fatbrain vector blog

THE VECTOR

Discussions on the direction of AI-assisted decision-making.

Healthcare

Posted April 12, 2020 by fatbrain

Viewpoint:  Why Hospitals Can’t Predict Risk in the COVID-19 Age, and Why Dynamic Risk Intelligence with AI is an Answer

No doubt that hospitals in New York, Detroit, New Orleans and other pending hotspots feel like a war zone. The pressure is real and intense on staff to care for patients, keep the community safe and keep each other healthly. No one in the healthcare system is taking COVID-19 lightly.

Our first-hand experience since the outbreak within the NY Healthcare system

We’ve experienced first hand the chaos and pressure placed on the healthcare frontline, seeing this pressure build since the beginning of the outbreak.

Over the last 3 weeks, we’ve witnessed the courageous response by one of New York’s finest regional hospital systems:

  • LOB staff of 5,
  • Supporting 10,000+ health care workers,
  • Across 6-hospital systems,
  • Managing 700+ COVID-19 hospitalizations,
  • 75+ hospital staff with COVID-19,
  • Less than 450 ventilators available for use,
  • Absorbing admissions from additional adjacent counties, bracing for the oncoming surge from New York City.

This NY hospital group is one of the smartest, most dedicated teams I have ever worked with. My gratitude goes to them. I don’t think I could walk 10 steps in their shoes right now, and feel fortunate that the right people are in place to do battle with the outbreak.

Health teams everywhere are scrambling and doing everything to understand “What comes next”, and caring for the ever-shifting flood of new patients. We’ve seen up close the struggle to assess COVID-19 risk as hospital staff don’t have the right toolset to get clear quality answers that can instill calm during the chaos. The COVID-19 playbook is being built (and rebuilt) in real-time across hospitals because the COVID-19 impact is so hard to predict.

Why COVID-19 is so hard to predict

Predicting COVID-19 risk outcomes is incredibly hard. It’s like a daily never-ending struggle to safely run airports without a modern reservation system, with passengers in great numbers overwhelming the shrinking volume of available seats in the airplanes, supported by the ever diminishing teams of rested pilots and crew.

To understand why COVID-19 impact is so hard to predict at the hospital level, Nobel Laureate Phil Anderson’s “More is Different” mission provides the right perspective. It urges an integrative approach to Science recognizing that Nature has a complex hierarchical structure built in accordance with fundamental laws, such that at each level of complexity, new properties, laws and behaviors emerge. That means examining in a new, integrated way epidemic modeling across populations, viral dynamics in individual patients, and therapeutic treatment alternatives, each with hierarchical level-specific risk and uncertainty. Pulitzer Prize-winning Columbia Professor Siddhartha Mukherjee dives deep into one aspect of the context for outcome complexities of viral dynamics inside patents in the recent “How Does CoronaVirus Behave Inside a Patient?” New Yorker article. The questions raised are legion.

Impact at the local hospital level – (re)building the COVID-19 playbook in real-time

Using “More is Different” approach as the base, when the basis of FiveThirtyEight’s article “Why It’s So Freaking Hard To Make A Good COVID-19 Model” is layered over top, the full spectrum of the Why COVID-19 is so hard to predict becomes more clear.

(source: FiveThirtyEight – https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/

Hospital staff is perpetually stressed to make safe, quantified risk decisions. Staff is challenged by the fact that by the time the data is manually integrated for daily decisions, the data has changed to form a new set of assumptions that need to be considered. This dynamic is what makes getting clear answers or getting to calm so difficult.

Getting to calm occurs when a hospital has confidence their risk-based approach will keep the community safe today, tomorrow, next week, next month, next quarter and next year. Getting to calm means having a clear path forward. The series of inputs and decisions inside this risk-based approach continues to expand:



Risks

  • Do I have a breaking point ?
    ( ie. When overrun)
  • At what point will risk levels dramatically increase for staff, patients, community?
  • How long are we in green across all risk factors?
  • When is the first yellow? When does that go back to green? Can it go red?
  • What is happening in neighboring areas? Will we need to take in an overflow?

Inputs

  • How much viral exposure (akin to radiation risk) to control for each hospital staff,
  • How much viral exposure for each kind of patient,
  • Policy choices to manage resource scarcity of PPE allocation and re-use e.g., irradiating used masks and gowns, while new ones are out of stock, on order.
  • Policy choices for: (i) days off work once staff tests positive, (ii) reuse of PPEs, when and method used (iiI) viral exposure control for each worker

Decisions and Insights

  • How to staff each ICU each day, including:
  • How many of what kind of specialist,
  • When to return the COVID-19 positive healthcare worker back into service e.g., 3, 7 or 14 days,
  • Each question and decision, yielding dramatically different staff availability and overall COVID-19 fighting capacity,
  • When is it advisable to lend a portion of our ventilators to help a neighboring hospital?

