Healthcare Information Technology Interventions to Improve Cardiovascular and Diabetes Medication Adherence
Published Online: December 22, 2010
Alexander S. Misono, BA; Sarah L. Cutrona, MD, MPH; Niteesh K. Choudhry, MD, PhD; Michael A. Fischer, MD, MS; Margaret R. Stedman, PhD; Joshua N. Liberman, PhD; Troyen A. Brennan, MD, JD; Sachin H. Jain, MD, MBA; and William H. Shrank, MD, MSHS
Nonadherence to essential medications is an important public health problem.1,2 Patients commonly do not adhere to chronic medication therapy, leading to substantial morbidity, mortality, and excess healthcare costs.3-6 Despite the significant impact of nonadherence on health, solutions are elusive. Meta-analyses evaluating interventions to improve adherence have consistently supported the use of resource-intensive, multifactorial approaches with multiple elements delivered over time, such as self-management plans, reinforcement, or rewards.7,8 In an effort to identify more cost-effective solutions, researchers and clinicians have begun exploring the role of healthcare information technology (HIT) in medication adherence interventions.
There are numerous potential applications for HIT in a medication adherence intervention. Electronic pharmacy data may allow for identification of nonadherence and facilitate data delivery to prescribers and pharmacists.9 Electronic systems might inexpensively remind patients and providers about refills. Interactive electronic systems may be used to educate patients about appropriate medication use, and widespread online connectivity should accommodate more sophisticated monitoring, interaction, and communication.
Although rigorous evaluation of the effectiveness of HIT adherence interventions is essential, little systematic assessment has been done to date. We conducted a systematic review of HIT interventions designed to improve medication adherence in cardiovascular disease and diabetes. Our goal was to assess the state of evidence in this field, identify interventions that were successful, and ascertain specific features of interventions that seem to be most associated with success.
We performed a systematic search of peer-reviewed journals between 1966 and 2010 using MEDLINE and EMBASE. We limited our search to randomized controlled trials.
Our search terms related to the type of study (ie, randomized controlled trial); adherence (ie, adherence OR compliance OR medication adherence OR treatment adherence); prescription drugs (ie, drug OR medication OR antihypertensive OR antihyperlipidemic OR hypoglycemic); and cardiovascular disease and diabetes (ie, myocardial infarction, coronary heart disease, heart failure, hypertension, hyperlipidemia, OR diabetes). Articles with at least 1 search term in 3 of the main categories (study type AND adherence AND either drug OR disease) met criteria for review.
Search terms and parameters were adjusted for both databases (MEDLINE and EMBASE) while maintaining a common overall architecture. Search results were then screened for duplicate entries, which were removed.
Studies were included if they reported results of randomized controlled trials studying interventions to improve adherence to medications used for prevention or treatment of diabetes or cardiovascular disease, the greatest source of mortality in the United States.10 We included only randomized trials in order to promote interventions based on the highest quality of evidence. Studies were limited to adult subjects (aged ≥18 years). Of these interventions, we included only interventions with any electronic component. Examples included the identification of patients with electronic tracking of adherence, electronic reminders to take medication, or electronically enhanced communication with patients or providers. Non-English studies were excluded.
After exclusions, 13 articles (Figure) were classified into 2 groups. The first group described the type of interaction with the patient: 1-way patient reminder systems, 2-way interactive systems, or systems to enhance patient–provider interaction. We selected these categories to assess how to best deliver interventions to patients, whether by simple patient reminders, engagement of patients with an electronic system, or enhancement of communication with the provider. Studies of reminder systems featured interventions providing audio and/ or visual reminders of medication dosing. Interactive systems included computer-based tools aimed at patient education, counseling, and/or promoting favorable patient behaviors. Systems interacted with patients either immediately or via delayed feedback (eg, customized reading material).
The second group described the type of physician engagement. These types of interventions included those in which no real-time adherence information was passed on to providers and those that incorporated real-time feedback to providers. We included this categorization in an attempt to assess the incremental value of delivering additional feedback to the provider. In a third type of intervention, providers (or in 1 case, research assistants) directly interacted with patients as part of the intervention.
Data were extracted by 3 investigators (ASM, SLC, WHS), with disagreements resolved by consensus. We assessed a number of variables related to the organization and outcome of studies including study design, setting, characteristics of population studied, number of participants, mean age (or age range) of participants, characteristics of the intervention, methods used to measure medication adherence, and medication adherence outcomes. Confidence intervals (CIs) were reported where they were available and P values where no CIs were available.
We identified those randomized controlled trials where means and standard deviations for medication adherence outcomes were presented. For these studies, we computed Cohen’s d effect size (ES) statistics, which can be calculated for outcomes that are either binary (eg, survey responses or predefined adherence cutoffs) or continuous (eg, proportion of days covered).11,12 The ESs compare the difference in effect between the study groups divided by the standard deviation of this difference. When standard deviations were not reported, we derived them from the P value or t test statistic.
Using standard methods, we considered an ES of less than 0.2 to be very small, an ES of 0.2 to less than 0.5 to be small, an ES of 0.5 to less than 0.8 to be medium, and an ES of 0.8 or greater to be large. We assumed that the estimated Cohen’s d statistics were independent of scale, sample size, and the standard deviation of the outcome studied.
PDF is available on the last page.