Thread by @LizSpecht

While the main point of this article is distressing to me, I feel there's also an underappreciated point here regarding positivity rates. [short thread]
theatlantic.com/science/archiv…

Positivity rates (# positive results versus # of total tests) have been so important for determining how much of a spike in caseload is truly indicative of higher viral spread versus merely a reflection of more testing.

Positivity rates now exceed 10% in many states. 2/n

But a line from this article made my head spin because it seems to indicate that the situation may be worse than it first appears from those positivity rates.

Unlike in March, April, or May, we are now testing huge numbers of people who are very unlikely to test positive. 3/n

"In mid-June, four changes hit all at once." The first 3? "Large companies began to test their employees en masse, hospitals started to test every patient who needed an elective procedure, and nursing homes started regularly testing their employees and some residents." 4/n

Only the 4th change —"The American public also seemed to seek out voluntary tests in greater numbers this month."— is akin to the way in which tests were used prior to June: someone seeks out a test because they have reason to believe they've been exposed or are symptomatic. 5/n

Routine testing in workplaces is akin to random community surveillance. You'd expect very, very low positivity rate among those tests.

This may be drowning out the signal from people who are voluntarily seeking tests because they feel ill or think they have been exposed. 6/n

Would love to know what fraction of tests are routine workplace/hospital testing vs. voluntary. If the former is a large fraction, the positivity rate we're seeing could indicate we're in a much worse spot today than when positivity rates were identical a few months ago. 7/n

Example: if 50% of tests today are routine workplace/hospital and those are typically only 1% positive, then a 10% positivity rate overall means that positivity among the voluntary tests is actually 18%. 8/n

We should be comparing only the *voluntary testing* positivity rates to the rates of Mar/Apr/May if we want a *comparative* sense of how bad the community spread is today vs. then. Random workplace/hospital testing positivity rate gives an *objective* measure of the problem. 9/n

Both of these measures are important and informative. But these categories of testing seem like they shouldn't be bundled into a single value for positivity rate. It confounds the interpretation of how bad things are. /end

@alexismadrigal Have you seen any data regarding what fraction of tests are random workplace/hospital testing versus voluntary tests sought out by individuals? Seems to have importance for interpretation of positivity rates (see thread).