aifred

Aifred is a clinical decision aid system that applies cutting-edge deep learning techniques to increase treatment efficacy in mental health, starting with depression. Aifred can help to bring expertise to resource-constrained healthcare environments, allowing for high-quality personalized mental health care to be accessed by a larger number of patients.

Depression is a common, serious mental illness that is projected to be responsible for the majority of DALY’s lost by 2030 (Ferrari et al., 2013), and which causes significant suffering and social burden (Kessler, 2012). 

Patients have extremely variable side effect profiles and responses to treatment. Some patients can spend years finding the right choice for them, and many patients won’t improve after the first treatment (Rush et. al, 2004).

There is no systematic, evidence-based tool that helps clinicians choose treatments in a way that is personalized to a given patient (Lam et al., 2013).

Aifred uses an atheoretical search through large amounts of high-quality trial and research data carried out through deep learning, which has been shown to be an effective approach to machine learning in healthcare contexts (Ledbetter 2016 http://bit.ly/2heqpvW , NVIDIA 2015 http://bit.ly/2vjXeOi ). 

Most studies using traditional statistics require hypothesis-testing that limits the identification of categories of patients across studies. Artificial neural networks are category-inferring systems, and are thus well suited to this problem. 

With better treatment effectiveness and side-effect prediction, personalized treatments can be chosen faster, saving on costs, limiting suffering, and empowering physicians and patients.

The impact of aifred will be personalized treatments for depression, leading to : 

1. reduced recovery times, 
2. minimization of adverse effects 
3. breakdown of stigma through better information for patients 

Less frequent health-care visits because of faster recovery will allow physicians to provide high-quality care to a larger number of patients, especially in regions where mental health treatment is inadequate or non-existent. 

Aifred will be deployed using a SaaS model, made available to clinicians via web, smartphone and SMS. While the initial aifred model will only be validated for North American and European populations, more data will lead to global validation.

Validated depression scales; sensitivity, specificity, positive predictive value, negative predictive value, and ROC curves; faster recovery times compared to treatment as usual. Validation: ease-of-use and safety studies and randomized control trials. - High-quality, more personalized treatment plans for patients with depression

Clinician numbers measured by number of licenses granted and their activity on the app; Patient numbers reported through the app - aifred has at least 20 enrolled clinicians used on over 1000 patients

Agreements in place with the departments of psychiatry and family medicine after successful pilot - aifred adopted at a major Canadian hospital network

Most approaches address diagnostics and have veered away from tackling the field of mental health because of its complexity. 

Aifred is embracing the challenge of mental health treatment prediction. We can determine patient categories in an atheoretical way that is different from classical statistical approaches. 

We’re innovating concepts like ‘Meticulous Transparency’ to deal with ethical issues that arise when deploying of AI in mental health. 

From a machine learning perspective we’re innovating to handle issues related to  data — including the concept of ‘Training Only Variables’ (e.g. placebo group variables, which are included only in the training procedure).

We’ll be developing aifred with clinicians and their patients—to enhance treatment decision-making. 

Human experts are involved right from the start at feature selection for the models, all the way to output interpretation. The machine learning and medical research teams in aifred work hand-in-hand, with human needs and concerns guiding all development. 

Given our strong focus on the ethical implications of the use of aifred ( http://bit.ly/2tPBDZr ) we are ensuring that the use of tools like aifred enhances human welfare while minimizing potential harms.

Aifred will be available to clinicians through a license; it will not be made available to non-clinicians, insurance companies or the general public. This is important because aifred requires input by a clinician to function, and outputs treatment suggestions that should be discussed by a clinician together with their patient.

We commit to scaling prices so that no clinician or health centre will be unable to access aifred to help their patients. It will be deployed as a SaaS model and made available via web, smartphones and even SMS, to be generally accessible to clinicians in our target populations.

All team members at the moment are either at school or have full-time jobs — we have used personal investments so far to sustain aifred. 

Going forward, during the pilot phase and as we grow & scale the solution, we will be looking for angel investment as a source of funds to sustain the venture. 

Additionally, the federal Canadian government offers several grants that are geared towards young entrepreneurs and also specifically towards startups applying AI technologies in novel areas. 

Other avenues that we have investigated include startup-friendly loans from local startup programs, startup competitions and accelerator programs that inject capital to jumpstart the venture. 

Aifred commits to applying industry best-practices to maintain the financial health of the company and re-invest all profits till a stable growth trajectory is achieved.

The key limiting factor is accessing enough high-quality data to train our machine-learning models. 

Additionally, inconsistencies in parameters acquired by different datasets across different healthcare systems must be reconciled accordingly. 

Applicability of the models is limited by the availability of data from each ecosystem. 

Financial resources will become a constraint as we need to pay for access to datasets from medical and pharmaceutical institutions and technology costs to maintain real-world deployments, ensuring their privacy and data security compliance across different regions of the world and as more team members transition to being full-time employees of aifred.

https://medium.com/@aifred

https://twitter.com/aifredhealth

https://www.facebook.com/aifredhealth/

Success of our solution largely depends on the amount of data we can acquire to feed into our deep learning models. By becoming a Solver, we hope to gain access to a network of experts who can provide us with access to larger datasets. We will benefit from resources to tailor technical and clinical needs for integration into healthcare systems across both developed and developing regions of the world. Connecting to national health databases would greatly accelerate the deployment of aifred in each ecosystem and exponentially increase the outcomes from these deep learning models.

Steinberg Center for Simulation, IVADO, ElementAI, Douglas Mental Health Institute, McGill University Health Center, Montreal General Hospital, Montreal Jewish General Hospital, McGill University Department of Psychiatry, District 3 at Concordia University, Building 21 at McGill University, Department of Psychiatry at University of Toronto, Department of Psychiatry at Queen’s University

Spring Health , Health Rhythms , CognitiveLeap