The odds of the three nons when an aptly prescribed medicine isn't working: non-compliance, non-absorption, non-response


  • Presented at the 3rd Meeting on Clinical Measurement and New Drug Development, Royal Society of Edinburgh, 13–14 June 2001.

Dr John Urquhart, 975 Hamilton Ave, Palo Alto, CA 94301 USA. Tel.: +1-650-321-3961; Fax: +1-650-324-9739; E-mail:


When an aptly prescribed medicine has not worked as expected, the potential explanations logically fall among three categories: pharmacodynamic, pharmacokinetic, and pharmionic. Pharmacodynamic reasons include problems in drug reception created by deficient or absent receptors, when disease or comorbidities are too severe for the agent's usual therapeutic benefits to express themselves, when unusually strong physiological counter-regulatory actions attenuate or nullify the drug's usual actions, and others – obviously a whole topic in its own right. Pharmacokinetic reasons include problems with drug absorption or unusually high rates of presystemic or systemic drug metabolism, or of other clearance mechanisms, resulting in unusually low concentrations of drug in plasma and other body fluids arising from the usually satisfactory, recommended dosage regimen. Pharmionic reasons include failure of the medicine to be taken in the doses, at the times, and in keeping with other dosing instructions that are needed, ceteris paribus, for satisfactory therapeutic action.


‘Pharmionic’ is the adjectival form of the name of a nascent discipline, pharmionics, which is concerned with factors that measure and control the use of prescription pharmaceuticals. The suffix ‘ionic’ comes from the Greek verb ‘to go’[1]. Pharmionics is akin to ‘avionics’, which is the discipline and technologies concerned with the measuring and controlling of flight in aircraft – ‘the going’ of the airplane.

Avionics began and grew in importance as advances in aircraft pushed speed and acceleration past the point that pilots could safely rely on direct vision, a compass, an altimeter, and ‘seat of the pants’ sensations for controlling flight. Today's means of measurement, communication, and control have superseded the simple approaches that can suffice when flight is slow, low, and in ideal weather. Analogously, one can reasonably say that pharmacotherapeutics has reached or passed the point where the power and complexity of medicines require formalization into disciplinary terms of the methods of measurement, communication, and control needed to improve the likelihood, if not assure, that powerful medicines are administered in appropriate doses, at appropriate times, under appropriate circumstances. These are the topics that comprise the subject matter of pharmionics.

The etymology of ‘pharmionics’ indicates its basic concern with ‘the going of the prescription drug – ‘going’ in the sense that a manufactured dosage form must, if it is to play its therapeutic role, travel from the package in which it is dispensed by a pharmacy into the patient's mouth, and down the oesophagus to the stomach. At that point, the discipline of pharmacokinetics takes centre-stage, dealing with subsequent conversion of drug into metabolites and movement into the bloodstream, toward receptors or other sites of action, which fall under the discipline of pharmacodynamics. The distances that the drug travels under the purview of pharmionics are a few metres more than the meter or so that drug travels under the purview of pharmacokinetics. One can envision a further spatial compression from metres to the angstroms involved in the receptor docking/undocking that could reasonably be considered the domain of pharmacodynamics, beyond which receptor-generated message(s), but usually not the drug molecules themselves, may travel locally or in some cases systemically. Undocked drug molecules return to the domain of pharmacokinetics, for metabolism and/or excretion.

In terms of what can go wrong in ambulatory pharmacotherapy, history shows that things can go wrong at every point along the medicine's travels from pharmaceutical factory to drug receptor. Most errors occur, however, after the dispensing of the prescription, when the patient becomes responsible for the penultimate steps in the whole chain from factory to receptor. Errors made at any point in, or segment of, those travels can nullify or otherwise pervert therapeutic intentions of manufacturers, prescribers, and pharmacists. This view was nicely codified by Carl Peck and the late John Harter in a valuable paper in 1991 [2], recently fine-tuned [3]. But given that errors at any point can nullify, thwart, or confound therapeutic intent, it might seem odd to a visitor from outside the pharmaceutical-medical world that attention is so markedly focused on just the pharmacokinetic phase of the overall process. This imbalance reflects probably several factors. One, of course, has been the phenomenal advances in analytical chemical methods during the past 3 decades, which have greatly facilitated studies of the disposition of ingested or otherwise administered drug within the body – ‘what the body does to the drug’. In contrast, pharmacodynamics – ‘what the drug does to the body’ – has been vexed by the methodologic difficulties of reliably quantifying the manifold types of drug action, many of which cannot be expressed in the numerical terms that most scientists have been trained to seek, but which must instead be quantified in clinimetric terms that relatively few scientists have been educated to develop and apply to clinically pressing problems.

