Medication Adherence

Abstract

Medication adherence is the extent to which a patient's drug-taking corresponds to the regimen agreed with the prescriber, and across conditions roughly a quarter of patients fall short. Because the shortfall is only weakly tied to how much patients know, nonadherence belongs with the intention-behavior gap rather than with ignorance. The field has moved from a compliance vocabulary toward adherence and concordance, treating drug-taking as a reasoned behavior shaped by patients' beliefs, and a standardized taxonomy now separates initiation, implementation, and persistence. This article covers the terminology, the measurement problem, the determinants, the clinical and economic consequences, and the cognitive models that explain why even the better interventions raise adherence only modestly. Three demonstrations model adherence measurement, the necessity-concerns balance, and the limited reach of interventions.

Keywords: medication adherence, patient compliance, necessity-concerns framework, ABC taxonomy, nonadherence

Medication adherence is where pharmacology meets the psychology of everyday behavior. A drug that works in a trial works in a life only if it is actually taken, and taking a medicine on schedule for months or years is a sustained behavior that competes with forgetting, doubt, side effects, cost, and the simple fact that chronic illness is often silent between doses. The question that organizes the field is deceptively plain — why do so many patients, prescribed an effective treatment they can afford, not take it as directed — and its answer has reshaped how clinicians think about the patient's role, moving from an expectation of obedience toward a recognition that drug-taking is a decision the patient makes and remakes every day.

What Medication Adherence Is

Medication adherence is the extent to which a person's medication-taking behavior — taking the drug, at the right dose, at the right times, for the prescribed duration — corresponds to the regimen agreed upon with a health-care provider. The emphasis on agreement is deliberate and comparatively recent. For decades the field spoke of compliance, a word that framed the ideal patient as one who obeyed the doctor's instructions and cast any deviation as the patient's failure. Dissatisfaction with that framing, both because it misdescribed the behavior and because it predicted it poorly, drove a shift toward adherence, which keeps the regimen as the reference point but treats the patient as an active decision-maker, and further toward concordance, which locates the standard not in the prescriber's orders but in a negotiated agreement between prescriber and patient (Chakrabarti, 2014).

The scale of the problem is what makes it a field rather than a footnote. A quantitative review pooling fifty years of research across many diseases and regimens found that on average about one patient in four does not adhere to medical recommendations, with adherence worse for longer regimens and for prevention than for cure, and strongly shaped by the disease, the complexity of the treatment, and the quality of the patient-provider relationship (DiMatteo, 2004). A clinical review aimed at physicians summarized the same reality in a memorable formulation — that drugs do not work in patients who do not take them — and laid out the layered reasons, from forgetting and cost to belief and side effects, that adherence is so hard to sustain (Osterberg & Blaschke, 2005). Figure 1 shows the three phases into which contemporary work divides the behavior.

Types of Medication Adherence

MeSH files medication adherence as a narrower descriptor beneath Patient Compliance, the broader heading for the degree to which patients follow any medical advice, of which taking medicine is one instance. Beneath medication adherence MeSH catalogues a single direct child. Because MeSH is an indexing vocabulary rather than a psychological taxonomy, this subtype reflects how the published literature is tagged, not a claim that adherence decomposes into cleanly separable kinds; the ABC phases in Figure 1 are the more analytically useful division. The one child names a distinct, heavily studied delivery strategy that has its own article on this site.

The single child is a reminder that most of what the literature calls types of adherence are really ways of measuring or supporting it rather than natural categories of the behavior. The phase distinction — initiation, implementation, persistence — carries more analytic weight, because a patient who never starts, one who takes doses erratically, and one who stops after six months have three different problems that call for three different responses (Vrijens et al., 2012).

Measuring Adherence

Adherence cannot be improved until it can be measured, and every available measure is a compromise. Methods divide into direct approaches — observed ingestion, or assay of the drug or a metabolite or a biological marker in blood or urine — and indirect approaches — pill counts, pharmacy refill records, electronic monitoring of bottle openings, and patient self-report (Lam & Fresco, 2015). Direct methods are the most objective but are costly, invasive, and defeatable, and they capture a single moment rather than a pattern. Indirect methods scale but each measures a proxy: a refilled prescription is not a swallowed pill, a counted pill may have been discarded, and a self-report is systematically inflated by social desirability, because patients tend to report more adherence than they achieve.

