The Case Against Percent Positivity

This post centers on how to interpret data published on the IDPH Site, media sites, and other sources. Later this week, I will post a series of guidelines on exposure and case management in light of recent changes issued by CDC.

What’s with all the differences in data?

There are many pulling, interpreting, and publishing data, including local media, with erroneous conclusions. The problem– the state releases data in two ways; 1) real-time counts of Total Tested and Positive updated throughout each day (these are pulled from the Totals row on the IDPH site), 2) counts by date continuously updated retrospectively for Tested and Positive (bar graphs within the IDPH site).

The media and others tracking on their own pick a time each day to update their numbers by simply marking the change in counts. The issue with this is the counts for previous days are never adjusted, the percent positive is inflated, and the totals quickly run out of alignment with the state. See illustration below – depending on when you pull the data, the counts are SIGNIFICANTLY different.

It takes a while to assign positive tests and tests performed to individual dates, but over time this is a more appropriate way to view the data. The problem, you have to be patient and the data are too dated for accurate and timely decision-making.

What do we do?

Watch the curve – mainly the 14d rolling total and trend on the NYT site. I like the Times because their trend is a 7-day, but even they have not completely addressed the data lag issue. https://www.nytimes.com/interactive/2020/us/iowa-coronavirus-cases.html#cases

What about school metrics?

There is no doubt using percent positive as a decision-making metric is problematic. The lag in assigning tests to days greatly impacts the percent positive sometimes showing a 15% day nearly two weeks after it occurs. I am in complete disagreement with the state on the incredibly high bar set as well as the duration at which that has to be sustained.

What is the alternative?

I’m becoming a big fan of cases per population. Here’s how this works:

• You have 490,161 people in Polk County.

• There was a total of 621 cases the last 7 days in Polk County.

• Rate of total new cases per 100,000 = 621/(490,161/100,000) = 127

The White House Task Force wants us under 100. We haven’t been there since June 30th.

The DMPS perspective:

• You have 490,161 people in Polk County.

• There was an average of 89 new cases each day the last 7 days.

•  Rate of new daily cases per 100,000 = 89/(490,161/100,000) = 18

DMPS wants us down to 1. Well, we haven’t been there since April 7th, so that’s probably not going to happen for quite some time.

Why the per 100,000 metric?

The denominator stays STABLE. The number of daily tests performed is matching, and at times outpacing, the number of positives. The result – the percent positive no longer detects surges in cases. See second graphic – if the number tested remained high, but stable, the last surge in cases would have been reflected (dashed line) in the percent positive. Since we substantially and continuously increase testing capacity, the percent positive did not detect or reflect the surge (solid line).

What do we do?

We keep increasing testing. Ultimately that finds cases which is a priority. We advocate for use of a different metric for keeping schools safe, though one more reasonable than proposed by DMPS. We watch our cases per 100,000 and continue to be vigilant.

Stay safe and healthy!