Federico Andres Lois Profile picture
Geek before it became trendy. Performance, C#, Deep Learning and Financial Modeling. Former Founder of @Corvalius.
Deplorable Skymom Profile picture fche Profile picture Phil Hershkowitz Profile picture Squints Profile picture DenysThorez Profile picture 6 added to My Authors
7 Nov
1/ The first rule of Lockdown Club is: You do not talk about deaths per million. The second rule of Lockdown Club is: You do not talk about deaths per million.
2/ Third rule of Lockdown Club: someone yells Sweden or herd immunity, you point out the other Nordics. Fourth rule: only two metrics to a discussion, cases and cases.
3/ Fifth rule: one lockdown per season, fellas. Sixth rule: no deaths, no herd. Seventh rule: lockdowns will go on as long as they have to.
Read 4 tweets
17 Oct
Controversial opinion: those that say its not possible to shield the vulnerable, also won't be able to prove if there is a difference (or lack of it) between the trajectory of the virus at Madrid and Stockholm. Who do you think has let it rip?
1/ There were many "Eureka" moments while working on our paper, but probably the most important of all happened pretty early. Non-linear models are highly sensitive to:
2/ We decided early on to eliminate as many parameters as possible. Location parameters are simple to fix, they are location parameters. Viral parameters also, you can go and say R0=3.3 and you made a choice. How many parameters are left if you do that?
Read 32 tweets
13 Oct
1/ Our preprint with @LDjaparidze is online at @medrxivpreprint
"SARS-CoV-2 waves in Europe: A 2-stratum SEIRS model solution"
medrxiv.org/content/10.110…
2/ We extended the SEIRS model to support stratified isolation levels for healthy <60 and vulnerable individuals.
3/ We forced the model to predict daily deaths curves and the reported age serology ratio for key metropolitan areas in Europe. The immunity level estimations obtained were: Madrid 43%; Catalonia 24%; Brussels 73%; and Stockholm 65%.
Read 11 tweets
2 Oct
0/n Thank all of you who participated in 'The demon game'. I am taking a screenshot because when knowing the whys it loses all value (there is no more asymmetry of information). These 182 responses are 'The sample'.
1/n You may have already known about this thought experiment you just run on, mainly because there are many different variants of it in the literature. This is the one that I have seen lately:
2/n This example is good because the results are clear-cut to show 2 typical sources of error. Poor experimental setups are the bain of our existence and there are myriad ways they can go wrong.
Read 13 tweets
1 Oct
I have my badge of honor. At this rate, the fat tail event is there won't be more twitter to block. :D Image
For context, this is why he got mad with me.
And I love the depth of the rebuttal. Image
Read 4 tweets
25 Sep
[1/n] Preparing everything to respond to the question: "Under our isolation epidemic model. Is it possible to correct government policy mismanagement starting at the end of July in 90 days for Madrid, Catalunya and a few other cities?" What do you think? Answers in an hour or so.
[2/n] For those that are new to this thread, you can prepare and hone your skills in modeling with the Harmless Virus Game:
[3/n] And the more difficult but also important for this new thread "The Vaccine Gamble" game:
Read 20 tweets
24 Sep
The arrival of a vaccine in the context of COVID can be modeled using game theory as a gamble over the expectation of the final death toll. Most countries will have negative payoff after August 2020. Change my mind. cc @LDjaparidze
Let's make it more interesting... Do you agree?
Context is king. For a gamble to exist we need to define clearly the parameters to observe the likely expected result. Let's start with vaccine efficacy (VE). What do you think is the range most manufacturers are looking for?
Read 24 tweets
23 Sep
If my 6yo could understand why universal mitigations like lockdowns dont work, so do you.