Refill-based measures have become the workhorses of large-scale research precisely because pharmacy data are routinely collected. The two most common are the medication possession ratio and the proportion of days covered, both of which express the supply of medication a patient obtained as a fraction of the days in a period, and both of which are conventionally dichotomized at a threshold — most often 80% — above which a patient is labeled adherent. The threshold is a convenience, not a biological constant; the clinically meaningful level of adherence depends on the drug's forgiveness, the margin by which its effect survives missed doses. The first demonstration lets the reader compute an adherence ratio from a dosing record and see how the 80% convention classifies it.

Over a fixed 180-day period, set how many days of medication a patient was dispensed. The medication possession ratio (MPR = supplied / period) is uncapped, so an early refill can push it past 100%; the proportion of days covered (PDC) is the stricter measure, capped at 100%. The dashed line marks the conventional 80% adherence threshold.

PDC = 83% — this patient is classified adherent at the 80% threshold

The 80% cut-point is a research convention, not a biological constant: the level of adherence that actually preserves a drug's effect depends on how forgiving the drug is of missed doses (Lam & Fresco, 2015). A ratio is also only as good as its source — a refilled prescription is not a swallowed pill.

Why Patients Do Not Adhere

The determinants of nonadherence are many and have been organized in several overlapping schemes. The World Health Organization's widely used framework groups them into five interacting dimensions — social and economic factors, the health-care team and system, the condition itself, the therapy, and the patient — and a clinical synthesis built on it stresses that nonadherence is rarely a single deliberate choice but the product of many small frictions, with cost and regimen complexity among the most tractable (Brown & Bussell, 2011). A review of systematic reviews that sorted hundreds of reported determinants concluded that no small set dominates, that the evidence is fragmented, and that the same factor can push adherence up in one setting and down in another (Kardas et al., 2013).

Against this sprawl, the most influential psychological account narrows the focus to beliefs. The Necessity-Concerns Framework proposes that a patient's adherence is driven by the balance between their belief that the medicine is necessary for their health and their concerns about taking it — fears of dependence, side effects, or long-term harm. A meta-analytic review across many long-term conditions found the expected pattern consistently: stronger necessity beliefs were associated with better adherence, stronger concerns with worse adherence, and the difference between the two predicted adherence better than either alone (Horne et al., 2013). This reframes nonadherence as frequently intentional and reasoned — a patient who doubts the medicine's necessity or fears its harms is behaving sensibly given those beliefs, however clinically mistaken the beliefs may be. The second demonstration animates this balance across the four belief profiles it generates. A qualitative scoping review of patients' own accounts reached a compatible conclusion, finding that adherence in chronic conditions rests on patients' understanding of the illness, their relationship with providers, and the practical and emotional fit of the regimen into a life (Kvarnström et al., 2021).

Set how necessary a patient believes their medicine is (N) and how concerned they are about taking it (C). The model predicts adherence from the differential alone: p = 0.5 + 0.04·(N − C), bounded to 0.10–0.90. Each point by which necessity outweighs concern adds four percentage points.

profile: Accepting — differential D = +7, predicted adherence 0.78

The two mixed profiles make the framework's point: an ambivalent patient (high N, high C) and an indifferent one (low N, low C) both land near the coin-flip, because the model reads the difference, not the levels. The differential predicts adherence better than either belief alone (Horne et al., 2013). Illustrative model, not fitted coefficients.

Consequences

Nonadherence matters because its downstream effects are large and measurable. In cardiovascular disease, where the medicines are effective and the stakes high, better adherence to therapies such as statins and antihypertensives is associated with fewer events and lower mortality, and poor adherence is a recognized contributor to avoidable hospitalization and death (Ho et al., 2009). Because the conditions most dependent on long-term medication are also the most prevalent, small average shortfalls in adherence aggregate into a substantial population burden of uncontrolled disease.

The economic consequences are correspondingly large. A systematic review of the cost of nonadherence across disease groups found that it imposes a heavy and largely avoidable burden on health systems, driven by the excess hospitalizations, procedures, and disease progression that follow from treatments not taken (Cutler et al., 2018). A clinical commentary framed the persistence of this burden as an unmet challenge: despite decades of attention, nonadherence remains common, costly, and under-addressed in routine care, in part because it is invisible unless deliberately looked for (Kleinsinger, 2018). The combination of clinical harm and financial cost is what justifies the large research effort devoted to intervention.

Interventions and Their Limits

If nonadherence is this consequential, the natural question is whether it can be fixed, and here the evidence is sobering. The landmark Cochrane systematic review of interventions for enhancing medication adherence concluded that even the most effective interventions were complex and labor-intensive and produced only modest improvements, that the methods of the trials were often weak, and that no single approach could be recommended across the board (Nieuwlaat et al., 2014). A later meta-analysis synthesizing 771 adherence-intervention trials quantified the same message: the average effect of interventions on adherence is small, though it varies by strategy, with approaches that build habit and target behavior outperforming those that simply deliver information (Conn & Ruppar, 2017).

