The rising prevalence of Central Nervous System (CNS) disorders due to a growing population and aging society has resulted in a major unmet health need.1 CNS drug development has stagnated in some areas despite concerted effort over the past few decades. With this lack of progress, the rate of treatment-resistant schizophrenia remains high, between 20 and 33%, while schizophrenia remains one of the leading causes of reduction in disability-adjusted life years.2
One common but overlooked challenge in schizophrenia clinical trials is the connection between medication adherence and patient dropout. The problem of patient attrition is particularly acute in schizophrenia clinical trials, with reported dropout rates over 65% and higher as the length of a trial increases.3 Patients who are non-adherent might be more likely to drop out of a study.3 Today, predictive analytics are being deployed as part of modern patient engagement platforms to identify participants at risk of poor medication adherence and possible discontinuation. Data derived from platforms such as AiCure Patient Connect support initiatives designed to keep schizophrenia study participants engaged and retained throughout the course of a trial.
Assessing medication adherence has become an increasingly important part of schizophrenia clinical trials, as patient dropout due to non-adherence can be a major cost and obstacle to validation. There are many potential factors that might contribute to non-adherence – including health literacy level, age and duration of treatment. Patients do not generally communicate their intentions regarding medication-taking to clinicians. There is no recognized non-adherent personality type, and traditionally there has been no standardized, universally valid and reliable approach to predicting adherence behavior. Race, sex and socio-economic status are not consistent predictors of poor adherence. Clearly, obtaining accurate information about a patient’s likelihood of adhering to therapy requires proactive effort from the research team.
Boehringer Ingelheim Deploys AiCure in Schizophrenia Clinical Trials
One innovative effort was initiated by Boehringer Ingelheim (BI) researchers who deployed AiCure’s predictive adherence technology in two schizophrenia clinical trials. The aim of the collaboration was to leverage AiCure adherence information to improve patient retention and data quality in clinical research.4
The pilot initiative utilized data from two phase 2 clinical trials studying the efficacy and safety of BI 409306 in people with schizophrenia or attenuated psychosis syndrome. AiCure’s smartphone mobile app sent alerts to help participants adhere to dosing protocols. The dose event was visually confirmed using the front-facing smartphone camera, plus AiCure computer vision and machine learning. This provided sites with a real-time view of adherence, allowing for targeted outreach and intervention.
Adherence data collected for the first 2 weeks was used to build quantitative, machine-learning models to predict the individual adherence over the trial. Predictive modeling utilizing different
monitoring periods (7-day, 10-day, and 14-day) and adherence cut-off points (0.8, 0.7, 0.6) were explored.
These predictive models were found to accurately identify and predict future poor adherers. The data created opportunities to plan possible intervention and mitigation strategies to improve patient adherence during trials, providing test drugs the best opportunity at proving efficacy.
Based on first 2 weeks of adherence data (combined from both studies), observation of
a participant’s adherence during Week 1–2 of a study predicted an individual’s average
adherence over the remainder of the trial. In addition, observation of a participant’s adherence for the latest 4 weeks of a study predicted an individual’s probability of drop-out (based on low adherence) prematurely from trial.
Watch this video to learn more about using medication adherence data in schizophrenia clinical trials to predict dropout probability from AiCure Chief Medical Officer Rich Christie, MD, PhD.
The BI studies highlighted a significant gap between eCRF compliance (pill count) and AiCure adherence data. In the case of one patient, a 99% adherence rate was reported using pill counts. But the AiCure platform determined an adherence rate of 75% after 25% of doses for this participant were disqualified by video analysis. This participant was determined to be intentionally non-adherent.
The predictive models developed by AiCure and used in the BI studies successfully identified high-risk patients, and can also help to identify those who are generally adherent, those who are adherent with continued support, and those who are intentionally non-adherent.
Finally, the predictive models can be deployed in real-time using AiCure’s validated system and platform, thus creating a unique opportunity to mitigate poor adherence in schizophrenia clinical research while a trial is still ongoing.
Other Therapeutic Areas Benefit from Predictive Models
Beyond schizophrenia research, the predictive analytics models are also enabling new insights into patient behavior in clinical trials across all therapeutic areas. For example, AiCure models are part of development programs analyzing medication dosing behavior to predict both patient adherence and dropout probability in cardiology, immunology, infectious disease, and respiratory disease. By identifying patients at risk of non-adherence or dropout early on, clinical trials can deploy targeted interventions to support retention and collect the most useful data to ensure study power and accurately assess outcomes.
Non-adherence is a major public health problem and an especially formidable obstacle in schizophrenia clinical trials that is likely to continue even if treatment advances are achieved. However, there is considerable promise in using computer vision and machine learning to identify and manage treatment non-adherence, particularly in patients with psychotic disorders such as schizophrenia.
While the reasons for non-adherence are complex, more must be done to quickly identify those at risk and implement interventions to promote long-term adherence. Applying predictive analytics and machine learning to patient data may help pinpoint who is likely to become non-adherent. By focusing research and clinical efforts on improving adherence participants in schizophrenia clinical trials will have a better chance of stable treatment response and recovery.
Please contact us today to discuss your clinical research study and how predictive models can improve medication adherence.
1 Howes, O.D., Mehta, M.A. Challenges in CNS drug development and the role of imaging. Psychopharmacology 238, 1229–1230 (2021). https://doi.org/10.1007/s00213-021-05838-3
2 Gonzague Corbin de Mangoux, Ali Amad, Clélia Quilès, Franck Schürhoff, Baptiste Pignon, History of ECT in Schizophrenia: From Discovery to Current Use, Schizophrenia Bulletin Open, Volume 3, Issue 1, January 2022, sgac053, https://doi.org/10.1093/schizbullopen/sgac053
3 Jørgensen, R., Munk-Jørgensen, P., Lysaker, P.H. et al. Overcoming recruitment barriers revealed high readiness to participate and low dropout rate among people with schizophrenia in a randomized controlled trial testing the effect of a Guided Self-Determination intervention. BMC Psychiatry 14, 28 (2014). https://doi.org/10.1186/1471-244X-14-28
4 Dooti Roy, Zheng Zhu, Lei Guan, Shaolei Feng, Kristen Daniels, Michael Sand. AI-based Adherence Prediction for Patients: Leveraging a Mobile Application to Improve Clinical Trials. Poster presented at the NEI Congress, Colorado Springs, Colorado, November 3–6, 2022.