Developing AI Bots to Manage High-Risk Populations in Pandemics

Develop, Train and Deploy Artificial Intelligence Bots to Manage High-Risk Immunocompromised Populations During Pandemics and Epidemics

Miro RabVass serves as Team Lead and CEO of Pandemics.AI: Partners for Deploying Artificial Intelligence to Protect High-Risk Populations in Pandemics, the organization established to develop the AI-Bot solution. 

Our solution focuses on managing immunocompromised populations during pandemics, when overwhelmed healthcare systems cannot provide equitable and timely access to critical care, real-time guidance and follow-up care to high-risk patients such as cancer patients, pregnant women and the elderly.

The total number of patients living with cancer exceeds 100 million globally with 19.3 million newly-diagnosed in 2020 alone, likely growing by 62% by 2040 [1]. A pooled analysis of 52 studies, comprising 18,650 patients with both COVID-19 and cancer, showed the probability of death for cancer patients was 25.6% (95% CI = 22.0–29.9%) [2]. 

There were over 100 million pregnant women in 2020 [3]. Recent meta-study of 192 peer-reviewed articles demonstrates pregnant women with COVID-19 have much higher odds of being admitted to intensive care and being put on ventilator than non-infected peers [4].

The global elderly population (65+) is currently above 750 million and is projected to double to 1.5 billion by 2050 [5]. An OpenSAFELY study  based on data from 17.3mm patients in the UK showed the COVID-19 mortality risk for the elderly is extremely high (hazard ratio (HR)> 10): 10x greater compared to 50-59 year olds (HR=1) and over 100x greater compared to 18-39 year olds (HR=0.06)[6].

Our solution serves two groups of users: 

(1) End users: cancer patients, pregnant women, and elderly

(2) Institutional users: hospitals, public health authorities, health insurers, and NGOs 

It helps the end users gain equitable and timely access to clinical care, real-time epidemiological guidance and follow-up care. It achieves that by enabling the institutional users to prioritize care provision to the highest-risk patients when healthcare capacity is severely strained by pandemics, while providing individualized guidance to the less risky patients and monitoring their status with follow-ups. 

We work closely with both end and institutional users in all stages of the solution development process (concept, prototype, testing, deployment). At present, we collaborate with US cancer clinics and with their patient populations to develop the AI bot clinical-assessment capability for cancer patients to evaluate their level of risk for hospitalization and adverse outcomes. 

Furthermore, we are collaborating on the following solution capabilities with our Trinity member-partners (partial list):

*  Mental-health risk assessment for high-risk populations with a leading university's Psychiatry faculty 

*  Vaccine confidence with a leading university in Asia

*  Maternal risk assessment with a global public health organization

(1) White paper discussing the approach, methodology, technology and implementation for developing and deploying AI Bots in pandemics with specific focus on informing hospital systems, public health authorities, health insurers  and public health NGOs

(2) Publish five research papers in peer-reviewed journals: two focused on building effective AI/ML/deep learning models for identifying and risk-classifying vulnerable populations in pandemics, two case studies on deploying and testing AI Bots in the US/UK (cancer patients) and Africa (pregnant women), one clinical trial (RCT) discussing the AI Bot development, testing and approval via US FDA’s regulatory path for Software-as-a-Medical-Device (SaMD).

(3) Unique Real World Evidence (RWE) datasets built by our team through the global operation of the AI-Bot solution and made available to the public for free to facilitate understanding how to fight pandemics and manage vulnerable populations during outbreaks.

(4) Our collaboration with multiple institutions to develop the solution will result in unique fusion of clinical, AI/ML, technology, public health, epidemiology, mental-health expertise that will lead to new multi-discipline spin-off research on the best approach for deploying AI in overall pandemic response and in protection of vulnerable populations.

(5) Artificial-Intelligence Bot solution provided as free-to-use public good to any country and organization globally

The AI Bots specifically benefit immunocompromised populations during pandemics by

(1) decreasing transmission, spread, mortality:  

They enable cancer patients, pregnant women and the elderly to have timely access to clinical care, individualized epidemiological guidance and follow-up/monitoring. The AI Bots achieve that by enabling hospitals, public health authorities, health insurers and NGOs to prioritize care provision to the highest-risk patients when healthcare capacity is severely strained by pandemic outbreaks, while providing customized directions to the less risky patients who should stay at home and avoid hospitals where they can be infected easily. Both the high- and lower-risk patients receive follow-up assessments by the AI Bots which also monitor their condition remotely and make referrals to clinicians, if their status deteriorates

(2) protecting the vulnerable and enabling their recovery:

The AI Bots specifically protect target populations during their recovery by monitoring remotely condition-specific risk factors and vaccine confidence/mental-health challenges. They do so by following up continuously with patients at risk of depression, self-harm and other mental-health conditions as well as Social Determinants of Health (SDH) needs and referring them to relevant resources as needed—limiting the long-term impact of mental illness/SDH.  Furthermore, they provide individually customized vaccine guidance and nudges to the hesitant.