Two lessons follow. The first is that information is rarely the bottleneck — because nonadherence is weakly tied to knowledge, education alone moves it little, and the interventions that work tend to reduce friction, build routine, or address the beliefs and barriers specific to the patient. The second is that a small average effect can still be worthwhile at scale when the outcome is as costly as uncontrolled chronic disease, provided the intervention is cheap enough and targeted at those who need it. The third demonstration models how a baseline adherence level and a realistically modest intervention effect combine, and why even a successful program leaves a large residue of nonadherence.

Set a baseline adherence level and the size of an intervention's effect in percentage points. The bars show adherence before and after; the shaded remainder of the “After” bar is the nonadherence that still remains. The default +5 points reflects the small average effect the trial evidence reports.

adherence rises from 55% to 60%, leaving 40% nonadherence unaddressed

Across 771 trials the average adherence intervention produced only a small improvement, with habit- and behavior-targeting approaches outperforming information alone (Conn & Ruppar, 2017); the landmark Cochrane review reached the same sobering conclusion (Nieuwlaat et al., 2014). A modest effect can still be worthwhile at the scale of chronic disease, but it rarely closes the gap. Illustrative values.

Worked Example

The Necessity-Concerns Framework can be made quantitative with a simple illustrative model. Suppose a patient rates, on scales from 0 to 10, how necessary they believe their medicine is (N) and how concerned they are about taking it (C). The framework's central claim is that adherence tracks the differential D = N − C — the degree to which perceived need outweighs worry. Model the predicted probability of adequate adherence as a linear function of that differential, p = 0.5 + 0.04·D, bounded to the range 0.10 to 0.90. A patient in perfect balance (D = 0) is modeled as a coin-flip adherer; each point by which necessity exceeds concern adds four percentage points, and each point the other way subtracts them.

The framework also generates four belief profiles from the two beliefs taken separately, and the model assigns each a characteristic adherence. An accepting patient (high necessity, low concern) sits at a large positive differential and high predicted adherence; a skeptical patient (low necessity, high concern) sits at a large negative differential and low adherence. The two mixed profiles are the interesting ones: an ambivalent patient holds both strong necessity and strong concern, and an indifferent patient holds both weakly, yet the model places both near the coin-flip, because what it reads is the difference, not the levels. Table 2 works the four profiles through.

The arithmetic makes the framework's clinical implication concrete. The accepting and skeptical patients are the easy cases — one will adhere, one will not, and the reasons are legible. The ambivalent patient is the one an intervention can most change: because both beliefs are strong, lowering the concern (addressing a specific fear about side effects, say) moves the differential sharply without touching the necessity belief, lifting predicted adherence from the coin-flip toward the accepting range. Pouring more necessity information onto an ambivalent patient, by contrast, may do nothing, because their problem is not that they doubt they need the drug but that they also fear it. This is why the meta-analytic finding that the differential predicts adherence is not a statistical nicety: it tells the clinician which belief to work on (Horne et al., 2013).

Discussion

The history of medication adherence research is, in large part, the story of abandoning a wrong model of the patient. The compliance framing treated drug-taking as obedience and nonadherence as deviance, which both insulted the patient and, more damagingly for science, predicted behavior poorly, because it looked for the cause in the patient's character rather than in the patient's reasoning and circumstances. The reframing toward adherence and concordance was not merely a change of manners; it relocated the explanatory burden onto beliefs, barriers, and the quality of the therapeutic relationship, where the evidence actually lies (Chakrabarti, 2014) (DiMatteo, 2004).

What the field has not achieved is a correspondingly powerful remedy. The gap between the clarity of the diagnosis and the modesty of the treatments is the central tension of the literature: nonadherence is pervasive, costly, and reasonably well understood, yet the best interventions move it only a little (Nieuwlaat et al., 2014) (Conn & Ruppar, 2017). The most defensible reading is that there is no single lever because there is no single cause — adherence is the net of many small forces, and an intervention that pulls on one leaves the others in place. That argues for tailoring: measuring the specific barrier a given patient faces and addressing it, rather than applying a uniform program, which is the direction both the belief models and the intervention meta-analyses point.