Against the foregoing two backdrops, the nascent discipline of pharmionics could hardly begin to be contemplated until the integration of time-stamping microcircuitry into pharmaceutical packages made it possible to compile reliable data on ambulatory patients' dosing histories [4–7]. That development, which by now has compiled an extensive range of published applications [8], represents the ‘technological push’ side of the story, as important to pharmionics as the advances in analytical chemistry since 1960 have been to pharmacokinetics. The medical need, or ‘pull’ for pharmionics, arises from the increasing power and specificity of drug actions, reflected by the increasing use of patients' responses to certain prescribed drugs as bases for medical decision-making. This aspect of medical practice is exemplified by, but certainly not limited to, the ‘stepped-care’ approach to treating arterial hypertension [9]. Stepped-care management includes the concept of first-, second-, and third-line agents. In general, a patient's failure to respond to treatment with a first-line agent is an indication to escalate the strength of the prescribed treatment. Escalation can variously be done by raising the dose of the already-prescribed agent(s), by adding a further first-line agent, by combining a first- and second-line agent, or by switching completely to a second-line agent. Persistent nonresponse is interpreted as an indication to proceed toward so-called ‘triple therapy’, i.e. three concomitantly prescribed drugs drawn mainly from 2nd and 3rd line agents. Similar ideas that nonresponse should drive progression through 1st, 2nd, and 3rd line agents toward a 3-drug ‘ultimate’ regimen is found in epilepsy, glaucoma, and other chronic diseases for which numerous choices of drug are available. Lip-service is, of course, given to the possibility that nonresponse may be consequent to poor compliance, but that sensible advice is nullified by the fact that, in the absence of electronic monitoring, the clinical ascertainment of patient compliance is aptly described as no better than a coin-toss [10].

Pre-electronic studies

Views about the impact and pervasiveness of poor compliance with prescribed or, in trials, protocol-specified drug regimens did not have to await the development of electronic monitoring methods. Four seminal studies presaged the growing importance of the topic with usually unreliable methods that were pushed by extraordinary efforts to a state of what might be called ‘marginal reliability’. Jointly considered, these studies give a foretaste of the explanatory power of reliable data on ambulatory patients' actual exposure to prescription drugs [11–14]. These studies are reviewed in [15].

With the advent, first of low-dose, slow turnover chemical markers [16–18], and then of electronic monitoring methods, the methodologic foundation has been laid for pharmionics to flourish. Its logic rests on the fact that, as the disciplinary home for reliable measurement and interpretation of patient compliance with prescribed drug regimens, pharmionics can claim to concern itself with one of the two biggest single sources of variability in drug response, the other being pharmacokinetics [2].

Therein, of course, lies the basis of my title ‘the odds of the three nons’. I use the lay sense of the term ‘odds’, to mean the probabilities that a disappointing response to a rationally prescribed drug will find its explanation as a problem in pharmionics, in pharmacokinetics, in pharmacodynamics, or in some combination thereof.

Pharmacological power – the essential element

An essential feature of the story is that prescription drugs now provide greater, more specifically focused interventional power than ever before. This evolution has occurred gradually, as a logical outcome of the highly competitive, economically attractive marketplace in which pharmaceuticals compete for share. Among all the varied attributes of a pharmaceutical product, therapeutic efficacy is naturally a principal focus in the quest for competitive advantage. The imposition, four decades ago, of the need to prove efficacy as a condition for product registration has assured that each pharmaceutical product enters the marketplace with an explicit claim of efficacy. Efficacy claims of course serve many purposes, one of which is to be a competitive target for future products. Efficacy can be proven in a few hundred or a few thousand patients in a circumscribed period of time. In contrast, the other principal focus of pharmaceutical development – defining the risk of harm that patients incur in using the product – is an open-ended, methodologically uncertain process involving several orders of magnitude more patients and much longer periods of observation than are involved in the proof of effectiveness.