And while you are at it, play the game before it expires:
Read 5 tweets
22 Sep
Say you have a completly harmless virus (IFR=0) that can spread at R0=3.3 and you can find via PCR for 19 days. How many deaths per million would you find if you test all deaths in an average european city? cc @LDjaparidze
So now that I got your attention. Let's narrow it down. Our harmless virus would be found during it's spread frenzy at a rate of
OK. It seems I have a few epidemiologists playing. Here is a curve ball. Would change the results if we "Do nothing" (let it spread unmitigated) or if we mitigate it ('lockdown, masks, etc')? I know it is harmless!! Play along.
Read 16 tweets
26 Aug
[1/n] Debido al post de @gummibear737 no va a quedar otra que explicar nuestra hipótesis de cómo las mitigaciones universales (por ej. cuarentena) extendidas en el tiempo causan más muerte que lo que hizo Suecia.
[2/n] Lo que voy a comentar está todavía en desarrollo y ni siquiera es el tema central del paper que estamos escribiendo con @LDjaparidze, así que no intenten leer entre líneas o en detalle.
[3/n] La idea general es: Como muchos fenómenos en física y biología, la curva de muerte es dependiente del camino. De una forma u otra se llega a la inmunidad comunitaria, y habrá muertes durante ese camino. La pregunta es siempre COMO llegamos ahí.
Read 13 tweets
26 Aug
[1/n] Well @gummibear737 pushed me into a position to have to explain our theory of how universal mitigations (for ex. lockdowns) kill over time more people than what Sweden did.
[2/n] Disclaimer ahead, this is early work and not even the central point of our paper with @LDjaparidze. So don't read too much into the details yet.
[3/n] The gist of it is: Like many things in physics and biology, the death toll is path-dependent. One way or another, you always eventually get to community immunity and accrue unavoidable deaths. The question is always HOW you got there.
Read 12 tweets
20 Jul
[1/n] While radical my current working hypothesis is that HIT is probably higher than what Oxford models suggest, even though we arrive at the same seroprevalence for cases curve concavity change. cc @MLevitt_NP2013
[2/n] I postulate that the observation of 15%-25% range for curve concavity is actually a measurement artifact that is not reflecting the actual seroprevalence at the population level.
[3/n] Rationale: Antibody level studies have 2 sources of error, one is calibration over clinical cases which affects the actual on the field sensitivity. Even the best assays will underestimate prevalence.
Read 11 tweets
6 Jul
[1/17] Let's talk about herd immunity, EVIDENCE only. That 70% usually quoted assumes 100% of people are susceptible to infection based on a very restricted model (SIR) that has served us well over time.
[2/17] That 70% can be composed of several immunity traits. From barrier immunity, humoral immunity (antibodies), or cellular immunity. They can be acquired via infection, vaccination, or cross-infection.
[3/17] We know that estimating when we reached that 70% is hard, because... cross-reactivity between pathogens could diminish the susceptible types, barrier immunity is difficult to quantify, ...
Read 18 tweets
28 Jun
Intriguing method. As a note, it is of particular importance the Non-Exponential Growth section to shed some light on serology anomalies like found in Barcelona's study and no seroconversion on intrafamilial exposure settings. [1/3]
In Barcelona's study 29 RT-PCR positive cases (8.53%) do not seroconvert. This may be signaling that serology studies are underestimating (probably by a lot) the underlying infection rate. [2/3] medrxiv.org/content/10.110…
Or worse, when positive confirmed cases infect secondary subjects and those even though potentially having compatible symptoms, never become positive at RT-PCR and IgG but show specific cellular immunity (confirming the infection). [3/3] medrxiv.org/content/10.110…
Read 4 tweets
30 Jan
1/5 With today's numbers, these are the expectations:
H1N1 - 20-40 per 100000 infected *
SARS - 11000 per 100000 inf.
MERS - 29700 per 100000 inf.
Coronavirus - 2076 per 100000 inf.
2/5 The assumptions:
- Everyone infected to date lives.
- Numbers after 8K infected don't change much modeling 2009 A(H1N1) outbreak behavior.
3/5 * Final mortality of H1N1 is just an estimation the latest actual measurement by the CDC shows 1250 per 100000 infected by week 31
Read 5 tweets