Cognitive Implications

Medication adherence is a textbook case of the intention-behavior gap and a problem of self-regulation rather than of knowledge. A patient may fully intend to take a medicine, understand why it matters, and still fail to enact the intention across the hundreds of occasions a chronic regimen demands, because enactment depends on memory, habit, cue, and the moment-to-moment weighing of effort against benefit. The finding that nonadherence is only weakly correlated with patient knowledge is the empirical signature of this gap: closing the information deficit does not close the behavior deficit, because the deficit was never mainly informational (DiMatteo, 2004).

Two cognitive models carry most of the explanatory load. The Necessity-Concerns Framework treats adherence as the output of a belief calculation, and its success in predicting behavior across conditions shows that drug-taking is reasoned from the patient's representation of their situation, not dictated by the prescription (Horne et al., 2013). The Common-Sense Model of self-regulation goes further, proposing that patients build a lay mental representation of their illness — its identity, cause, timeline, consequences, and controllability — and that adherence is the behavioral output of coping with the illness as that representation defines it; a patient whose common-sense model says the illness is acute and symptom-defined will stop a preventive medicine the moment symptoms resolve, exactly as the model predicts and exactly as adherence data show (Leventhal et al., 1992). In both models the medicine is taken or not according to an internal account of the illness, which is why interventions that change the account outperform those that merely repeat the instruction.

Current Directions

The most active front is digital measurement and support. Electronic monitoring, smart pill bottles, ingestion sensors, and smartphone applications promise to replace the crude proxies of refill data and self-report with a near-continuous record of dosing, and to intervene in real time with reminders and feedback. The evidence that these tools capture dosing more accurately is strong; the evidence that they improve clinical outcomes is still thin, and the qualitative literature cautions that technology helps only when it fits the patient's life and beliefs rather than adding another demand (Kvarnström et al., 2021). The phase taxonomy has sharpened this work by letting monitoring distinguish a persistence problem (the patient stopped) from an implementation problem (the patient takes doses erratically), which call for different responses (Vrijens et al., 2012).

A second direction is economic and systemic. As health systems absorb the measured cost of nonadherence, adherence has become a target of value-based care, with interventions judged increasingly on cost-effectiveness rather than efficacy alone (Cutler et al., 2018). Because the intervention literature shows small average effects, the economic case depends on cheap, scalable, and well-targeted programs — simplified regimens, fixed-dose combinations, synchronized refills, and belief-focused counseling directed at the patients most likely to benefit. The convergence of the digital and economic trends points toward a model of adherence support that is measured continuously, tailored to the individual barrier, and deployed selectively rather than universally (Conn & Ruppar, 2017).

Common Misconceptions

Patients who do not take their medicine simply do not understand why they should.

Nonadherence is only weakly related to patient knowledge; much of it is intentional and reasoned, driven by doubts about the medicine's necessity and concerns about its harms rather than by an information deficit (Horne et al., 2013).

Nonadherence is a minor problem affecting a small minority of difficult patients.

A quantitative review across fifty years and many conditions puts average nonadherence near one patient in four, making it a pervasive feature of treatment rather than the property of a few outliers (DiMatteo, 2004).

If we just had the right program, we could largely eliminate nonadherence.

Even the most effective interventions are complex and produce only modest improvements, and a meta-analysis of 771 trials found small average effects, because adherence has many causes and no single lever moves it far (Nieuwlaat et al., 2014).

A patient who collected the prescription is taking the medicine.

Every adherence measure is a proxy: a refilled prescription is not a swallowed pill, pill counts can be gamed, and self-report is systematically inflated, so a collected prescription establishes supply, not ingestion (Lam & Fresco, 2015).

Glossary

ABC taxonomy.

A standardized vocabulary that divides adherence into initiation, implementation, and discontinuation, with persistence defined as the time from initiation to discontinuation.

Adherence.

The extent to which a patient's medication-taking corresponds to the regimen agreed with the prescriber, framing the patient as an active decision-maker.

Common-Sense Model.

Leventhal's self-regulation model in which adherence is the behavioral output of a patient's lay mental representation of their illness and its treatment.

Compliance.

The older term for following medical instructions, largely superseded because it framed the patient as an obedient rule-follower and predicted behavior poorly.

Concerns beliefs.

A patient's worries about a prescribed medicine — side effects, dependence, long-term harm — that lower adherence in the Necessity-Concerns Framework.

Concordance.

A framing in which the standard for drug-taking is a negotiated agreement between prescriber and patient rather than the prescriber's orders alone.

Directly Observed Therapy.

An adherence-support strategy in which an observer watches the patient take each dose, the single MeSH child of medication adherence.

Discontinuation.

The point in the ABC taxonomy at which a patient stops taking the medicine; the end of the persistence period.

Implementation.

The phase describing how closely a patient's actual day-to-day dosing matches the prescribed schedule between initiation and discontinuation.