With progressively more power and specifically acting pharmaceuticals have come new understanding of disease, facilitation of previously impossible surgical procedures, previously impossible (and unthinkable) preventive manoeuvres, reliable family planning, reduced mortality rates at all ages of life (although the death rate ultimately remains at one per person), and improved management of many diseases. Two natural concomitants of these benefits are increased usage and increased prices of prescription drugs, as natural extensions of the basic economic principle – applicable both inside and outside the healthcare arena – that effective products command wider usage and higher prices than ineffective products.

The steadily increasing power and pricing of pharmaceuticals are, together with the dose and time dependencies of drug actions, the four factors that have jointly brought patient compliance with prescribed drug regimens from a trivial to a key role in ambulatory care.

In effect, then, the need for the discipline of pharmionics is a de facto tribute to the success of medicinal chemistry, pharmaceutical development, and pharmacologic and pharmaceutical science since the origins of these disciplines roughly a century ago.

Perforce …

A natural consequence of the present role of prescription drugs is that, when aptly prescribed, correctly taken, and continued for a sufficiently long period of time, they are more likely than not to have beneficial medical consequences. How much more likely depends upon the dynamics of the underlying disease and for how long the treatment persists. The proof of efficacy carries the corollary that prescribing the drug increases the odds of a good outcome, but there is a set of provisos. The main provisos are that the patient: (a) is properly diagnosed (b) rationally treated (c) complies satisfactorily with the prescribed drug regimen, and (d) persists with the treatment regimen for an adequately long period of time. Proviso (b) includes a regimen that assures adequate absorption and effective concentrations of drug at sites of action, and satisfactory reception of drug. If these provisos are not met, the outcome may still be good if the disease is one from which patients recover without treatment. If not, however, then each of the provisos approaches the status of being a conditio sine qua non for a good outcome.

In contrast, when prescribing is inept, or otherwise inappropriate, or when a properly prescribed drug regimen is ineptly executed (i.e. poorly complied with), or if persistence with the regimen is too brief, patients are likely to have bad outcomes, which may be further complicated by hazards arising not only from failure to treat in an effective and timely way, but also from the holiday pattern of drug exposure, which can trigger hazardous rebound effects or recurrent first-dose effects [19–23].

Thus, the achievement of a good or poor outcome of treatment correspondingly tilts the retrospectively assessed odds of the status of the various provisos. Good outcomes are not likely in patients who have complied poorly or who have persisted only a short while with their prescribed treatment. Conversely, patients who have had a good outcome are likely to have complied satisfactorily and persisted with prescribed treatment for an adequate period of time. These outcome-dependent probabilities can be considered as therapeutically pertinent biases when patients are selected for some new therapeutic adventure based on the outcomes they have had from prior treatment. When recognized as such, these biases may be used to advantage when, e.g. one seeks to assemble a cohort of unusually uniformly compliant and persistent patients. Conversely, they may complicate a trial designed to show that a new medicine can satisfactorily treat patients who had failed to respond to a competing agent. If a substantial fraction of such patients failed to respond to a prior treatment because of inadequate compliance with the prescribed drugs, then it is likely that they will carry their poor compliance over into the new trial or treatment program. If their prevailing level of compliance is too poor for the new medicine to work, then it, too, will fail, as will any medicine, however, powerful.

These tendencies to select for unusually good or unusually poor compliance need to be seen in the context of the usual patterns of good, partial, and poor compliance found in medically unselected patients, exemplified by patients who have only recently embarked on a first course of primary preventive treatment, without special attention having been paid to their compliance with the prescribed regimen(s). The dosing patterns observed in such patients by unobtrusive electronic monitoring, without any attempt to feed results back to patients or otherwise influence their dosing patterns, is approximately as follows [24]:

  • One-sixth of the patients execute the regimen with strict punctuality.

