Please create an account to participate in the Slashdot moderation system

 



Forgot your password?
typodupeerror
×
Medicine Math

'Why Modeling the Spread of COVID-19 Is So Damn Hard' (ieee.org) 117

Slashdot reader the_newsbeagle writes: At the beginning of the pandemic, modelers pulled out everything they had to predict the spread of the virus. This article explains the three main types of models used: 1) compartmental models that sort people into categories of exposure and recovery, 2) data-driven models that often use neural networks to make predictions, and 3) agent-based models that are something like a Sim Pandemic.
"Researchers say they've learned a lot of lessons modeling this pandemic, lessons that will carry over to the next..." the article points out: Finally, researchers emphasize the need for agility. Jarad Niemi, an associate professor of statistics at Iowa State University who helps run the forecast hub used by the CDC, says software packages have made it easier to build models quickly, and the code-sharing site GitHub lets people share and compare their models. COVID-19 is giving modelers a chance to try out all their newest tools, says biologist Lauren Ancel Meyers, the head of the COVID-19 Modeling Consortium at the University of Texas at Austin. "The pace of innovation, the pace of development, is unlike ever before," she says. "There are new statistical methods, new kinds of data, new model structures."

"If we want to beat this virus," says Mikhail Prokopenko, a computer scientist at the University of Sydney, "we have to be as adaptive as it is."

This discussion has been archived. No new comments can be posted.

'Why Modeling the Spread of COVID-19 Is So Damn Hard'

Comments Filter:
  • Can't get a straight answer

    Like economic modeling, nobody accounts for the emotional human

    • by burtosis ( 1124179 ) on Sunday September 27, 2020 @09:24AM (#60547810)
      Did the model account for testing? Because if we stop testing now, we would have very few cases. I plan on using a similar tactic to fly by obliviously walking off a ledge while refusing to look down.
      • Re: (Score:3, Informative)

        A 'case' is not an 'infection'.

        This distinction is incredibly important for modeling.

        Infection diagnosis should involve two types of laboratory tests to eliminate specificity errors.

        A 'case' is a non-clinical term that can vary by jurisdiction, and a case does not need to be an active or contagious infection. Due to the prevalence of coronavirus antibodies from various prior cold-like infections, case assignment by antibody test is highly unreliable.

        The potential result of substandard testing an
    • by jythie ( 914043 )
      Human emotion is at the core of most modern economic models.
      • Correction;

        The modeler's emotions are at the core of most modern economic models.

        Ok, yeah, I'll admit it, the modelers are human too...

    • by hey! ( 33014 )

      Not human, but human*s*. There is a spectrum of behavioral patterns; the proportions of population that exhibits these behavioral patterns are model parameters.

      Models at best give you a *contingent* prediction, although calling it a "prediction" is a bit of a stretch: if these conditions hold then it's reasonable to expect such-and-so. But from a policy standpoint what is often most important isn't a model's predictions per se, but developing insight into how sensitive a situation is to each of its param

      • by rtb61 ( 674572 )

        You can not properly model the spread of a virus beyond really basic blob models because any attempt at accurate modelling will results in wildly varying results every time and the more they vary the more accurate they are. You are attempting to model chaos, the actual spread, will be tied to how resistant or not unknown people are and how those clusters will propagate. Hit a low resistance cluster of people and it will spread like a wild fire, hit a high resistance cluster and it dies right down but contin

  • by Pinky's Brain ( 1158667 ) on Sunday September 27, 2020 @09:25AM (#60547812)

    They try to fit their models of transmission to their retroactive models of population proximity and then fit variables till it somewhat corresponds to modelled infection numbers ... all incredibly haphazard.

    So much supposition, so much under/overfitting, so few controlled experiments.

    • by cirby ( 2599 ) on Sunday September 27, 2020 @09:40AM (#60547848)

      It's even more chaotic since they're finally figured out that the susceptibility of different people varies wildly, and not just from age or preexisting conditions.

      How much vitamin D you have in your system, how much of various trace elements, what your blood type is, what other coronaviruses you've been exposed to, what your normal T cells can handle, et cetera, et cetera.

