Innovations In Clinical Neuroscience

CNS Summit 2016

A peer-reviewed, evidence-based journal for clinicians in the field of neuroscience

Issue link:

Contents of this Issue


Page 4 of 23

[ V O L U M E 1 4 , N U M B E R 1 – 2 , J A N U A R Y – F E B R U A R Y 2 0 1 7 , S U P P L E M E N T ] Innovations in CLINICAL NEUROSCIENCE S5 ARTIFICIAL INTELLIGENCE/ MACHINE LEARNING E valuating the use of an artificial intelligence platform on mobile devices to measure adherence in subjects with an acute exacerbation of schizophrenia Presenters: Shafner L 1 , McCue M 2 , R ubin A 2 , Dong X 2 , Hanson E 2 , Mahableshwarkar A 2 , Hanina A 1 , Macek T 2 Affiliations: 1 AiCure, New York, New York; 2 Takeda Development Center Americas, Inc., Deerfield, Illinois Objective: The need to minimize medication nonadherence is particularly important in central nervous system (CNS) clinical trials. An artificial intelligence (AI) platform was assessed in measuring and increasing medication adherence in subjects with schizophrenia in a Phase 2, randomized, double-blind study. Design: Subjects in the TAK-063 study who were stable after three or more weeks of inpatient treatment and were discharged were followed up as outpatients for the remainder of the six- week period. Subjects were given devices with the AI application downloaded and asked to use the application for dosing administrations. The primary adherence measure for the study was based on scheduled pill counts; AI platform adherence data were tested for exploratory purposes. Results: The AI platform was used by 26 subjects for 372 subject days; 744 adherence parameters were collected. Five subjects discontinued early (19.2%). For subjects who completed the trial, mean (standard deviation [SD]) cumulative adherence rates based on visual confirmation of drug ingestion (AI application) and on pill count were 82.5 percent and 99.7 percent, respectively. The mean time to use the AI platform was 86.8 seconds per pill. Conclusion: Subjects with acutely exacerbated schizophrenia who were eligible for discharge from the inpatient setting and who completed the study demonstrated high rates of adherence using the mobile AI application. Subjects were able to easily use the technology. Use of the platform did not appear to increase the dropout rate. This study demonstrates the feasibility of using AI platforms to ensure high adherence, provide reliable adherence data, and rapidly detect nonadherence in CNS trials. D isclosures/funding: Adam Hanina and Laura Shafner are employees of AiCure, New York, New York, and consultants to Takeda. Xinxin Dong, Elizabeth Hanson, Thomas A. Macek, Atul Mahableshwarkar, a nd Maggie McCue are employees of Takeda Development Center Americas, Inc., Deerfield, Illinois. Anne Rubin is a former employee of Takeda. NetraMark: predicting placebo and drug response for pharma Presenters: Geraci J Affiliations: NetraMark Corp., Ontario, Canada Objective: We wished to showcase precision medicine through novel machine learning models based on molecular and psychiatric scale data that can impact clinical trial success through placebo response and drug efficacy prediction. Design: Utilizing data sets, we discovered models of placebo response robust enough to reproduce (within the specific treatment paradigm of a trial) and models that demonstrated the patient population purification process for predicting response in a precise way. Results: We found two models of placebo response that had an accuracy of over 85 percent. For some situations we were able to predict if someone was a placebo responder with an accuracy of over 95 percent. Clinical scales alone were used for one of these models. We also based models on miRNA expression data that were capable of accurately predicting response for a particular subpopulation of major depression patients. This provided evidence that our technology is capable of dealing with heterogeneous patient populations. This could be valuable during Phase 3 scenarios for drug protocols that only work for a portion of the treatment population. Conclusion: NetraMark developed a machine learning system to help pharmaceutical companies who must deal with complex patient populations. NetraMark accomplished this by providing accurate models of placebo and drug response. In situations where the efficacy of a drug is limited to a subpopulation whose effect could be "washed out" by the United States Food and Drug Adminisation (FDA) statistical paradigm, we provided what we call a Patient Population P urification process. This process is capable of predicting who these patients are with a high level of accuracy. Our unique system has the ability to help companies get potential medications on t he market through our approach to precision medicine. Disclosures/funding: We are now working with several pharmaceutical companies. At this time, we are unable to disclose all our current relationships. BIOMARKERS AND IMAGING Comparison of actigraphy endpoints for estimating nocturnal scratching duration in patients with atopic dermatitis Presenters: Peterson B 1 , Moreau A 1 , Anderer P 1 , Cerny A 1 , Ross M 1 , Almazan T 2 , Craft N 2 Affiliations: 1 Philips Healthcare, Bend, Oregon; 2 Science 37, Los Angeles, California. Objective: Wrist-worn actigraphy measures of total sleep time (TST), wake after sleep onset (WASO), and sleep efficiency (Eff) have been used as indicators of nocturnal scratching. We compared those indicators with an algorithm that uses high-resolution actigraphy and neural networks (NN) to determine the duration of scratching events. Design: Six healthy controls and 18 patients with atopic dermatitis (AD) wore actigraphy devices on each wrist and were video recorded during one night in a sleep lab. The videos were scored to determine the true duration of scratching (% time). TST, WASO, and Eff were calculated using standard actigraphy algorithms. A regression line of each endpoint against the true scratching duration was used to estimate the duration of scratching events from the sleep endpoints. A new NN algorithm calculated scratching duration directly from the actigraphy data. The error was the difference between the estimate and the true value. Results: The true scratching duration (% time) was 0.1±0.1 standard deviation (SD) for healthy subjects and 3.6±6.2 for patients with AD (p<0.03). All endpoints correlated with the true scratching duration

Articles in this issue

Archives of this issue

view archives of Innovations In Clinical Neuroscience - CNS Summit 2016