Hacker News Reader

Comment by timr | Hacker News Reader

> 2 was published in Feb 2020, so not sure how helpful that is...

It's a meta-analysis of many other papers. The estimates for the other ILIs have not changed significantly in 2020. Again, this is nitpicking that misses the forest for the trees: the differences in estimated R0 between these viruses is not so precise as to be useful for claiming the effectiveness of something like "masks and distancing".

> 1 is not normalized data. It does not account for testing frequency

This is incorrect. The data is normalized across all samples. Your hospital's non-participation may change the coverage of the sample, but that doesn't really matter when it's just showing you that rhinovirus hasn't gone away.

View on HN

In the paper associated with 1 it says:

The FilmArray RP test utilization rate (TUR) metric is defined as the non-normalized number of RP patient test results generated each week across the Trend sites (computed as a centered 3-week moving average). To calculate the pathogen detection rate (as displayed in Figure 2 [second data view] and on the Trend website), we compute the rate for each organism at each institution as a centered 3-week moving average. To adjust for the capacity differences between sites, a national aggregate is calculated as the unweighted average of individual site rates. Only data from sites contributing more than 30 tests per week is included to avoid noise from small numbers of tests. Because the calculation of pathogen detection rate includes results from patients with multiple detections, the detection rate for all organisms can, in theory, add up to greater than one. In practice, this does not occur.

These are rates. This does not account for testing frequency. If they only tested 1 person in the United States tomorrow, and they had rhino/entero, the graph would show 100% (obviously it wouldn't and they won't only test 1 person). That means that if testing decreased, because for example there were less patients with symptomatic viral infections, but the % positive stayed the same, the chart would not appear to change, even though there were fewer cases...

paper: https://publichealth.jmir.org/2018/3/e59/

The TUR is not the metric plotted on the website. Read the very next sentence in the section you're quoting:

> To calculate the pathogen detection rate (as displayed in Figure 2 [second data view] and on the Trend website), we compute the rate for each organism at each institution as a centered 3-week moving average. To adjust for the capacity differences between sites, a national aggregate is calculated as the unweighted average of individual site rates. Only data from sites contributing more than 30 tests per week is included to avoid noise from small numbers of tests. Because the calculation of pathogen detection rate includes results from patients with multiple detections, the detection rate for all organisms can, in theory, add up to greater than one. In practice, this does not occur.

The TUR is the the raw number of hits across all sites. They normalize this, which is why they talk about it possibly exceeding 1.0. It's why the Y-axis of the plot is labeled in percentages. The height of the bars is a moving average of detection rates across their network.

Regardless, this is still nitpicking. The point was that these other pathogens have not gone away. They have not. Even if you were right, it wouldn't change the argument.

Reply on news.ycombinator.com