  • One-sixth of the patients take virtually all the prescribed doses but with some fluctuations in dose-timing.

  • One-sixth of the patients occasionally omit a single day's doses, with fluctuations in dose-timing.

  • One-sixth of the patients have a drug holiday (the sequential omission of 3 or more days' doses) 3–4 times per year, together with occasional omissions of 1–2 days' doses.

  • One-sixth of the patients have a drug holiday monthly or more often, together with frequent omissions of 1–2 day's doses.

  • One-sixth of the patients take few or no doses, while maintaining the appearance of satisfactory, if not perfect, compliance.

These categories are, of course, approximations, but this ‘rule of sixes’ is a useful approximation of what to expect of drug exposure when dosing histories are compiled by electronic monitoring methods in medically unselected patients. These approximate the dosing patterns that, e.g. underlie the 50-fold higher conception rates characteristic of ‘typical’ use, compared to ‘ideal’ use of oestrogen-progestin combined oral contraceptives [25] and the fact that only about 27% of treated hypertensive patients in the US achieve satisfactory control of their blood pressures [9].

Selection for unusually good compliance – two examples

Cardiac transplant recipients

An example of selection for unusually compliant patients is provided by the work of de Geest and her colleagues [26], who measured compliance in 105 patients who had successfully undergone a cardiac transplant. She found virtually flawless, punctual compliance in over 90% of the patients.

These findings contrast starkly with the ‘rule of sixes’ described above, in medically unselected patients, in which only one in about six patients are punctually compliant. The patients studied by de Geest et al.[26] had been selected initially as candidates for cardiac transplantation on the basis of their having unequivocally severe congestive heart failure. They then had to wait in a queue for some months before an immunologically compatible heart became available. Severe heart failure is a medically precarious condition, with a naturally downhill trajectory and high mortality rate, in which suboptimal treatment can result in accelerated deterioration and death. One would naturally like to see complete data on both prescribing and the patients' compliance with prescribed drug regimens through the entire pretransplant period of an inception cohort of patients selected as candidates for transplant. Unfortunately, all we have at the present are data on the compliance patterns of patients post-transplant.

In any case, the exceptionally high prevalence of punctual compliance naturally poses the question: did the exceptional nature of their medical condition cause many patients to convert to punctual compliance, or, alternatively, did most of those whose compliance was suboptimal fall by the wayside during their generally long wait in the queue, either dying or becoming too sick to be an acceptable surgical risk? Certainly the latter case is medically plausible, but it remains for future studies to show which is the more apt description of why heart transplant recipients are an exceptionally compliant group.

Long-term use of drugs with holiday-induced hazard

Petri and I [27] monitored compliance prospectively in patients of a group of Dutch general practitioners who had prescribed, through the pharmacy associated with the group, β-adrenergic receptor antagonists as first-line treatment for their hypertension. Most of the agents used were of the category that lack intrinsic sympathomimetic activity. Excluded from the study were patients whose hypertension was complicated, as evidenced by review of their medical records and of their prescription dispensing histories. The only chronic-use concomitant medication the patients could receive, and still be included in the study, was a thiazide diuretic. Thus any patient who had previously developed angina pectoris, congestive heart failure, or a myocardial infarction was excluded from the study. Enrolled patients were, after informed consent based on a brief explanation that the purpose of the special package was to keep a count of their doses, given their β-adrenergic receptor antagonist in an electronically monitored package (MEMS®, APREX). An exceptionally high prevalence of excellent compliance was found in this group of patients, raising the question of how this study differed from the usual wide diversity of compliance with prescribed antihypertensive drugs [24, 28–45].

