Public Health and Data in the Time of COVID-19

Melody Wu

Experiential Ethics Final Project

Public health is a data-driven but also people-contextual field.

History of Epidemiology and Contact Tracing

How Data and Models Came Into Public Health

Brief History of Epidemiology and Data

Field emerges

Hygiene & infectious diseases focus

Disease spread @ community level

Subfields emerge with more tools

1600s

1800s

1900s

2000s

First feature: population data about cases of death specifically in the context of the plague

Looking at spread of IDs based off mapping of cases; bacteria identified as major causes; John Snow, Louis Pasteur, Florence Nightingale

Marked by 1918 flu pandemic; public health emerges; focus beyond ID on minimizing general health problems in communities

Molecular epidemiology emerges w/ more knowledge about underpinnings of diseases and immunology research; internet and data tools

History of Contact Tracing

Used for centuries…

How about now?

• “Typhoid Mary” - asymptomatic carrier
• Traces people who have come into contact with someone with the disease
• Specific focus on diseases transmitted through the air in the 19th century (i.e. smallpox, eventually TB, etc.)

• Electronic reporting increases
• Mobile apps and electronics

Why modeling?

• “Not just to predict, but also to persuade”
• Why does it matter that we get past the “first wave” before the “second wave” comes?
• Determine interventions and recommendations for government to reduce cases but also allow for testing and hospital capacity to be reached
• Better understanding of the spread/basic reproductive # (R0) of the disease, length of immunity following infection
• Allows various factors to be considered

What are models, and how can we think of models in terms of cultivating public trust, and why should we (ethically) care?

COVID-19 Cases Models and Data

How They’re Used and What’s at Stake

What do the models say…

IHME Model

Los Alamos National Labs

UGeneva and Swiss Data Science Center

COVID-19Sim

COVID-19 Forecast Hub

MIT COVID-19 Analytics (ORC)

Conclusions: These models actually seem to under-predict; models can’t predict behavior; each assumes different behaviors and factors

• Uses mechanistic model and curve fitting approach
• Predicts up to end of Dec. depending on scenarios, policies, & resources
• Predicts ~180K deaths end of Aug. (as of Aug. 11th)

• Statistical growth model; probabilistic
• Assumes current interventions continue
• Predicts 172.4K deaths end of Aug. (as of Aug. 11th)

• Exponential and linear statistical models fitting
• Assumes social distancing policies in place continue
• Only predicts 1 week ahead (so much more accurate)

• SEIR model (Susceptible–Exposed–Infectious–Recovered)
• Dependent on social distancing measures
• Predicts 178K by end of August

• Uses 32 models to predict
• Makes predictions only up to 4 weeks out
• Predicts 172,446 deaths by end of August (now, 181K)

• SEIR model
• Assumes once cases reach a threshold interventions will continue
• Predicts 178.4K by end of August

Cumulative Deaths in the U.S.

(2 weeks ago)

​

~July 20th

August 11th

Prediction made one week before wk of July 20th

~July 20th

src: COVID-19 ForecastHub

Cumulative Deaths in the U.S.

(next week)

​

August 11th

Actual: 161, 842 total

~August 16th

Prediction made 4 wks ahead of August 16th;

​

When 1 wk ahead, pred. much closer

~August 16th

src: COVID-19 ForecastHub

Cumulative Deaths in the U.S.

(next three weeks)

​

August 11th

Actual: 161, 842 total

~August 30th

Actual: 182,714 total

Predicts higher typically, but could be closer

src: COVID-19 ForecastHub

SEIR Model over time

Which model do you trust?

None of them / Maybe COVID-19 Forecast Hub…

The Controversial Model from March

Conclusion: You can’t rely on one model and you can’t predict human behavior

• White House primarily used the IHME Model (as shown) — which has turned out to be very wrong
• Predicted about 82,000 deaths in August; we are double that at 162K deaths now
• Can’t really predict beyond 4 weeks out from a given week

What can we trust then?