      The good news is that it's starting to look like the population that has caught COVID so far is mostly the "very susceptible" cohort, and the ones who haven't caught it might be somewhat resistant - or, in many cases, naturally immune. They're just now getting around to quantifying how big (or small) those numbers are. We're not at herd immunity in the US yet - but we might be a lot closer than most people think.

      • "We're not at herd immunity in the US yet - but we might be a lot closer than most people think."

        Herd immunity is not a strategy or a goal, it's the almost inevitable result if you do nothing.

        It's like someone is falling from a plane and your plan is they hit the terminal velocity of a person without a chute. The theory is it will stop them from accelerating. You are correct, they stop accelerating mid-fall, and hit the ground in a splat. By doing nothing at all, you stopped the fall and avoided falling

  • The source data is so dirty that reliable modeling is not possible at this time.

    As soon as laboratory diagnostics became available in the US, the apparent crisis dropped below epidemic threshold in nearly every State and has remained below that threshold since May. This suggests that early alarm for the general population - largely based on domestic 'case' assignment by non-laboratory diagnostic indicators and fatalities highly concentrated in elder care and skilled nursing facilities - may have resulted
  • It's because algorithms are and always will be glorified calculators that are only capable of linear logic. They will always fail in the face of real-world realities. Tracking is another matter, but duh, you don't *need* to model a virus in this fashion.that just isn't how viruses work. Millennial scientists (I use the word loosely) and engineers need to go back to school and actually learn something. You can run models until you are blue in the face - it won't change a damned thing in this regard. A cup w
    • by jythie ( 914043 )
      Well, yeah, models are imperfect, anyone who works in a field that uses them know that. They are also the underpinning of the modern world. Do you think, when people build a bridge, building, or airplane just sorta 'guess' and use 'common sense'? No, they use imperfect models to design things. Millennial scientists and engineers learned a lot in school.. kinda sounds like you did not.
  • by joe_frisch ( 1366229 ) on Sunday September 27, 2020 @10:26AM (#60547940)

    Despite the huge and highly regulated medical system in the US, we did a terrible job of having carefully crafted reporting schemes, and that has made a lot of the data nearly useless. Combine that with (in this case) misplaced privacy rules, and there is no high quality data to analyze. Pour party politics on top of this and we are doing about as well as a medieval village during the plague.

    • On the money. Decisions are being driven by mathematical models largely fed with garbage data. We have no way of knowing how well the model might work with reliable data. As for privacy, we can tell people to stay home, force people to die alone without the comfort of family and friends and destroy people's life work. But god forbid we violate their privacy be telling people they caught covid. It tells you something about the limited range of values that are informing these decisions.
  • The biggest problem is differentiating between covid-19 and 'regular' flu (which appearantly isn't really possible). Looking at figures of the 'regular' flu in our own country where in a 'normal' year around 6000-7000 people die because of it, this year, including the 'corona' deaths we haven't had that amount yet, but are close to it. Knowing that the tests also might show positive in a lot of cases when you have/had the flu, how accurate is this whole covid-19 scare? We never tested this much for the 'nor
  • At this point, the disease is clinically over. All the graphs are showing that while the false positive PCR test results are exploding, the actual hospitalizations and deaths are hovering close to zero in the UK, France, Germany, Canada and elsewhere in Europe - the US will soon follow.

    Exploding false positive cases: https://www.google.com/search?... [google.com]

    No actual deaths: https://www.google.com/search?... [google.com]

    The graphs clearly show that whatever people are doing with obsessive compulsive hand washing, masking

    • Pants on fire. The number of deaths in most European countries, even here in Germany, has gone up, back to the June levels. You simply don't understand that infected people don't immediately drop dead after they have been tested, it takes several weeks.

      • You obviously didn't look at the linked graphs.
      • Two deaths in Germany yesterday. The cases have started going up weeks ago. The deaths remain close to zero. https://www.worldometers.info/... [worldometers.info]

        The case numbers are mostly false positives and the people are not really sick at all. What you have is a Testdemic, not a pandemic, due to bad tests.

        • The number of deaths is always low on the weekend. There have been 17 deaths on wednesday. This kind of numbers is what we previously had in June. France had 150 deaths on friday. The cases also started going up because everyone returning to Germany from abroad had the option of a free test so more asymptmatic cases were caught.

