For people interested in college policies re: COVID in the fall, I strongly recommend reading this preprint article draft analyzing the effects of different test regimes: https://www.medrxiv.org/content/10.1101/2020.07.06.20147702v1.full.pdf
The authors (people at the Yale School of Public Health) model different frequencies of universal student screening for COVID-19, with different assumptions about how bad the overall situation is (varying Rt values, along with varying numbers of cases introduced from external sources). In their base case, assuming a Rt of 2.5 and 5 externally introduced cases per week in a population of 5000 residential students, they conclude that the optimal strategy would be to test students every 2 days with a relatively cheap, relatively low sensitivity test (i.e. relatively large numbers of “misses,” negative results for people who are actually infected). They assume that all positive cases are moved to an isolation dorm, where they are cared for and no spread is permitted (e.g. no contact whatsoever with non-isolated students for the duration of infectiousness). Daily testing reduces case numbers further, but they argue that it fails a cost-benefit test; testing every 3 days increases case numbers and fails the cost-benefit test in the other direction. Weekly testing results in close to uncontrolled spread, with approximately 70% of the students infected by the end of the semester.
In the worst case scenario (Rt of 3.5 and 25 externally introduced cases per 14 days), daily testing is cost-effective and can adequately control infection — holding total infections down to less than 5% of the student body, only somewhat higher than the base case with testing every other day. However, at those assumptions about spread, testing every every 3 days fails almost completely (60% total infections), and testing every other day doubles the number of cases and fails a cost-benefit analysis.
In the best case scenario, weekly testing is adequate and the optimal cost-benefit strategy.
Other notable findings: sensitivity is not very important (how likely tests are to correctly identify infected cases), and it’s cost-effective to use cheaper, less sensitive tests more frequently than more expensive, higher sensitivity tests. However, specificity (the likelihood that a positive result is not a false positive) is extremely important — even modest declines in specificity create big problems, and those probably understate the case because they don’t take into account things like the morale effect and changing willingness to participate when people get frequent false positives.
We shouldn’t rely on a single model in our decision-making, and we need to incorporate other data. Also, I’m not an epidemiologist — I can understand and interpret this model, but I’m not qualified to evaluate its quality. But here’s my take-aways if we assume this model is accurate enough: We have no business opening any university to on-campus education without at least testing of every student every third day, sufficient isolation dormitory space to isolate all positive students, and a willingness to increase testing to daily if there are signs of spread. (The authors didn’t model approaches where testing frequency is varied based on testing results, but it’s an obvious way of dealing with uncertainty.). Even with strong efforts to hold down Rt (mask-wearing, social distancing, frequent hand-washing, etc.), the risks in a university setting are high — Rt is likely to be higher in that context than in the general population. That means that we need to plan around a base Rt before testing and isolation that’s higher than the Rt in a non-university context.
Not to put too fine a point on it, but if this article is close to accurate, Western Michigan University’s plan to reopen with large numbers of students on campus and no testing of non-symptomatic students is unconscionable and extraordinarily likely to fail. Applying something like the base case assumptions, with no non-symptomatic testing, would predict that out of the 20,000 or so students, roughly 14,000 students would be infected over an 80 day semester — and about 7 would die. More realistically, without regular testing of all students, it is very likely that a substantial outbreak would occur, resulting in the university switching to online education only and sending all students home (and dispersing their infections into their home communities). If other efforts to model this produce similar results, universities must engage in frequent testing of all students — or must not have students on campus. There’s no room for relying on testing only symptomatic students.