Q) Adoption: Historically, government health departments have been terrible at consumer marketing. How is your solution any different?
A) (i) Role of Govt and campuses: The ongoing covid pandemic has identified key areas of weakness across levels of society and opportunities for governments to improve. We have seen 20-30% adoption of our PathCheck contact tracing apps within days of official branded launch by governments. We expect the government to promote and incentivize the population with our toolkit. We are not directly responsible for increasing adoption but do help share best practices. Our specific and focused role is to build the toolkit (open source software for app, server, dashboard, and engagement platform). We partner with several regional health IT partners who deliver it to the government or large campuses. This model has worked well during our rollouts over the last year. We built a vibrant network of regional health-IT companies who want to help government officials and institutions.
(ii) Only 5% adoption required: Navigation apps make estimates of traffic density with well under 5% drivers contributing. While the larger goal is to achieve higher adoption rates, we can make statistically significant and meaningful predictions with as little as 5%.
Q) Is this yet another app?: Healthcare is littered with failed consumer health apps (i.e. low adoption, unknown effectiveness), why is this different?
A) Closed-loop with public health: We expect our toolkit to be used for a specific purpose and at a specific time: during the outbreak when stakeholders are pushing it as one of the official and easy way to share data and to stay engaged. Exposure notification (EN) apps as well as other crisis apps usually have very good adoption for that limited window. Some of PathCheck’s EN apps reached 20% - 30% adoption in just days in many of the states. See our work in Guam and Minnesota where public health created a closed-loop system for engaging with infected citizens.
Beyond a diary: We are not building an app that is focused on just UI/UX or health questionnaires. We are building a mechanism for the governments to launch a solution within days of an outbreak and be able to guarantee to their citizens about a solution that is well tested and proven for privacy, security, quality, behavioral factors in addition to epidemiology. We will provide the modular tech stack for NoPeek privacy, crowdsourced data analytics and models.
Q) Why can’t Google, Apple, or Facebook on their own build this kind of crowdsourcing for the pandemic response?
A) Privacy perception and brand risk: Any entity with an expansive view of people’s daily lives can play a positive role in a pandemic. But it is challenging for these for-profit companies to ask for even more sensitive personal health and activity data. Big tech companies care about their brand and are concerned about the implications of their privacy perception, even if algorithmically NoPeek is privacy-preserving.
API, not app: Because of this privacy brand risk, for exposure notification, Apple and Google provided the APIs and minimal app skeletons rather than launching a full-fledged company-branded product. We think they will continue to play this important enabling role. For pandemic response, many new sensor APIs will be required: mask-wearing detected during facial recognition phone unlock to compute risk scores and to estimate aggregate mask-wearing percentage, Oxygen saturation levels (as well as pulse rate) recorded using a finger on the camera by adding another wavelength, improving GPS protocols to create a private diary of GPS trail even when the app is in the background and so on.
Privacy Expectations: We are in touch with these companies, and our team members have worked at these three companies recently including Facebook health and the Apple Privacy team. As far as we know, beyond exposure notification APIs, these companies have yet to implement NoPeek solutions for consumer health. If the big tech companies implement NoPeek for some consumer health, customers may start to expect such privacy guarantees for other services like email, maps, news, social media, and e-commerce. This will challenge their data and targeting-driven business models.
Gathering momentum is more important than deciding who delivers it ultimately: In any case, we can develop these NoPeek and engagement technologies in the specific context of pandemic response. We can scale them to the deployment of billions of people or partner/sell the ruggedized solution to big-tech companies. Either way, the world needs a ready-to-deploy privacy-preserving crowdsourcing solution for health crisis response, and we are passionate to get started and play a catalyst role for the whole industry.
Q) Many symptom trackers now exist with websites or mobile apps. While these are for generic symptoms, why wouldn’t they be a better platform for pandemic alerts rather than a specific-purpose PathCheck solution?
A) We are in touch with various for-profit companies in the symptoms-to-guidance space. Self-reported symptom data is one specific aspect of our solution. It can come from these platforms or be integrated with our app to create a seamless experience. Over time, approved apps can access each other’s online ‘health vault’ with NoPeek. We agree that adoption and daily engagement are tricky problems that each jurisdiction will solve on its own. Our EN apps are used by a large fraction of the population within days of the launch so that gives us hope that in a crisis, citizens look for trustworthy solutions recommended by the governments.
Q) What is your actual evidence of feasibility?
A) There are multiple sub-questions to consider:
(i) Does citizen data and personalized engagement help in quickly taming a pandemic?: Some Asian countries did show that aggressive test-trace-isolate works to tame the spread within three months with non-pharmaceutical intervention. (However, the citizen data collection was invasive and engagement was coercive). Separately, peer-reviewed papers have demonstrated that one life is saved for approximately every 200 citizens participating in crowdsourced exposure notification and significant benefits at even 15% adoption [Abueg2020].
(ii) Can governments and institutions nudge citizens to do the right thing to achieve pandemic orchestration? Behavior change is challenging. We know it works in Navigation map apps if the solution can achieve the quality of engagement and immediate tangible benefits (or involves compliance fines).
(iii) Will citizens trust the privacy guarantees? And will they understand NoPeek is better than consent or pseudo-anonymity?: The awareness about privacy layers and their implications among citizens may be limited. However, we learned from deploying contact tracing apps that governments in democratic countries will deploy an official solution only if it is highly privacy-preserving and has no risks of data breach, security, or fraud. This awareness among government and public health leaders has led to strong support for our solutions. We need strong technology and public education as the alternative scenario of no privacy or data centralization is too dire.
(iv) Why will citizens bother downloading or using yet another app? Will there be an immediate tangible benefit? We have shown for EN, that citizens are very excited to use our new digital solutions in a crisis if recommended by trusted entities and adoption can reach 20-30% within days. The good news with crowdsourcing is that we need a small fraction of the population to participate to create statistically meaningful estimates. For example, with navigation map apps, a small fraction of app users on the road is enough to help us estimate traffic blockage versus a smooth flow.
Q) What evidence do you have for data analysis?
A) We have developed machine learning and deep learning models that can predict prevalence, and forecast outbreaks based on self-reported population-level aggregated symptoms data. Furthermore, we have also developed multi-agent reinforcement learning models for prescribing non-pharmaceutical interventions, given the current state of the disease, economic impact, and more (Glorioso et al., 2021; Patwa et al., 2021; Sukumaran et al., 2021). Our DeepABM system makes use of Graph Neural Networks (GNNs) to scale agent-based models to run simulations with 100,000s agents in less than a few seconds (Chopra et al., 2021). Our outbreak predictions and NPI prescription models were ranked among the top models globally in renowned competitions organized by Facebook and XPRIZE. Members of our team have also been at the forefront of creating privacy-preserved machine learning models. We have published these works as numerous research papers in top-tier AI/ML venues (NeurIPS, ICLR, CVPR, and more).