Developing benchmarks for adherence studies

In the context of evidence-based medicine and increasing pressure on limited healthcare resources, more attention has recently turned to extending the information from clinical trials to clinical practice. A number of factors relating to physician, patient, drug and system factors have to be weighed into the translation of safety and efficacy data into effectiveness information [1]. More recently, in the wake of controversial evidence on cyclooxygenase-2 inhibitors, much of the national discussion has now gravitated toward questions of how best to assess safety and efficacy as necessary conditions for marketing a drug [101]. At the same time, by definition, or by virtue of the clinical trial design, these conditions cannot be fully satisfied prior to marketing. It is only after a critical period of time of utilization that effectiveness can be fully established, sometimes even varying across subpopulations.

One critical consideration in establishing effectiveness information is adherence to medications. A correct assessment of adherence in a given population can be as critical to risk–benefit as to cost–effectiveness studies [2]. Correct understanding of adherence patterns is of great interest to researchers, physicians, other providers, drug sponsors, regulators, managed care and government payers alike. The data provided by clinical trials are often based on the premise of near full adherence. In fact, clinical trials control for a given level of adherence, and often have a run-in period through which they weed out subjects who are not compliant with the trial regimen.

Why is it important to measure adherence?

Adherence to medication regimens is one of the most significant determinants of therapy outcomes. It has been defined as the degree to which a patient’s behavior coincides with the clinician’s recommendations for a certain treatment [3]. Although a number of studies have shown that adherence to prescribed medications improves disease control [4], only approximately a third of patients are estimated to adhere completely with their healthcare provider’s advice [5].

Full adherence to prescribed drugs can lead to successful therapy outcomes provided there is an accurate diagnosis and an appropriate prescription of medications. Indeed, if correlations between adherence and outcomes are to be appropriately established, it is critical that there be a correct method to measure how appropriate the prescription is, perhaps in relation to established clinical guidelines. Adherence studies have shown that rates may vary depending on the term of treatment, and that rates of adherence with long-term treatments are approximately 50% regardless of the disease being treated [6]. On the other hand, while compliance with short-term medication treatments is commonly considered to be higher than with long-term therapy, fast declines in adherence are generally observed in the initial days of short-term therapy [7].

It has been argued that nonadherence is the most common cause of no reaction to medication [3]. In addition, there is strong evidence suggesting that patients who adhere to therapy have better outcomes than patients who discontinue their treatment, even when being administered placebo [3]. It is possible that in these instances, adherence with medical advice relating to drug treatment serves as a proxy for compliance with other guidelines for behaviors related to correct health maintenance, which, in general, sustain favorable positive health outcomes.

Adherence from different perspectives

Adherence measures and interventions depend heavily on the perspective of the stakeholder. From the patient perspective, nonadherence to therapy may lead to inferior health outcomes, possibly directly or indirectly, affecting functions of daily living and other quality-of-life factors. Studies have shown that patients with low compliance are at relatively higher risk of morbidity and mortality, are more likely to show dissatisfaction with care and may even incur higher out-of-pocket expenses than patients who adhere to treatment recommendations [8,9]. A patient’s decision whether or not to adhere to therapy is presumably based on an informal and personal risk–benefit analysis. Often the final decision, whether reached passively or actively, is a function of several factors relating to drug side effects, mode of administration, dosing schedule, the patients’ priority to take the drug, their perception of the seriousness of the disease, their belief in the efficacy of the medication and their acceptance for a change [10]. From the clinician’s perspective, it is of utmost importance to know whether the patient is following directions and taking the drug as prescribed, to check for side effects, adjust dosage specifications, add other drugs if required or switch to alternative therapies. Physicians often try to improve adherence by searching for modalities to increase patient compliance. Those modalities include simple regimens and prescriptions that minimize side effects while optimizing health outcomes. From the perspective of managed care payers, it is essential to understand adherence within certain populations and across given drugs. Adherence patterns can help guide formulary decision-making. Indeed, poor adherence patterns may eventually translate into higher healthcare resource utilization leading to higher medical and transaction costs [11]. For example, a retrospective cohort study by Hepke and colleagues found that increased adherence to pharmaceutical therapy was associated with a decrease in medical care services utilization (not with lower costs) in a diabetic population [12].

Factors affecting adherence

A broad variation of nonadherence rates exists in the literature. This may be caused by the variability of contributing factors in different settings, the application of different method-ologies for measuring adherence rates and different views in the interpretation of adherence.

Factors affecting adherence include lack of understanding of the physician’s instructions, lack of affordability of the therapy, side effects, dosage and prescription modality and socio-demographic factors [13,14]. For example, a recent study by Piette and colleagues found that out-of-pocket costs and insufficient prescription coverage may lead to low adherence issues in chronically ill patients [15]. Furthermore, recent studies have found that nonmedication-related issues may also influence adherence. The importance of communication between the patient and the physician is underscored, for example, by a new interest in cultural fluency programs [16]. It has been shown that patients are more likely to be adherent with therapy if they perceive their physician to be approachable, display concern and communicate well with them [10].