Possible explanations, of course, were (a) that these patients and their physicians enjoyed a uniquely close and harmonious relation that naturally fostered good compliance with prescribed drug regimens, or (b) something about the informed consent procedure changed their usual compliance patterns. A third explanation was suggested by the finding, which had not been evident when the study was planned, that, several years prior to the study, this group of physicians had ceased using β-adrenergic receptor antagonists as first-line treatment for hypertension, and had switched to other agents, mainly angiotensin converting enzyme antagonists. Thus, the patients included in this study had been started on their present treatment prior to that change in prescribing policy, and thus were multiyear veterans of treatment with β-adrenergic receptor antagonists. This finding raised the question of whether partially compliant patients, prone to multiday lapses in dosing (drug ‘holidays’), who would have been repeatedly exposed to hazardous rebound effects, were gradually selected out of the inception cohort by incident coronary artery disease. There is clear evidence that sudden interruptions in dosing with non-ISA β-adrenergic receptor antagonists trigger hazardous rebound effects that include incident coronary artery disease [19–23]. This hazard is usually discussed in relation to physician-directed cessation of treatment, but it must also be so that the recurrence of drug holidays among about a third of medically unselected patients, who are holiday-prone, constitutes a recurring gauntlet of elevated risk of incident coronary disease. In the course of time, these serial episodes of elevated risk could be expected to take their toll, which, in terms of the study done by Petri and me meant the occurrence of incident coronary disease, and selection out of the group of drug recipients who had uncomplicated hypertension.

Of course, β-adrenergic receptor antagonists are prescribed to reduce the overall risk of incident coronary artery disease, with evidence from large trials of an aggregate net benefit [9]. The finding of net benefit in the aggregate is not inconsistent with the same treatment's adding hazard in a minority of patients whose dosing patterns expose them to recurrent episodes of elevated risk, created by recurrent rebound effects that persist for a few days to a week or more, depending on the length of drug holidays and the dynamics of the rebound phenomenon. Safety considerations such as this are one of the reasons why reliably measured compliance with protocol-specified drug regimens in randomized, placebo-controlled trials should be used to develop covariate information, even if intention-to-treat analysis, which excludes such consideration, is used as a primary analysis of efficacy [46–51]. The need for reliable measurement of compliance in active-controlled trials is a story that I have told elsewhere [52].

To avoid such uncertainty, one would obviously like to be able to see the entire story unfold in an inception cohort of patients, starting from the first diagnosis of hypertension, through the decision to start treatment, and onwards, informed by reliable data on compliance with prescribed antihypertensive agents, as patients respond variably, whilst continuing, changing, or discontinuing treatment.

An example of selection for unusually poor compliance

Recently, Burnier, Brunner, and their colleagues in Lausanne reported that, among the patients referred to their university hospital clinic for diagnostic evaluation of persistently drug-resistant hypertension, slightly over half proved to be clinically unrecognized noncompliers [45]. Almost all of the patients studied had been escalated to ‘triple therapy’ in the face of persistent hypertension. All had a record of consistent nonresponse to any prescribed antihypertensive agents. When seen at the university hospital clinic, the patients were informed that, before the usual diagnostic tests were performed, they would first undergo a new diagnostic test, which was a 60 day period of electronic monitoring of their daily dosing with their three prescribed antihypertensive drugs. Then that record would be reviewed with them, and would be the basis on which the decision would be taken to continue with other, more costly tests, which included a renal angiogram, a search for a pheochromocytoma, and an assessment of intracranial pressure. They were then presented with three electronic monitors (eDEM® Monitors, AARDEX Ltd) each containing one of their three prescribed antihypertensive agents, On the grounds of Cabot's Dictum, which is a Bostonian rephrasing of Occam's Razor (‘simplest explanations are best’), one could reasonably conclude that renal angiograms and other complex, costly diagnostic tests are not indicated when persistent drug regimen noncompliance is positively identified, which is uniquely possible, as Burnier et al.[45] pointed out, by electronic monitoring.

