead of 1 fill in the 3 months prior and another fill 4-9 months prior), but the difference-in-difference results were unchanged because trends were similar for exempt and nonexempt veterans.

The decline in adherence to OHAs was significantly greater for veterans required to pay copayments than for exempt veterans, but the change in adherence to antihypertensives and statins was more modest. It may simply be that patients value their cardiovascular medications more than their diabetes medications, as has been shown in prior studies.

 A study of commercially insured populations by Goldman et al

 found that patients with diabetes had a greater reduction in days supply in response to a doubling of copayments than patients with hypertension or hyperlipidemia (25% vs 10%), which is consistent with our results. However, a study of nonelderly, disability-eligible Medicare beneficiaries by Soumerai et al

 reported similar rates of cost-related nonadherence for beneficiaries with diabetes and beneficiaries with hypertension. Finally, our statin findings are consistent with a recent study of veterans from 1 VAMC taking lipid-lowering medications,

 despite several study design differences that make comparison difficult, including that study’s significantly larger sample size, less equivalent control group, lack of propensity score matching, and more limited covariate adjustment.

We attempted to improve upon prior studies of cost-sharing and adherence by including a colocated control group

 instead of a nonequivalent, geographically distinct control group,

 by reducing potential bias through propensity score matching, and by contrasting pre—post changes of treatment and control groups.

 We also examined medication adherence across a number of conditions to illustrate a range of clinical impacts of the copayment increase, which provides further evidence that the adherence decline that we observed in the diabetes sample was due to the copayment increase and not other factors.

However, several limitations remain. We were unable to track veterans’ use of non-VA medications. It is possible that veterans with Medicaid or private insurance who became nonadherent to VA-acquired medications were simply obtaining them elsewhere, but this omission is likely to be minimal because VA copayments were lower than prevailing rates in commercial insurance and Medicare Part D plans at the time, and Walmart prescription drug programs were not available until 2006. The generalizability of our results is somewhat limited because our sample was drawn from 4 large VAMCs. However, we chose geographically dispersed VAMCs to reduce small area variation biases. Despite the propensity score matching, there remained a few differences in observed factors between exempt and nonexempt veterans, although we significantly reduced the extent of covariate imbalance and controlled for remaining imbalances directly in the adherence regression. However, the longitudinal natural experiment enabled us to control for fixed person-specific effects and time trends to minimize unobserved confounding, whereas propensity score matching and covariate adjustment reduced the likelihood of observed or unobserved confounding. These results appear to be robust.

It appears that a $5 copayment increase from $2 to $7 was sufficiently large to adversely impact medication adherence among veterans who have greater comorbidity and lower incomes than the general US population. The VA has since increased medication copayments to $8. Future copayment increases need to be considered carefully. If implemented, copayment increases should be matched with nonfinancial interventions to offset cost-related nonadherence. The VA may want to consider linking copayments with the clinical value of medications, because copayment reductions for high-value medications have been shown to reduce nonadherence in commercial populations.

 Given that many veterans taking diabetes, hypertension, and hyperlipidemic medications are elderly or near elderly and are on fixed incomes, such copayment reductions may generate a sizable response if adherence changes mirror the decline observed in this analysis in response to a $5 copayment increase.

This work was supported by the Office of Research and Development, Health Services Research and Development Service, Department of Veterans Affairs, project number IIR 03-200. The views expressed are those of the authors and do not reflect the views of the Department of Veterans Affairs.

Dr Maciejewski reports serving as a consultant for Takeda Pharmaceuticals and reports owning stock in Amgen. The other authors (CLB, MP, DKB, FEC, JCF, SLK, KTS, NDS, and C-FL) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

A copayment increase from $2 to $7 adversely affected veterans' adherence to statins, antihypertensives, and oral hypoglycemic agents.


John C. Fortney, PhD

Articles

A Web-Based Clinical Decision Support System for Depression Care Management

Increasing Copayments and Adherence to Diabetes, Hypertension, and Hyperlipidemic Medications