Tuesday, October 28, 2014

Non-adherence And Other Reasons To Doubt Clinical Trials

In a word - adherence or what we used to call compliance.  That is whether the drug is taken at all or whether it is being taken according the the prescription.  Adherence is an important aspect of the working alliance with any physician.  The only accurate way to determine a response to a medication depends on whether it is being taken correctly and the person taking the medication is describing a true treatment response and true side effects.   Many patients  do not take the medication as prescribed for various reasons.  The question of adherence takes on an equally important role in clinical trials.  If non-adherence is present it affects the statistical power of the study.  Sample sizes are often calculated to bring the study into a range where the sample is considered to have an adequate number of subjects.  Non-adherence rates can adversely affect that estimation.  There is also the question of whether non-adherence reflects another issue related to the medication like side effects.   The related research question is whether non-adherence is evenly distributed across the groups taking the study drug and the placebo group.  In previous discussions of the placebo response in drug trials of psychiatric drugs I have not seen adherence discussed as a critical factor.

A brief news report in Science this week by Kelly Servick provides a good discussion of the adherence issue from a number of perspectives.  The central graphic is from a paper by Blashke, et al (2) showing summary data of electronic monitoring of medication adherence from 95 clinical trials that shows decreased adherence rates in terms of taking the medication and taking the medication as prescribed.  Both fall off significantly over time.   By 100 days 20% of subjects have stopped taking the medication and about 30% are no longer taking it as prescribed.  Those are substantial numbers especially if the active drug can be identified by a specific effect or side effect and discontinued on that basis.  In a field where there is a significant placebo response among subjects with mild to moderate illness non-adherence can lead to significant problems in the final outcome and overall worth of the study.

According to Servick the typical approach used in the past has been to recruit enough subjects to counter the low adherence rates.  This is problematic for a number of reasons.  Subjects these days are often from nonclinical samples.  On college campuses this can be a problem with some subjects volunteering for multiple studies.  With psychiatric drug trials, the recruitment criteria are subjective, obvious, and selection is often coordinated by non-physician research coordinators whose job it is to get the required number of volunteers  in a specified period of time.   In the drug trials that I am personally aware of only Alzheimer's disease trials asked for corroboration from sources other than the patients on how they were able to function on a daily basis.  It would be very interesting to obtain that kind of data on subjects recruited from University campuses who were still attending classes especially if some incentive was involved.  In a related matter, one of the investigators in this area created a database to identify potential subjects who came in for screening at various sites where he was an investigator.  Up to 7.78% of the subjects across 9 sites were identified as duplicates (3).   Because of the potential negative effect of duplicate subjects the authors suggest that a nation wide database of subject should be considered.

The article looks at a number of measures to determine the level of adherence in a study.  The first take home message is that pill counts are relatively meaningless.  I have certainly talked with research subjects who told me that their blister packs were empty because they just threw the pills away.  In a study that compared pills counts with blood levels of the drug the sample sample had an adherence rate  of 92% by pill count but only 70% by blood levels.  The author cites medication side effects as a reason for non-adherence, but in a complex sample of patients with varying levels of motivation and insight the reasons can be very complex.  Several electronic approaches to adherence have been devised that vary from a chip in the pill bottle cap that records when the bottle is opened and closed (MEMS system) to a chip in a pill that records when it is ingested.

Adherence measures are another dimension to look for when reading the results of clinical trials.   I don't recall seeing any commentary on this important issue in Cochrane reviews and probably with good reason.   Non-adherence rates this high are probably at least as important as what Cochrane typically discusses as technical problems like small sample size and measurement problems.  Blaschke is quoted in the Servick article that many of the researchers in this area feel that the problem is bigger than one that can be detected by surveillance and databases.  To me this comes down to the limitations of clinical trials and a problem that cannot be potentially solved.  Certainly the days of research units where subjects could be supervised in inpatient settings for months is gone.  In most cases, persons with severe psychiatric disorders can only get that kind of treatment if they can personally pay for it or the state they live in has a state psychiatric facility.  Even then they often have to undergo civil commitment.  A practical solution would be to eliminate the obviously non-adherent subjects and not include them in any intent-to-treat analysis and use a standard adherence measure such as blood levels where appropriate.  Ambivalence about taking a drug in a research protocol is not the same thing as stopping an FDA approved drug in a clinical setting, but that conscious state has not been adequately studied.

George Dawson, MD, DFAPA

Supplementary 1:  In researching this article I was very pleased to find the full text of Blaschke, et al online but also a reference to the National Academy of Sciences Committee on National Statistics.  That site contains a report The Prevention and Treatment of Missing Data in Clinical Trials that was references in the  original Science article.  The full article can be obtained at no cost form that site with registration.

1: Servick K. Nonadherence: a bitter pill for drug trials. Science 17 October 2014: Vol. 346 no. 6207 pp. 288-289  DOI: 10.1126/science.346.6207.288

2: Blaschke TF, Osterberg L, Vrijens B, Urquhart J. Adherence to medications: insights arising from studies on the unreliable link between prescribed and actual drug dosing histories. Annu Rev Pharmacol Toxicol. 2012;52:275-301. doi:10.1146/annurev-pharmtox-011711-113247. Epub 2011 Sep 19. Review. PubMed PMID: 21942628

3: Shiovitz TM, Wilcox CS, Gevorgyan L, Shawkat A. CNS sites cooperate to detect duplicate subjects with a clinical trial subject registry. Innov Clin Neurosci. 2013 Feb;10(2):17-21. PubMed PMID: 23556138.

4: Czobor P, Skolnick P. The secrets of a successful clinical trial: compliance, compliance, and compliance. Mol Interv. 2011 Apr;11(2):107-10. doi: 10.1124/mi.11.2.8. PubMed PMID: 21540470; PubMed Central PMCID: PMC3109858.

5:  General search on adherence related articles


  1. Thank you very much for this article. This reinforce what I have been posting at 1BOM about RCTs. Which is that archangels with no COI could be doing this work under current standards and still produce junk science.

    The prevalence of this problem also supports the case for using this:


  2. James - could not agree more. The implications of 30% non-adherence and 7% duplicate subjects are huge. The study of any subject influenced by human consciousness is full of potential methodological challenges. The device looks interesting and to succeed it just has to be competitive with the usual blood levels and electronic devices and be competitive. I think it would be more acceptable to research subjects than swallowing electronic chips.

  3. Everything in Ego and the Mechanisms of Defense (written by the practical Freud, Anna) seems to comport with the daily realities of practice and what we see in patients. I've never seen a person who uses projection and splitting as primary defense mechanisms who is more psychologically well adjusted than someone who uses humor and sublimation. Yet nothing in that book has been confirmed by RCTs. This would be an example of practical knowledge clinically useful outside the scope of such research that cannot really be ignored. This really applies to all practical observable psychoanalytic ideas.