          So yep; you are a liar. Or a covidiot. Maybe both.

          • Dude, please don't stop taking your meds, or ask your mom to hand it to you each morning.

            If you could be bothered to click the link - that finger next to your thumb, on the left mouse button, then you would see the nice graphs which are public information, not made by me.

            Any rational person can see that the cases have been going up for weeks, with no corresponding rise in deaths a few weeks later.

    • If you think that the only reason death counts go down over time with a new disease then you're a fucking idiot. Overnight, this has become the most studied disease on the planet, and doctors have been constantly trading and revising information on the best way to treat patients. Hospital capacity has increased— as hospitals reach capacity, the number of deaths will increase at a significantly greater rate per infected person because infected people are receiving inadequate care. There are many factor

  • I've been modeling since punchcards into 370 RJE. Covid was irresistible to model even back in January. One (or more!) lines per day, whatever calcs you wanted on the line (upward references) and copy down. Easy-peasy.

    The problem is it always blows up (sometimes down). With whatever delay function, sooner or later it just goes. But that does not match the data of rising-to-peaks from a slow simmer. Something is very wrong with the bug (unlikely) or reporting (uncertain can be large enough).

    • The trouble is with the tests. The tests are actually fine, but the results are used wrong.

      The PCR test is a negative test: It Rules Covid Out.

      If the result is negative, then you can be 100% sure that you are not infected. However, if the result is positive, then you are only 5% sure that you are infected. So on its own, it is not good to indicate sick people.

      At this time, most people have some dead viral DNA fragments in their noses and the PCR test picks this up, causing an explosion in false po

      • As I understand it, the opposite is true. John Hopkins found there is a VERY high number of false negatives. As high as 85% in the first few days after exposure and 20% even for people showing clear symptoms. That is based on real life testing results, not research labs.
  • The models fail to predict massive, coordinated, secret, and well-funded efforts to continue spreading the damn thing on purpose.

  • The OP left out Cubic Models, created with the Excel curve-fitting function. Very useful.

  • The reasons are both complex and extremely simple, but all tied to this: fatally flawed data - by design. No conspiracy theory is involved required.

    The Chinese government front-loaded this thing with their not only incompetent, but deliberately wrong data. This stuff was pumped through WHO where their non-doctor puppet pushed it out to the world as seemingly neutral and reliable info.

    Governments all over the world then produced bad data. In many countries this was from sheer incompetence, understandably - m

    • "No conspiracy theory is involved required. The Chinese government front-loaded this thing with their not only incompetent, but deliberately wrong data. " Great segawy into a conspiracy theory.
  • The most harmful projection ever made was the one made by the Imperial College London that grossly overestimated deaths by covid-19 and was the driving force behind all lockdown. Never before in history have one single prediction had so negative impact
    • So true. Nobody should ever again believe anything coming from Imperial College London.
    • by ebvwfbw ( 864834 )

      My understanding is worse - it was based of a researcher's Daughter's science fair experiment.

      Then - no other scientist in the world spoke up that should have spoken up and said - this is shit.

      Then in spite of scientific facts showing that Hydroxychloroquine works, they pan it because Trump said it might work. How idiotic.

  • In the United States, COVID was spread by rich Americans who flew to China on Business trips to put more American workers out of work. It was the poor people who got sick and died the most, likely from the lack of medical care.
  • The government response to covid is a matter of public policy, not science. Science can help inform those decisions or it can harm those decisions. In the case of these models, they are mostly just noise with their apparent certain outcomes just misleading the public and decision makers. "Garbage in, garbage out" is just one of the many problems with using them to inform decisions. The countries that have been most successful are the ones that relied less on abstract academic science and more on concrete ex
  • The board game put out by Z-Man games seems to do the job quite well.

    Up to 4 players.   And it is 12 years old!

    Fun for the whole family!

    Crazy thing about the game unlike other games, it teaches players to COOPERATE rather than compete against each other.  Players fight the game engine.  Either all Players win, or the Pandemic does.

Passwords are implemented as a result of insecurity.

Working...