today, dr rochelle walensky went on CNBC to “lose the script” and speak about her “recurring feeling of impending doom.” she said “right now, i’m scared.” this was all driven by “the case numbers.”
well, it’s been a minute since i looked at this in any detail, so this fearful admonition made me wonder just what data dr walensky was seeing to cause such trepidation and to very nearly bring the CNBC money-bunny who was reporting on this to tears.
as many of you may know, the covid tracking project stopped publishing data in early march and this has made good US data for testing levels and case counts trickier to come by in workable form. fortunately, being a friendly gato with high functioning data hawk pals, i happen to know some folks who are pulling all the same state data that the CTP used to and are aggregating it in the same fashion. so i pulled their data.
as many of you who know me from the last year know, i’m a real stickler on a key point of covid data: reporting case counts without adjusting for testing level is tantamount to lying. this is stats 101 stuff and the fact that it is still rife is beyond bewildering. no one can still not get this fact. it’s being done on purpose and it turns the data into gibberish
sample rate matters. consider a simple example: counting red cars on the highway. let’s say that 10% of cars are red. a million cars drive by. we expect 100k red ones. but we cannot count them all. say we count 10,000. we get 1000 red ones. 10%! great. our count is working. let’s say we up our sample rate to 20,000 over the same group of cars. now we get 2000 red ones. you’d need to be a fool to mistake that for the number of red cars doubling. it’s just the same 10% sampled at twice the rate. but this is EXACTLY what US public health officials and media outlets have been doing with case reporting. we did 116k tests on march 31 2020. we were up to 2.3 million by december 18th. comparing that raw data is meaningless until you adjust for that differential. we need an adjustment to normalize the sampling rate. so i started publishing one. let’s pick up where i left off:
here we are today. this data is through march 27th and all trailing data is normalized to that day’s testing level so it can all be compared. one can easily see that this mirrors the % positive rate on tests quite closely.
one can also see that there is no uptick. in fact, we’re very near the lows since covid began and are essentially flat for about a month. this is a very interesting data outcome because it was, in fact, predicted by my longtime pal @Hold2LLC who has done quite a lot of work on hope-simpson seasonality.
he made an interesting discovery. when you take this chart of seasonal flu by latitude:
and you combine the 2 northern regions to mirror the US, you get this:
he published this in dec/jan. it predicted a basically flat march and after a sharp feb drop which has been just what we got. it also predicts another big drop in april. so that will provide a nice forward test of the predictive power of this model.
a possible confound is that the PCR testing modality we’re using looks to have a significant false positive and non-clinical positive rate. it has struggled to stay under 4% in the past. so, it’s possible we’re hitting a positivity floor based on flaws in the test.
we see the same trend in CDC hospital reporting for emergency room visits. feb drop, mar plateau. this may provide a good check in april as well.
but, at the risk of being contrary, i fear that dr walensky may, indeed, have lost the script. there is just nothing scary here at this time. this sort of irresponsible fear mongering has no place in public health when the data so clearly fails to support it. if rochelle finds this frightening, i fear that perhaps she is not up to handling an actual crisis. this is exactly how one erodes what little faith people have in public health. it really needs to stop. this is not the way to generate public trust, confidence, or encourage willing compliance. this is how you alienate everyone and make them think you’re misleading them or are simply way over your head in a role you cannot manage. the CDC needs to do better than this if they expect to ever be taken seriously in the future.