Our solution has the potential for transformational impact on millions of lives globally. Our target population are cancer patients, pregnant women, the elderly. There are over 100 millions patients living with cancer globally [1], over 100 million pregnant women each year [3] and approximately 750 million elderly (65+) as of today [5].

Our strategy for scaling our solution globally is based on the partnership model: leverage multiple partner institutions active in or headquartered out of the countries where deployment is planned. We are already utilizing the same strategy with our multiple collaborating institutions to develop and test the solution: e.g., hospitals to provide both clinical expertise, but also feedback on deployment and serve as pilot sites and public-health NGOs in Africa to provide local expertise, program management and relationships with local governments.

 As of the time of application submission, we have 15 partner institutions on four continents and there are 2 more in advanced discussions with us. To enable our strategy, we further plan to seek partnership with international and government organizations such as the World Health Organizations and the UK’s NHS.

We use 4 categories of metrics to measure progress against our impact goals. Some are relevant to later stages.  Example metrics and current status where applicable:

(1) Metrics to measure progress of solution development and implementation 

—No. of bots customized to an institutional user's needs

—No. of partners working with us to develop and implement the solution: 15

—No. of countries in which the bot has been tested, implementation started

—No. of publications based on our work: 6 planned

(2) Metrics to measure solution adoption 

—No. of institutions in which the bot has been implemented

—No. of patients using the AI Bot

—Frequency of engagement with the AI Bot

(3) Metrics to measure solution accuracy and effectiveness

—Accuracy in evaluation and classification: 100%

—Accuracy in responding to patient queries/ referral provision: 85-98%

—Feedback on solution from patients and institutions   

(4) Metrics to evaluate impact on care

—No. of patients that received timely care due to the solution's capabilities

—No. of patients who reported reduced anxiety, depression

—No. of patients that accessed SDH services due to solution referrals

—No. of institutions showing improved capacity management due to the solution

We believe the key potential obstacles that our team and partners need to tackle over the next three years in the process of deploying our solution globally are as follows:

(1) Trust: we need our target populations of cancer patients, pregnant women and the elderly to trust our solution to

—deliver what it promises in times of urgency

—interact with them in a culturally sensitive manner appropriate for the local conditions and histories

—verifiably protect the privacy and confidentiality of their personal and health data

—provide full transparency on solution development, technology, data storage and collection 

 To achieve the above, we work closely with both our target populations and the institutions that serve them to tailor the solution to local conditions. 

(2) Institutional buy-in: we need our target institutional clients (hospitals, public health authorities, health insurers, NGOs) to be willing to test and deploy our solution. To accomplish this, we leverage our partnership model to receive safety and effectiveness endorsements by leading institutions via pilots and clinical trials.

(3) Political climate: governments in multiple countries downplay the pandemic risks and thus may delay solution deployment. We plan to partner with the WHO and other relevant organizations to facilitate access.

Miro RabVass’ formal affiliations: 

Cambridge University:

*School of Clinical Medicine

*Department of Psychiatry

*Behavioral & Clinical Neuroscience Institute

*Centre for Data-Driven Discovery (C2D3)

Truman National Security Forum (US): AI Fellow

Madison Policy Form (US): Cybersecurity Fellow

Harvard Kennedy School: Dean’s Leadership Advisory Council

Alpha Regnum Investments(US): Chief Executive Officer

(1) Partners:  our strategy for developing and scaling our solution globally is based on the partnership model— to leverage multiple partner institutions with unique expertise and resources. The Trinity Challenge enabled us to connect to and collaborate with 7 member organizations that we may not be able to partner with otherwise.

(2) Publicity and media coverage: scaling our solution on a global scale will be greatly facilitated by media coverage generated through The Trinity Challenge. Officials and potential partners in various countries will be more willing to collaborate with us, when we have established public profile through a prestigious competition such as TTC.

(3) Trust: The Trinity Challenge will enable us to build trust with institutional users and future collaborating partners, since many of them are familiar with distinguished individuals involved with the competition.

(4) Mission alignment: Trinity Challenge’s  mission to support solutions delivered as global public goods to all of humanity’s benefit is closely aligned with our vision to develop and deploy our solution worldwide for free.

(5) Funding: since we will deliver our solution for free to the world as public good and won’t have commercial revenues, raising funds through innovative programs such as TTC’s  is crucial.