Model Assumptions / Perspective and Inform Potential Interventions

src: CDC COVID Forecasting page

2 Major Model Assumptions

“The Auquan, CMU, DDS, Columbia-UNC, ERDC, ESG, Geneva, GT-DeepCOVID, ISU, Karlen, LANL, LNQ, LSHTM, MIT-CovAlliance, MIT-ORC (DELPHI), MOBS, Oliver Wyman, NotreDame-Mobility, QJHong, RPI-UW, STH, UA, UCM, UM, UMass-MB, USC, and UT forecasts assume that existing control measures will remain in place during the prediction period.

​

The Columbia, COVID19Sim, GT-CHHS, IHME, JCB, JHU, NotreDame-FRED, PSI, UCLA, and YYG forecasts make different assumptions about how levels of social distancing will change in the future.”

Why do interventions matter?

src: IHME Model

Why do interventions matter?

src: IHME Model

Why do interventions matter?

src: IHME Model

Why do interventions matter?

src: IHME Model

Why do interventions matter?

src: COVID19Sim

NPR’s 9 Takeaways from IHME Model

1. The coronavirus is on track to be the third leading cause of death in the U.S.

… if IHME's projection holds true, the coronavirus will likely be the third leading cause of death in the U.S. for 2020 — behind only heart disease and cancer.

2. The hardest-hit states probably won't bend their curves much

… In some of the hardest-hit states such as Arizona, Florida and Texas, people have already modified their behavior enough to bend the curve… [but] "We don't expect a sharp decline in those states. We expect that deaths will come down a little bit and then we will sort of see a slow, steady set of numbers there." This is due to a pattern his team has noticed when it comes to Americans' behavior...

3. There could be a roller coaster effect.

"When things get bad in their community, individuals are more likely to wear a mask, more likely to be cautious. And that helps put the brakes on transmission." But the flip side of that is that once there is an improvement in daily death tolls, people tend to ease up too quickly.ster effect

4. Starting in November, cold weather could turbocharge this cycle

… when the weather is colder the virus appears to transmit more rapidly. This is a statistical analysis — so it doesn't explain the cause. For instance, it could be that when the weather turns cold, people spend more time indoors. Or it could be that the virus thrives in colder air. But whatever the reason, the impact is massive, according to Murray… “at the peak, which will be the first week of February, we would see approximately a 50% increase in transmission." And he says the effect will kick in starting in November.

5. Things could be worse than projected if hard-hit states don't return to lockdowns

forecast assumes 50% of American schools will be sticking to online-only instruction for the entire 2020-2021 school year… forecast also assumes states will shut nonessential businesses and institute stay-at-home orders once their daily death counts get to the uncomfortably high metric of eight daily deaths per million residents. Four states — Arizona, Florida, Mississippi, South Carolina — have already passed that mark… By November, 16 states are projected to reach it…. None of the states that have reached the threshold so far have gone into full stay-at-home mode.

6. Things might be better than projected if mask use takes off

currently about 50% of people in the U.S. are wearing masks when they are out and about. The team then ran a simulation to see what would happen if starting today, that share was increased to 95% of Americans wearing masks. They found that this would cut the number of deaths by Dec. 1 almost in half — saving 66,000 lives... IHME's team estimates that when officials make masks mandatory, use increases by 8 percentage points. And when the mandates include penalties, there's a 15 percentage point bump.

7. Even with universal masking, many states may need to lock down

In the case of four states — California, Kentucky, Louisiana and Missouri — if 95% of the population started wearing masks, the state would no longer reach the IHME threshold for imposing stay-at-home orders by December. But for the remaining 18 states that are currently at or projected to reach the threshold by December, near universal mask use would only delay the point at which they reach it by an average of six to eight weeks.

8. New solutions could change the model

Murray says… "I do believe that as we get closer to the fall, absolutely the most important question for many states will be, 'Is there something that is less intrusive on people's ability to work and their lives that will still provide enough protection to avoid the death rate getting to a high level?'”.... He adds, "is it enough to have a mask mandate, bar closures, indoor-dining closures, businesses aligned on practices to try to keep their employees safe? And can we model out the effect of that versus the more draconian stay-at-home orders?" Similarly, he says, it will be a priority to estimate the impact of the patchwork of online and in-person instruction in schools and universities, as well as to determine how long lockdowns really need to be kept in place.

