The term that David Sackett used to describe adherence bias in his 1979 paper “Bias in Analytic Research” was “compliance bias”. However, in describing the bias he referred to “experiments requiring patient adherence to therapy”, and it is clear that adherence to therapy was what he was talking about. Adherence is important, because questions about the efficacy of an intervention become confounded by the problem of non-adherence, for example when, as in the example he gave, “it is the high-risk coronary patients who quit exercise programs.”
Participants who adhere to an intervention may differ from those who do not, and in ways that may affect the risk of the outcome being measured. This means that adherence is a confounding factor in assessing the effect of the intervention. Furthermore, suboptimal medication adherence is a major determinant of poor treatment response.
The term “adherence” in this context has replaced the older terms “compliance” and “concordance”, which, for different reasons, are considered unsatisfactory.
It is worth noting, to avoid confusion, that “compliance” is still a relevant term when referring to mandatory systems. For example, there are certain legal requirements with which a company must comply before it can market a medical device, and the UK’s Medicines and Healthcare products Regulatory Agency (MHRA) has issued guidance on how compliance with the regulations is assessed and enforced. This is not adherence but compliance.
Previous uses of the term “compliance bias” in the relevant literature have referred to what should be called “adherence bias”. For example, an essay on “compliance bias” by the late biostatistician Alvin Feinstein (Feinstein, 1974) began: “Most discussions of compliance are concerned with a patient’s maintenance of an assigned therapeutic regimen.” He later pointed out in the same essay that “A substantial literature has begun to develop on compliance with therapeutic regimens.” He then went on to discuss “six features of compliance that can affect the biostatistical data and interpretations … regimen compliance, the evaluation of compliance, the non-compliance ‘control’, protocol compliance, the ‘compliance sample’, and the ‘compliance-confounded cohort’”, each dealing with aspects of what is now called “adherence”
Adherence to a medication may be a surrogate for factors that improve health outcomes, such as fractures. Little is known about the size of this potential “healthy adherer” effect. One study evaluated the hypothesis that greater than 80% adherence to placebo as a dispensed study medication was associated inversely with bone loss and fractures among women participating in the Fracture Intervention Trial (FIT) using daily medication diaries.
Among 3169 women randomized to placebo, 82% had high adherence. Bone loss at the total hip was lower in the adherent placebo-treated women (P = 0.04), and there was a nonsignificant reduced risk of hip fractures in women with high placebo adherence compared with those with poor adherence (adjusted hazard ratio = 0.67, 95% confidence interval = 0.30-1.45).
Adherence to prescribed treatment has been shown to reduce adverse outcomes in COPD and ulcerative colitis research. It is therefore important to know about and account for adherence when investigating the effects of an intervention.
In 1980, the Coronary Drug Project Research Group explored the effect of adherence on mortality in the long-term treatment of coronary heart disease. Good adherers to clofibrate (defined as patients who took 80% or more of the protocol prescription during the five-year follow-up period) had substantially lower five-year mortality than poor adherers (15% vs 25%). There were similar results in the placebo group (15% mortality for good adherers vs 28% for poor adherers).
Good adherence to interventions by study participants should be encouraged, for example by using medication reminder methods, such as text messages via mobile phones, smartphone apps, and electronic medication boxes, in order to obtain the best evidence about the potential benefits and harms of the intervention being tested in the study.
Clinical trialists should try to collect data on adherence, and while intention-to-treat analyses should be included, exploratory secondary analyses investigating the impact of non-adherence will help inform the minimum degrees of adherence that should be necessary when assessing the outcome of interest