Indeed, Burnier et al.[45] were able to get about two-thirds of their clinically unrecognized noncompliers to execute their prescribed regimens well enough to bring their blood pressures, for the first time, into satisfactory control. The improvements occurred in two ways. Some of the patients, on grasping the importance which the university hospital experts ascribed to the correct taking of the prescribed medicines, spontaneously corrected their compliance, and returned 60 days later with a record of good compliance and, for the first time, satisfactorily controlled blood pressures, although several had difficulties with postural hypotension, as triple therapy was evidently rather more treatment than they needed, given satisfactory compliance. Another subgroup of patients returned with a record of poor compliance and persistent hypertension, about half of whom were able to improve their compliance, after a review of their dosing chronologies, together with examples of what a proper dosing chronology should look like. There was a further subgroup, amounting to just under half of the entire group of patients, who returned with a record of good compliance with persistent hypertension – true drug-resistant hypertension. There remained a subgroup, which amounted to about 1 in 5 of the clinically unrecognized noncompliers, who were unable or unwilling to improve their compliance, and so remained hypertensive. While this last subgroup is a challenge for future research, the overall results represent, if confirmed elsewhere, an important advance in the management of drug-resistant hypertension. It gives the promise of eliminating about half of the diagnostic tests run for drug-refractory hypertension, brings an appreciable fraction of otherwise troublesome patients into satisfactory control of their blood pressures, and allows some excessive prescribing to be eliminated. Test-avoidance and reduction in the number or amount of prescribed drugs can offset the costs of the electronic monitors, which presently run at circa $80 for a monitor that has a 3 year lifetime. One or two of the monitors bought for the initial evaluation can, of course, be used for ongoing monitoring to watch for evidence of backsliding among patients who have been newly converted to good compliance.

It is noteworthy that the Lausanne group are not alone in having achieved satisfactory compliance in a substantial majority of patients with a major medical need for correct drug use. Cramer & Rosenheck, at Yale, have reported correction of compliance in a similar proportion of alcoholic psychotics whose initial compliance had them omitting roughly half their prescribed doses of antipsychotic medication; they were able to achieve satisfactory compliance in about 80% of a group of such patients [54]. While it is beyond the scope of this paper to dwell on techniques for improving compliance, the two key elements in both studies have been the objective, continuous record of day by day dosing times, which can be easily interpreted by most patients, and the fact that the monitoring is ongoing, with continuing periodic reviews with the prescribing physician, a pharmacist, or a nurse. The term for this new approach is Measurement-Guided Medication Management (MGMM), and is a clear break with past attempts to improve compliance based on unsatisfactory methods of measurement that afforded patients the easy ability to exaggerate their compliance. In contrast, electronic monitoring makes it a very demanding task to create a false record of good compliance: the immutable nature of the electronic record requires the patient to execute the requisite maneuvers with the drug package at each scheduled dosing time in order to create a false record of good compliance. While not impossible, it requires the intersection of two relatively infrequent attributes: (a) strict punctuality and (b) sustained malevolence. Much of course remains to be done to confirm these promising early findings.


Figure 1 sketches the changes in the proportion of unsatisfactorily compliant patients as one goes from the medically unselected, to those selected on the basis of a successful outcome of prior treatment with prescription drugs, and those selected on the basis of an unsuccessful outcome of prior treatment. Figures 2 and 3 show how a group of medically unselected patients, initially divided between roughly 1/3rd and 2/3rds having, respectively, compliance too poor and good-enough for a satisfactory outcome of treatment, would apportion themselves when categorized as having had a satisfactory (Figure 2) or unsatisfactory (Figure 3) outcome of treatment.

Figure 1.

Schematic of how the proportions of satisfactory and unsatisfactory compliers shift from the roughly two-thirds/one-third in medically unselected patients, to a very high proportion of good compliers found among patients who have not only survived but maintained their status as an acceptable surgical risk during a long period of waiting for a cardiac transplant in the face of severe congestive heart failure. The other extreme is represented by patients who have persistently failed to respond to escalating doses of antihypertensive medicines. Both situations are discussed in the text.

Figure 2.

Schematic of how the numbers of satisfactory compliers change, according to the success or failure of an imposed treatment. While Figure 1 shows proportions, this figure represents how an inception cohort of patients in the two compliance groups might be expected to change, depending on success of a rationally prescribed course of ambulatory drug therapy. The absolute numbers are to be interpreted metaphorically, not literally.

