Recently I have been writing a lot on some recent studies detailing some dismal results in improving adherence and outcomes using low tech (REMIND Trial) and high tech solutions (HeartStrong Trial). With that in mind, I wanted to tackle a new study from AiCure, a company that has built a platform using AI to help improve medication adherence. Their technology focuses on using the smartphone's camera to watch a patient take their medication. Directly observed therapy is nothing new, it's used in many clinics and in public health agencies. The issue is you are paying a staff member to watch patients drive out to the clinic to take their medication. AiCure's innovative approach via the smartphone camera is novel in a possible solution that is low in cost and based on previously used methods.
The study was a was aimed at the anticoagulation population, where arguably, nonadherence can quickly lead to negative outcomes.
Patients with a previous ischemic stroke and on oral anticoagulation therapy (Warfarin or DOAC) were eligible for the study. Those that were included were randomized into a parallel-group, controlled single-site study. Patients were randomized into the AiCure group or control group, which was based on standard of care.
Patients were followed for 12-weeks, with visits to the clinic every 4-weeks. At visits, blood plasma was collected for PT, aPTT, INR, and blood concentrations for drug level. Pill counts were also conducted.
In total, 28 patients out of 117 screened were included in the study. One patient dropped out at the very start, leaving 13 patients in the control group and 15 in the intervention group. Demographics were primariliy femal (54%), and minority (89%), with an average age of 57 years. A total of 8 patients were on warfarin, and 20 on DOAC.
Adherence was found to be 97.2% in the intervention group and 90.6% in the control group based on pill count. However, based on blood plasma level for drug concentration, rates of adherence then were 100% vs 50% in favor of the intervention group. When only looking at patients on DOACs, the intervention group had the higher adherence rate (100%) with only 50% of those on DOAC being adherent in the control group, while those on warfarin were found to be adherent. Interestingly, when looking at the data from AiCure, visual confirmation was 90.1% for adherence in those receieving DOAC.
This small study struck my interest primarily because AiCure has been off my radar until I heard they actually were producing results. In this case, what interests me the most is the measurement of adherence. We have 3 measurements occurring here (in the intervention arm)
- Pill Counts which are not always accurate. As noted in past experiences and other studies, patients have been found to count out medications before a visit.
- Blood Concentrations of drugs, which can be an issue in some cases. Name in this scenario DOACs has a predictable PK/PD, which is why they don't require routine monitoring, versus warfarin with its erratic PK/PD which may explain why the 4 blood draws demonstrated adherence versus DOACs.
- Directly Observed Therapy (DOT) via the AiCure platform which relied on patients to show themselves taking the medication.
Interestingly, looking at only the intervention arm DOAC patients (n=10), we find that they had a 96.4% adherence with pill count, 90.1% adherence with DOT, and 100% based on blood concentrations. This to me demonstrates that even this tactic of measuring blood concentrations may lead to variable measures of adherence and that the pill count vs DOT do not align when it would have been expected.
So whats next?
AiCure will stay on my radar, and they do have a few other trials being enrolled currently in Hepatitis and Opioid studies, which will be interesting to see.
I'm curious how AiCure will perform in larger populations. The amount of work to collect this data with lab tests is an interesting approach I haven't seen before, and on a larger basis based on DOT vs Pill counts if they continue to do so I wonder if we will still see some discrepency in data.