The Mystery of Patient Compliance Levels

We don’t monitor levels of patient compliance in the vast majority of clinical trials.

When we refer to patient compliance we count the total number of tablets (injections, doses, etc.) taken versus the total number available to be taken. Some patients take 100% of their doses, while others take 0%, and others take anywhere in-between. So 50% compliance means that only 1 in 2 doses were taken across all the patients. For the sake of clarity in this article, we will use the number of tablets as a surrogate for all types of doses.

So if at 50% compliance, our drug cures (improves) 10% more patients than an inert substance (i.e., a sugar pill), then it potentially could have cured twice as many vis a vis the inert substance had all patients taken all their tablets. If our drug is measured to be 10% better than a sugar pill, it is actually upwards to 20% better, for those who complied with the drug regimen. The upper range (20%) is most likely to happen where the 50% compliance happened because ½ the patients took no tablets at all, and the other half took all their tablets (referred to here as the drug compliance profile). Typically patients taking 80% or more of their tablets would be expected to see a similar cure as someone taking 100%. Anything less than 80% leads to reduced cure, but still some cure. In patients beneath a certain compliance threshold, say 25%, it's unlikely any cure is pharmacologically-driven: they might as well have been given a sugar pill.

It doesn’t matter whether or not patients actually took the inert substance (i.e., their compliance profile) because in either case their ‘cure’ cannot be attributed to a pharmacological intervention. Whether they took the sugar pill or not does not affect their rate of (spontaneous) cure. It could be argued that those who took the sugar pill had a greater placebo effect, but we set that aside as a contributing factor for the purposes of this analysis. Only the compliance of patients taking a drug influences the calculated rate of cure (i.e., the difference vis a vis the inert substance).

Conversely, if at 50% compliance, our drug harms (i.e., adverse events) 10% more patients than an inert substance, then it potentially could have harmed twice as many vis a vis the inert substance had all patients taken all their tablets. We under-report serious, non-reversible, side effects in all our clinical trials that are not monitored for compliance.

The argument is that compliance levels in clinical trials are representative of those seen once the drug is marketed to the general public, so we needn’t worry about compliance level effects on either cures (i.e., efficacy) or harms (i.e., adverse events). If compliance in the clinical trial is at 50%, then compliance in the open market is assumed to be 50%. Our claim is that in the aggregate, the influence of compliance on efficacy and safety in clinical trials will be representative of what will be seen once the product reaches the market. There is no evidence to back this claim and lots of reason to doubt it. In an early HIV drug clinical trial it took an alert investigator to discover that patients were cutting their compliance in half in order to share the precious medications with their mates. This discovery saved the product from being discontinued due to lack of efficacy (and seemingly increased HIV drug-resistance resulting from the under-medication).

In run-offs with existing therapies, i.e., our drug vis a vis the competitor’s, it is assumed that the compliance levels are the same for our drug and the competitor’s. Again, there is even less reason to rely on this claim, since different pharmacological interventions drive different levels of compliance.

Compliance effects on cure rates are not linear. What if our drug at 80% compliance is actually much better than the competitor’s drug at 80% compliance, but slightly worse at less than 80% compliance? If the overall compliance for our clinical trial is 50%, then it becomes very difficult to say which drug is better, since we win with almost all patients who were 80%+ compliant, and lose with most patients less compliant.

Again, the argument is that this clinical trial is representative of the overall compliance rates, and hence of cure rates we should see in the open market. The much better drug loses because irresponsible individuals in society swamp the responsible. The argument is that this forces drug manufacturers to invest in compliance measures (i.e., reminder containers, pill boxes, blister packs) and to invest in clinical trials that demonstrate increased compliance in the aggregate due to those measures. We force industry to take care of those who are irresponsible in taking care of themselves (i.e., yet another hidden tax on the responsible).

In comparator trials we don’t know which group is better because we don’t know which individuals are taking the drugs. Compliance for asymptomatic drugs is widely reported to be about 50%, well below the 80% threshold deemed reasonable for a reliable measurement. Conversely, patients with side effects may just cut down on their medication, giving a false sense of the level of safety of the product. They may only sheepishly report side effects if they had to reveal they stopped taking the medications.

Interestingly, I participated in what may be the only true test of compliance across different arms of a clinical study, as part of a study sponsored by DoseCue Company ( It ran in the basement of Joe's Neighborhood Pharmacy (illustrative name) and was a randomized, double-blind study with a container that surreptitiously captured the date-time of the open-closes of the cap. Various drug tablets were involved, but for the most part they were in drug classes that are asymptomatic (e.g., lowering cholesterol, lowering blood pressure). During the ‘wash-in’ period of 30 days, both arms of the study were unaware (and had no hint from the container itself) that the intent of the study was to measure compliance. The results are shown in the following table.

DoseCue Clinical Trial on Compliance Month 1 Month 2
Control Group 92% 93%
Study Group 89% 97%*

* Reminder notifications revealed (and usage instructions given) to the study group participants for this month.

So in this one instance, compliance across arms was similar, and compliance in the control group stayed constant, assuaging concerns voiced earlier in this article. However,

  • Joe's patients were already highly compliant, so difference across the arms (3%) was limited simply due to the math
  • The study length was two months, so we have only two data points

Only one patient was under 50% compliant, the rest were above 80%. Since average compliance was above 80% for both arms, clinical differences due to compliance would most likely be minimal. It is assumed these compliance rates are extraordinary, given the vast body of literature that indicates open market rates are near 50%.

It is unknown whether or not mere participation in this study drove patients to higher levels of compliance (i.e., the Hawthorne Effect). If that were the case in general, then it could be argued that lower compliance rates in the open market should drive greater safety and lesser effectiveness for a drug than what you see in clinical trials. Again, we don’t measure compliance, so it’s merely a conjecture.

Further Reading