9. Not all forecasts are as pessimistic as IHME's

Reich says the forecasts diverge because they are based on differing computer models "that are incorporating different data sources. Some of them incorporate data on recent trends in neighboring states. Some are incorporating information about which age groups are getting infected." Others are not. "All of those different data sources," says Reich, "mean that some models in certain states may be more pessimistic and other models might be more optimistic.

Contact Tracing Models and Data

How They’re Used and What’s at Stake

Is your contact tracing program making an impact?

• Makes basic assumptions about the disease, rate in which it spreads, and the community in which contact tracing occurs
• Allows us to see potential impact of the CT program

src. conTESSA

Contact Tracing Challenges

• Privacy and Stigma
• Testing Delay and Capacity
• Cost and Personnel
• Surge Capacity -- When Cases Surge and When Cases Fall
• Ability to Isolate

Can Technology/Other Tools Combat this? (This impact can also be modeled!)

• Privacy and Stigma → Social Media Communications
• Testing Delay and Capacity → Quicker Tests, Community Level Sewage Testing
• Cost and Personnel → Increasing funding, Community Level Sewage Testing
• Surge Capacity — When Cases Surge and When Cases Fall → Mobile CT Apps, Volunteers
• Ability to Isolate → Non-profit organizational support, Economic relief

What other factors would you consider in modeling the impact of CT?

Communicating Models and Data to Inform Health Decisions

How We Communicate Data Influences Public Health & Politics

Hydroxychloroquine

Let’s start by going way back in the timeline…

Hydroxychloroquine

I’d like to note I was particularly inspired by a friend’s Instagram story on something like this where she did the digging into facts from a heavy scientific paper and boiled it down

• Check out this timeline of the hydroxychloroquine controversy by Beckers Hospital Review and this one too by ABC News
• Conclusions: The debate is still going on and is still very political (most recent update August 3rd)
• I’d like to highlight this particular date’s comments:
• June 4: The Lancet retracted the study it published May 22 that claimed hydroxychloroquine was linked to higher mortality rates in COVID-19 patients.
• On the same day, The NEJM retracted a separate study showing that blood pressure medications were safe to take for COVID-19 patients. Both studies used data from analytics company Surgispher, which refused to share its raw data with study authors or a third-party auditor after questions about its accuracy arose.
• Some thoughts: I think it’s crazy how one singular study can cause such commotion, and it just goes to show how careful scientists and engineers should be when creating therapeutics but also testing them

Hydroxychloroquine - Trump’s Cited Controversial Study (by the French)

March 21: Trump cites success of small French study, publisher later says data 'did not meet its standards'

​

Gautreta et al. - Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an openlabel non-randomized clinical trial

​

• How valid is this study? Not very valid… “The ISAC Board believes the article does not meet the Society’s expected standard, especially relating to the lack of better explanations of the inclusion criteria and the triage of patients to ensure patient safety.” (Statement on IJAA paper - ISAC)

​

• What are the consequences of this statement? March 24th - Husband and wife poison themselves trying to self-medicate with chloroquine (LiveScience)

Hydroxychloroquine - But where did it come from? People say… FOX News

March 16th: FOX News Ingraham Angle program cites chloroquine

Trump administration not ruling out domestic travel restrictions amid coronavirus pandemic (Ingraham Angle Program on FOX News)

“Well, according to a new study, there is such a drug. It's called chloroquine. And that study found that use of chloroquine and its tablets is showing favorable outcomes in humans infected with coronavirus, including faster time to recovery and shorter hospital stays. CDC research shows that chloroquine also has strong potential as a prophylactic preventative measure against coronavirus in the lab, and while we wait for a vaccine to be developed.

​

Tonight, joining me now is one of the coauthors that study Gregory Rigano.”

​

Who is Gregory Rigano? He’s not a doctor. (The Hucksters Pushing A Coronavirus 'Cure' With The Help Of Fox News And Elon Musk - HuffPost)

​

​

I wonder… maybe… they got SARS CoV-1 confused with SARS CoV-2?