Figure 3.

Schematic of how the numbers of unsatisfactory compliers change, according to the failure of an imposed treatment. While Figure 1 shows proportions, this figure represents how an inception cohort of patients in the two compliance groups might be expected to change, depending on failure of a rationally prescribed course of ambulatory drug therapy. The absolute numbers are to be interpreted metaphorically, not literally.

Pharmaceuticals differ in their ability to maintain therapeutic action in the face of occasional omissions of one, two, or more doses. Such differences can shift the boundary between ‘good enough’ and ‘not good enough’ compliance for a satisfactory outcome. The attribute of maintaining therapeutic action in the face of missed doses is called ‘forgiveness’, and has been discussed previously [24], but suffice it to say here that a drug whose duration of action is little longer than the prescribed interval between doses, e.g. 27 h for the progestin-only minipill with a 24 h interval between once-daily doses [55] is very unforgiving of erratic timing, let alone missed doses. In contrast, reserpine, with its several week duration of antihypertensive action is exceptionally forgiving of missed doses in its once-daily dosing regimen [24].

Time is in multiple ways an important element in determining the consequences of variable compliance with prescribed drug regimens. Obviously time since a last-taken dose is a critical parameter in determining when drug action will decline to subtherapeutic levels. Then the length of intervals spent without therapeutic action will influence net outcomes of treatment with widely differing situations, depending on drug and disease, e.g. periods of uncontrolled hypertension and their effects on left ventricular hypertrophy and other cardiovascular phenomena, an episode of breakthrough ovulation followed by conception, a sudden upsurge of viral replication with the possibility emergence of drug-resistant virus, acute rejection of a transplanted organ, or rapid retention of salt and water in congestive heart failure.

Time also plays several roles when hazardous rebound, recurrent first dose, or rubber crutch effects occur, creating temporary periods of exaggerated risk. The longer a period of exaggerated risk lasts, the more likely it is that the patient will incur harm from it. As treatment time lengthens, allowing time for drug holidays to recur, each such period of elevated risk puts the patient once again through a risk gauntlet, within which the patient may or may not incur harm. Successive passages of a group of patients through a risk gauntlet can be expected to pluck individual patients from the category of unharmed but at risk, to the category of harmed. As time progresses, the number of the unharmed but at risk dwindle, either to zero or to some small group of exceptionally hardy or lucky people.

Depending on how one opts to include or exclude patients from a particular grouping, and on how long the aforementioned selection mechanisms have been at work, a resulting group may be comprised, as we have seen, of exceptional proportions of good or poor compliers. Thus, the finding of unusual proportions of good or poor compliers should trigger inquiry into the kinds of prior selection mechanisms that may have been operating. Changes in exposure times may also be expected to change the ‘odds of the nons’ by allowing certain risk factors to operate for longer or shorter periods of time. For example, one might expect compliance problems after cardiac transplantation to rise dramatically, perhaps to be as frequent with cardiac transplants as they presently are with renal transplants [56, 57], if the waiting time to receive a cardiac transplant were dramatically reduced. Such a change would allow less time for the risks of poor or partial compliance with heart failure treatment regimens to undermine such patients' health, thus presently precluding their receipt of a transplanted heart.

At the root of the matter is the power of prescription drugs. The greater their power, the greater the consequences of suboptimal use, and the larger the potential shifts in the odds of the nons as a consequence. Time, always at work, assists.

The author is indebted to Professor Walter S. Nimmo, the organizer of the conference on Clinical Measurement at which parts of this paper were presented orally as the Third Drug Absorption Foundation Lecture. For over a quarter-century, we have been engaged in discussions on various aspects of therapeutics, drug delivery systems, the history of medicine, analgesia and anaesthesia, and many other topics of mutual interest. I would also like to acknowledge Dr Jean-Michel Métry, my 15-year long colleague in the development of electronic monitoring systems for measuring drug exposure in ambulatory patients. His boundless enthusiasm and energy have overcome many obstacles in bringing a new modality of clinical measurement into productive use.