Each decision is myopic and un-coordinated, posing daily uncertainty and unquantified risk. This results in a playbook being rewritten on a daily basis because the risk factors are changing on a daily basis, and impacting others by either delaying their onset or accelerating the day of breakage. Chief Economist for the Bank of England, Andy Haldane highlighted the burdens of such unmanageable complexities in “The Dog and the Frisbee” and suggested the way forward in a broader context of regulating risk must involve a simple, learned approach.

There is no standard template.

(source: FiveThirtyEight – https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/

Not having the right toolset is resulting in significant time being consumed understanding the interdependence of the risks, inputs, and decisions. By the time the consolidated risk analysis is complete and ready for circulation and decisions, there is a new set of assumptions. It’s a constant never-ending cycle of catch up in real time.

Not only are the internal risk factors changing daily and externally as well. Each region, whether a city, county, state and/or country is different. Different inputs drive different infection rates, different hospitalization rates (and the % needing ventilators) and different morbidity rates.

The external data need to assess the capacity risk differs across regions, including:

  • Age cohorts
  • Pre-existing conditions
  • Degree of social distancing adherence
  • COVID-19 infection stage when patient seeks treatment
  • Strain concentration within a specific geo

To build a better model, a hospital needs to be able to benchmark against a region that closely resembles their external makeup and service area. No central spreadsheet exists to help with this effort, and the ‘ideal’ match may be across the country or the world, not necessarily ‘next door’.

And, there is the crux of the issue: At the intersection, there is no standard template.

Dynamic Risk Intelligence with AI is Best Suited for Rapid-Evolution Environments like COVID-19

At FatBrain, we’ve learned the limitations for institutions operating in a siloed template. In financial services, that template comprises a rules-based system modeled to catch financial crime in the form of money laundering, terrorism financing and fraud. These systems are terribly inefficient and an ineffective match for ever more sophisticated criminals.

Interestingly, there are similarities between complex money laundering schemes and pandemic risk outcomes being managed inside hospitals today. In each case, the input dynamics are changing rapidly, resulting in ‘learnings’ that need to be captured and fed back to turbo charge the assessment models. When this is done, outcomes improve dramatically with less risk and uncertainty, and no more additional efforts.

Each COVID-19 risk factor the hospital is dealing with is interdependent. Therefore, as one changes (improves or deteriorates), a material impact can be felt on other risk factors. Which is why a) COVID-19 modeling is so difficult and b) near impossible to assess risk when done with uncoordinated, manual efforts.

Recognizing the complexity challenge facing hospitals, we knew we could help. Seeing these similarities firsthand enabled us to quickly retool our dynamic risk intelligence framework used by top 10 banks to fight financial crimes in a new platform called FatBrain Odin Health. To control risk, risk factors need to be quantified in an integrated, explainable manner.

Dynamic Risk Intelligence with AI is ideal for rapidly changing conditions. FatBrain’s biomimetic approach to AI is grounded in human thinking and brain biology, making it best suited for managing risk controls and quantified decision-making within the hospitals facing COVID-19 today.

Working around the clock with regional NY hospitals and staff over the last 10 days, we’ve launched FatBrain Odin Health to boost the global fight vs. COVID-19. And we’ve made the Essential Edition Free to any hospital in the world (we’re also paying the hosting charges). Why? We believe that simply, hospitals don’t have the luxury of time to get the essential non-PPE purchases approved. It’s part of our duty to join the fight and to provide support, because helping reduce the stress of the front line and ultimately save more lives is the right thing to do.

Access Odin Health on our website: fatbrain.ai/odin

We are not stopping there. We continue to work fast without sacrificing privacy, security, auditability, and Enterprise governance controls. To these and additional innovative ends, Odin Health will continue in active development over the course of the crisis, always improving, always getting better so that if the outbreak takes another turn, hospitals can be ready.

Endnote – Building Safe Zones to Regain the #OldNormal.

Battle vs. COVID-19 will be won at the local level. Pockets of ‘safe zones’ will emerge there, then spread to create a growing safe zone. In other words, the voice of ‘All clear’ will build from local leaders, supported by federal and global efforts to feel safe in each community.

Dynamic Risk Intelligence with AI is a path forward to the #OldNormal. Ensuring the health systems don’t break is the strategic imperative to regain the calm we had lost on December 31, 2019.

Let’s get down to business.

Learn how FatBrain will help you effectively navigate your ever-growing data maze to quickly monetize troves of transactional data with real-time decision making.

Talk to FatBrain