On this slide, I’d like to note that I started a COVID-19 Pandemic Evolving Guide back in March when we were sent home, and chloroquine had come up in a conversation and noted on the doc through a friend (before Trump touted it even)

2005 Paper: Chloroquine is a potent inhibitor of SARS coronavirus infection and spread (Vincent et al. Virology Journal 2005)

• Conclusions: “Chloroquine is effective in preventing the spread of SARS CoV in cell culture. Favorable inhibition of virus spread was observed when the cells were either treated with chloroquine prior to or after SARS CoV infection. In addition, the indirect immunofluorescence assay described herein represents a simple and rapid method for screening SARS-CoV antiviral compounds.”
• To be honest, I think this confusion by Fox News is very possible considering instances where politicians have also confused COVID-19 as the “19th” coronavirus when it’s named after 2019 or the year. But the more likely idea is that (hydroxy)chloroquine is one of those drugs in the library of FDA-approved drugs that scientists suspected could have some impact or just wanted to test since it’s FDA-approved.

Okay… but we’re past all this hydroxychloroquine talk now right?

NOPE. Let’s zoom forward now…

• July 2nd: The International Journal of Infectious Diseases published a study conducted by researchers at Detroit-based Henry Ford Health System that claimed COVID-19 patients who received a small dose of hydroxychloroquine within the first two days of their hospital stay were more likely to survive. (Adams, Beckers Hospital Review)
• July 8: STAT published an article highlighting the viewpoint of many clinical experts quick to point out the flaws in Henry Ford Health System's study. (Adams, Beckers Hospital Review)

Okay… but we’re past all this hydroxychloroquine talk right?

NOPE.

July 31st: 'America's Frontline Doctors' tout hydroxychloroquine: Who are they?

• The thing I also hated about this video and article too was how much people talked about just Dr. Stella Immanuel (when there was a whole group of white people right behind her) and how much this also results in black people being considered a monolith again for crazy facts — when, as Trevor Noah on the Daily Show states this is not the case. To add to that, Republicans/conservatives now use her almost like the “my one black friend” concept.

Okay… but we’re past all this hydroxychloroquine talk right?

NOPE.

August 3rd: Two Henry Ford Health System executives wrote in an open letter that the persisting political climate has made any objective discussion about hydroxychloroquine "impossible." (Adams, Beckers Hospital Review)

• I wonder how much longer this will go on… in my head I’m just thinking — find another therapeutic, we can’t just have one anyway!!

What are the consequences of this talk and buzz for so long? … $$$ and lives

Jeez… did do we care that about hydroxychloroquine?

Apparently so.

We also spent so much time and money on just chloroquine.

Data show panic, disorganization dominate the study of Covid-19 drugs (STAT News)

1. A little news and the words of a controversial president can go a long way… and
2. scientists have to be more vigilant in the way they talk about and communicate data…
3. Oh and we should probably do better in educating Americans on data, statistics, and “correlation does NOT equal causation”...

​

My takeaways…

Hydroxychloroquine

I’d like to note I was particularly inspired by a friend’s Instagram story on something like this where she did the digging into facts from a heavy scientific paper and boiled it down

• If I had more time, I think it would be interesting to do an actual read-through of the different papers and studies and hydroxychloroquine and the data behind that but I don’t… :(

​

​

Other Topics Where Data Informs

• Impact on Young People
• Impact on Vulnerable Groups
• Vaccines - Is it Ready?
• Vaccines - Who Should Get it First?

How do you think data might inform public health decisions?

Remaining Questions

Do you think my project offers perspective on the ethics of data use to inform public health decisions?

How well do you think data is usually communicated today?

Do you think looking at data matters when policymakers may choose differently?

Data itself is factual, but the data interpreters are biased…

• Like public health, the models and facts we use and state are data-driven but also people-contextual and people-influenced
• Questioning the facts presented and not taking one model or fact at face value is something that we don’t always do as reactionary human beings!! But we could miss a whole perspective or picture otherwise!
• Models influence our daily lives not only in terms of thinking about our privacy but also in terms of thinking about our decisions

Thank you!

CREDITS: This presentation template was created by Slidesgo, including icons by Flaticon, and infographics & images by Freepik.

​

​

What are models, and how can we think of models in terms of cultivating public trust, and why should we (ethically) care?

Supp Slides to Give Analogy

Google functions that people never knew about

Really this addresses this question of the challenge of what we currently deal with communications-wise:

THE INFODEMIC