Measuring adherence

There are several methods for measuring adherence. These include the time-to-prescription-refill method and the self-reported method. In the first method, data are usually collected from claims databases and rely on counting the days elapsed between consecutive drug prescription claims; nonadherence is defined as a discontinuation of the therapy (i.e., a failure to refill). In the second method, primary data are collected directly from the patient. There is some evidence that when refill data are available, the time-to-prescription-refill method may be a better approach to measure adherence [17]. These types of data are usually modeled by survival analysis techniques such as Cox proportional hazard models or logistic regression models, to compare the relative likelihood of adherence with certain therapies. The difference between these two modeling techniques is that the first is time dependent whereas the second is time independent. Some researchers favor more of a cross-sectional approach and use the Medication Possession Ratio (MPR). MPR is defined as the ratio of the cumulative number of the drug days supply, over the period of time elapsed between the first day of the first prescription and the end-of-day supply of the last prescription. This ratio should be interpreted with caution, since it provides an estimate of the proportion of time that the patient had possession of the drug, but no indication as to consecutive use of the drug. The MPR may be more appropriate for drugs used for chronic conditions (e.g., calcium channel blockers or diuretics) than for drugs (e.g., antibiotics) indicated for acute conditions, where timing of the drug is very important and predictive of outcomes.

Related studies of adherence

Numerous studies have addressed the association between adherence to drug therapies and health outcomes. Some of the conditions covered are diabetes [18–20], depression [21–23], respiratory diseases [24–26], hypertension [27,28], cerebrovascular [29,30] and heart disease [31,32]. Most studies showed that good adherence favored better health outcomes. For instance, the study by Cheng and colleagues examined the pattern of adherence to statin therapy in a cohort of Chinese patients at high risk of coronary heart disease. This group also looked at the association of adherence to statin therapy and the control of serum low-density lipoprotein cholesterol. They found high adherence to statins but a weak association between adherence and low-density lipoprotein reduction [32]. Another study by Williams and colleagues estimated the proportion of poor asthma-related outcomes attributable to inhaled corticosteroids (ICS) nonadherence. Their findings suggest that poor adherence to ICS among asthma patients is correlated with poor asthma-related outcomes and that most asthma-related hospitalizations are caused by less than perfect adherence to ICS [25].

Benchmarks for adherence

Most of the literature has focused on methods of measuring and improving adherence, but very little is known about how much adherence is appropriate. This concept is related to what is commonly referred to as the forgiveness of the drug. Drug forgiveness is defined as the threshold of adherence above which the marginal benefits of additional adherence are negligible. An underlying assumption within this concept is that a correlation exists between adherence levels and health outcomes, and that prescribed treatment reflects guidelines and predicts given improvements in health outcomes. For example, there is a multitude of studies that investigate the association between treatment adherence and health outcomes for patients with Type 2 diabetes mellitus. However, there appears to be little to support a target desired threshold level of adherence in these patients. Most of these studies reported adherence rates to blood glucose-lowering therapy between 65 and 85% [33–40,101]. Few have found rates between 36 and 54% for specific patient populations and dosage regimens [39,41–43].

It is commonly agreed that an 80% level of adherence is usually acceptable to label a patient adherent. However, most studies comparing relative adherence levels appear to imply that higher rates of adherence are preferred, often with no reference to a baseline acceptable benchmark.

In addition to a need for a clinical benchmark for adherence rates, it is important to understand and evaluate the cost–effectiveness/benefit of improving adherence before initiating programs aimed at increasing adherence rates, if and when those rates are already adequate. One possible approach could be to collect data about adherence rates and outcomes of interest and use threshold regression techniques in order to estimate a threshold above which the marginal benefit of adding a point for the adherence rate would be negative. Such regression techniques were developed by Hansen and are currently used in cross-sectional studies in economics [44]. Hansen’s threshold regression technique, based on least squares estimation of the coefficients, splits the sample endogenously, and tests for the statistical significance of the threshold level that deternmines the split. If we let y_i and x_i be the vectors of dependent and independent variables respectively, the threshold regression model is then represented as Equation 1:

where p_i, which may or may not be included in x_i, is the threshold variable that splits the sample. Hansen uses a sequential search algorithm to locate the value of γ and the estimates for β₁ and β₂, and employs a Lagrange multiplier to test for the statistical significance of the split.

Threshold regression models would be of interest to determine the threshold of adherence or forgiveness of the drug. However, such studies should be cautiously designed since it is important to apply a valid measure of adherence and to control for all factors that may influence clinical outcomes. In addition, those studies may be sensitive to drug dosage, the types of patients and drug characteristics.

Conclusion

Adherence to drug therapy is essential for an effective response to therapeutic treatments. It is critical, however, to have established reference levels of adherence benchmarks, by disease, for a proper interpretation of results. More research is needed to develop valid techniques for adherence measurement and benchmarks. In this review, examples of studies performed on adherence from the literature were highlighted. The authors also highlighted the need for further research in the area of drug forgiveness, the need to look at how much adherence can be considered to be adequate in effectiveness studies, and to analyze the cost–effectiveness of additional programs for adherence improvement. The authors introduced the concept of a threshold regression technique that can be used for adherence threshold estimation.