Thread by @Bob_Wachter

1/ Covid (@UCSF) Chronicles, Day 70

First, @ucsfhospitals, 15 cases, 6 on vents (Fig L). Mild uptick. SF hospitalizations stable-to-down (52 now). Cases plateaued, still averaging ~40/d – no change in past mth. No SF deaths for 10d (total=40)(Fig R). Overall, picture is stable.

2/ New @WSJ piece on confusing Covid test results on.wsj.com/3c4ri1a. Hard to cover this complex issue in 1000 words, so here's a tweetorial. We'll start with a few core principles, then try to walk through three confusing scenarios. Yes, there's some math, but still fun.

3/ With more testing & various versions of 2 different types of tests, the confusion is unsurprising. Last month's “Can I fly?” or “Can I touch the mail?” questions are increasingly like this: “My daughter tested positive, then neg, then… What does that mean?”

4/ I won't review 2 major tests for Covid: viral vs antibody. Lots of good primers; eg, bit.ly/3ekJIMB. To interpret a test, one must understand the differences (@CDCgov may have forgotten this: bit.ly/2LZpLyK), including the timing: viral early; antibody later.

5/ But even good primers like this one (by @UKRI_News bit.ly/3ekJIMB) necessarily simplify things: a positive test means X, a negative test means Y. If only it were that simple.

6/ In interpreting any medical test (an x-ray looking for cancer, EKG for heart attack, a prostate exam), key point is that no test is perfect. Some (eg, HIV tests) are close, but all tests can be wrong, both ways: negative when someone has the disease, positive when they don’t.

7/ Terms to know: "Sensitivity"=Positivity in Disease, or the proportion of people with a disease who have a + test. If a treadmill for coronary artery disease is 80% sensitive, 8/10 people with CAD have a + result. Thus, 20% of pts w/ actual CAD will have “false negative” result

8/ "Specificity"=Negativity in Health, or the proportion of people who DON’T have the disease who have a neg test. If a test for antibodies to SARS-CoV-2 is 99% specific, then 1/100 positive results will be false positives: the test says they have antibodies, but they truly don’t

9/ Enter Reverend Bayes, whose theorem is key to diagnostic reasoning bit.ly/2TGhAeZ. To interpret any test result, said Bayes, you must know 2 things: a) How good is the test (sensitivity & specificity) & b) How likely was the person to have the disease BEFORE the test.

10/ There are complex formulas & online calculators to help w/ the math (I like bit.ly/3d5IhBH, use 2nd set of boxes). The bottom line: unless a test is always correct (100% sensitive AND 100% specific), you’ll misinterpret results unless you apply Bayesian reasoning.

11/ OK, let's go to Covid test results. We internists apply Bayesian reasoning about 50 times a day. But for a newbie, this is wildly confusing stuff, made worse by all kinds of Covid-specific head fakes: asymptomatic infection, viral vs antibody, time & location dependency...

12/ Below, I’ve listed stuff we now know about Covid tests that influence the interpretation of results. Note that all 3 inputs (sensitivity, specificity, & prevalence, which get plugged into the calculator bit.ly/3d5IhBH) are nuanced & often require as much art as math.

13/ OK, with that in mind, let’s get to 3 common testing scenarios. I have a brief discussion in the tweets, more detail in the panels below (sometimes 280 characters just doesn't cut it). I’ll post this later to @Medium to make it easier to read the text.

14/ The first case: A 40-year-old female has flu-like symptoms. She thinks she has Covid (as do all patients with pretty much any symptoms these days). Her viral PCR is negative.

See Part 1 of my discussion in the panel below.

15/ Much depends on the prevalence in her community & how long she’s had symptoms. She probably doesn’t have Covid, particularly if she’s in a low prevalence area. But, depending on risk factors, it would be prudent to isolate & consider retesting in a few days.

Part 2 below.

16/ OK, let’s try a 2nd case: A 65-year old male was diagnosed with Covid 3 wks ago with a positive viral test. He is required to have a negative test to return to work, but a repeat test, taken 16 days into his illness, is still positive. He is a bit achy but mostly feels better

17/ A common and vexing scenario. Studies have now confirmed that Covid patients are no longer infectious 11d after illness began. Cases of persistent PCR positivity represent dead virus, no longer capable of transmission, as shown in this Singapore study bit.ly/3c6IxPB

18/ OK, here’s one final case: A 54-year-old woman has a positive SARS-CoV-2 antibody test, done as part of a screening program at work. She doesn't recall any symptoms in the prior three months.

19/ Antibody testing introduces its own twists & turns – it’s crucial to channel the Reverend Bayes to make the right call. In the panel below, I show that if this woman is in low-prevalence San Francisco, there would be a ~71% chance that her antibody result is a false positive.

20/ This panel shows that it’s even more complex, since the results on one test (a viral test) influences the math we use for a subsequent test (eg, antibody test). Each new data point – prior tests, the patient's course, what’s happening in the community – shifts the odds.

21/ I find Bayesian reasoning to be among the most fascinating topics in medicine – and it’s absolutely essential to unraveling the mysteries of Covid testing. It’s also key to interpreting other tests, like political polls, as @NateSilver538 knows.

22/ I know this one was a little dense. I hope you found it useful.

Sadly, not everyone who could use the education is likely to get to the end, as @sarahcpr illustrates, hilariously bit.ly/3c9lIL0.

Back for Grand Rounds Thursday.