Initiation.

The phase marking whether and when a patient takes the first dose of a newly prescribed regimen.

Medication possession ratio (MPR).

A refill-based adherence measure expressing the supply of medication obtained as a fraction of the days in a period.

Necessity beliefs.

A patient's conviction that a prescribed medicine is needed for their health, which raises adherence in the Necessity-Concerns Framework.

Necessity-Concerns Framework.

Horne's model in which adherence tracks the balance between belief in a medicine's necessity and concern about taking it.

Nonadherence.

Medication-taking that falls short of the agreed regimen, whether by not starting, dosing erratically, or stopping early; may be intentional or unintentional.

Persistence.

The length of time a patient continues a regimen from the first dose until discontinuation.

Proportion of days covered (PDC).

A refill-based adherence measure, generally preferred to the MPR, giving the share of days in a period on which the patient had medication available.

Social-desirability bias.

The tendency of patients to report more adherence than they achieve, which makes self-report a systematically inflated measure.

Key Researchers

Vicki S. Conn

(University of Missouri, Sinclair School of Nursing). Her meta-analyses synthesized hundreds of adherence-intervention trials, quantifying how small the average intervention effect is and which strategies outperform. ORCID · Scholar · Wikidata

M. Robin DiMatteo

(University of California, Riverside). Her quantitative review of fifty years of adherence research established the field's baseline nonadherence estimate and the roles of disease, regimen, and the patient-physician relationship. Personal

R. Brian Haynes

(McMaster University). A founder of evidence-based medicine and senior author of the Cochrane reviews on interventions for enhancing adherence, which established how modest and effortful the effective interventions are. ORCID · Faculty · Wikipedia

Robert Horne

(UCL School of Pharmacy, University College London). Developed the Necessity-Concerns Framework and the Beliefs about Medicines Questionnaire, recasting nonadherence as a reasoned weighing of perceived need against concern. ORCID · Faculty · Wikipedia

Howard Leventhal

(Rutgers, The State University of New Jersey). Developed the Common-Sense Model of self-regulation, which frames adherence as the behavioral output of a patient's lay mental representation of illness and treatment. Faculty

Bernard Vrijens

(AARDEX Group; University of Liège). Led the ABC taxonomy that standardized adherence terminology into initiation, implementation, and persistence. Scholar · Wikidata

Frequently Asked Questions

What is medication adherence?

Medication adherence is the extent to which a patient takes a medicine (the right drug, dose, timing, and duration) in line with the regimen agreed with their prescriber. It is distinguished from the older term compliance by treating the patient as an active decision-maker rather than an obedient follower of orders (Chakrabarti, 2014).

How common is nonadherence?

A quantitative review pooling fifty years of research across many conditions found that on average about one patient in four does not adhere to medical recommendations, with adherence worse for longer regimens and for prevention than for cure (DiMatteo, 2004).

What is the difference between compliance, adherence, and concordance?

Compliance frames drug-taking as obedience to the prescriber; adherence keeps the regimen as the reference but treats the patient as an active decision-maker; concordance locates the standard in a negotiated agreement between prescriber and patient rather than in the prescriber's orders (Chakrabarti, 2014).

Why do patients stop taking medicines they need?

Much nonadherence is intentional and reasoned: patients weigh their belief that a medicine is necessary against concerns about side effects, dependence, or long-term harm, and when concern outweighs perceived need they take less, whatever the clinical reality (Horne et al., 2013).

Is nonadherence mostly a knowledge problem?

No. Nonadherence is only weakly related to how much patients know, which is why education alone changes it little; it is better understood as an intention-behavior and self-regulation problem shaped by beliefs, habits, and barriers (DiMatteo, 2004).

How is adherence measured?

By direct methods such as observed ingestion or drug assays, and indirect methods such as pill counts, pharmacy refill records, electronic monitoring, and self-report. Each is a proxy with its own weaknesses, and self-report in particular is systematically inflated (Lam & Fresco, 2015).

Do interventions to improve adherence work?

They work modestly. The Cochrane review found even the best interventions complex and only modestly effective, and a meta-analysis of 771 trials found small average effects, with habit-building and behavioral strategies outperforming information alone (Nieuwlaat et al., 2014) (Conn & Ruppar, 2017).

Why does nonadherence matter beyond the individual patient?

Because its effects aggregate: poor adherence worsens outcomes in high-stakes conditions such as cardiovascular disease and imposes a large, largely avoidable economic burden on health systems through excess hospitalization and disease progression (Ho et al., 2009) (Cutler et al., 2018).

References