By Allison Proffitt

July 7, 2020 | Instead of swallowing a pill, you can spit it into a glass of water. Or maybe wedge it into a straw as you take a sip. Or just reach in your mouth and grab it.

It turns out there are many inelegant ways to avoid taking study medication. Ed Ikeguchi knows—he’s seen them in real time with real patients during real trials.

“It’s crazy! And it gets crazier when you see thousands of these! Not just one or two, which you would imagine, but thousands.”

Ikeguchi, CEO of AiCure, has quite a collection of video examples of non-adherence. That’s not surprising. The company’s genesis was in using computer vision to track adherence to study protocols. The AiCure app walks a patient through the process of taking a study medication: at the scheduled time, open the app, show the pill, put it on your tongue and swallow it.

But the non-adherence the company expected was more innocent. Doses were forgotten. Patients preferred for someone to walk them through the process. Ikeguchi says he was surprised at the level of clumsy subterfuge on display.

When the AiCure platform flags an issue, it is reviewed by a human reviewer. We have the ability to zoom in and slow down video, Ikeguchi says. We can see clearly when the dose is missed. The company then alerts the site and sponsor of the missed dose.

Sites’ responses, Ikeguchi says, can be disappointing. “You’d be surprised; not a lot of the sites respond. And in fact, some of the sites actually are almost insulted. They feel like it’s impossible that any of their patients—or a particular patient that they know—would possibly go and spit out the pill. It’s a little bit of an uphill battle.”

To help with the climb, AiCure enlisted Ken Getz and the Center for the Study of Drug Development at Tufts University.

On the surface, adherence in clinical research seems fairly straightforward. Why lie about a study for which you volunteered? But we’ve long known it isn’t that simple. Non-adherence can take many forms. Study volunteers may not take any of their study doses, they may be partially non-adherent, or they may take the investigational treatment but at the pharmaco-dynamically inappropriate time in violation of protocol requirements. A literature review published in 2018 found that the majority (75%) of participants concealed or fabricated information and provided inaccurate self-reports of adherence.

To look at this more closely, AiCure shared de-identified data on the past 23 studies of psychiatric, neurological and neuromuscular diseases with the Tufts team for analysis.

“We basically did a data dump with Ken,” Ikeguchi said. AiCure shared de-identified data on the past 23 studies of psychiatric, neurological, and neuromuscular diseases. “That was the number of studies where we had the most solid data on this behavior.”

From those data, the Tufts team made nearly 260,000 dose observations of nearly 3,000 study volunteers. They only considered events that were intentionally non-adherent: either all information confirmed non-ingestion—the participant visibly removed the study drug from the mouth—or there was some missing information but suspicious activity and additional review can generally confirm non-adherence.

Ikeguchi says the findings were truly surprising: 4% of all confirmed doses were intentionally non-adherent, 48% of all study volunteers had at least one intentionally non-adherent dose, and 5% of study volunteers were intentionally non-adherent for more than one-third of all doses required. The findings were published in Therapeutic Innovation & Regulatory Science in May (DOI: 10.1007/s43441-020-00155-x).

Within those data are some trends. Participants in Phase I clinical trials are less likely to have one or more intentionally non-adherent dose than participants in Phase IV clinical trials. The longer the trial, the more likely there will be non-adherent doses; the larger the trial, the more likely there will be non-adherent participants. But the more pills required by the protocol did not increase likelihood of non-adherence. “Participants in clinical trials that required five pills to be taken concomitantly were less likely to have at least one intentionally non-adherent dose than participants in clinical trials that required only one pill per dose,” the authors wrote.

Participants who are intentionally non-adherent from the first dose are more likely to repeat that behavior. “The mean intentional non-adherence rate was 26.3% for study volunteers whose first dose was intentionally non-adherent compared with a mean of 4.5% for those participants whose first dose in the clinical trial was adherent,” the authors write.

Even geography revealed trends. “Investigative sites based in the Midwest are more likely to have lower relative intentional non-adherence rates compared to those operating in the Western United States,” the authors observe. “And investigative sites in the North East and South East have higher relative intentional non-adherence rates.”

The “why” behind all of these findings will need to be studied further. “That deserves its own exploration,” Ikeguchi says. But these levels of non-adherence have real impacts on how well a drug performs in trials, what adverse events are uncovered, and how fit it is for the population. “Non-adherence leads researchers to underestimate both the true dose response and toxicity from investigational treatments and to overestimate the true dose level required,” the study authors write.

For AiCure, Ikeguchi is pleased with how comprehensively and quickly the platform caught non-adherent instances. These data can now be used to refine the machine learning algorithms that drive the platform. And Ikeguchi has at least one recommendation for protocol development now: before randomizing a participant, give him or her a week to take a placebo and track their adherence to the protocol. Non-adherent patients can either be removed from the study or excluded from statistical analysis.