1. I'm a strong advocate of testing as an important pillar of the coronavirus pandemic response.

There is often confusion about what testing offers, and I think it's helpful to distinguish between three roles that testing can play.

2. First, *Testing for Personal Health*. If you present with symptoms, your doctor will want to know if have COVID-19 (even though treatment options are limited) and you want to know so you can avoid possibly transmitting to your friends and family.

3. Here it's fairly straightforward who should be tested. Ideally, anyone who suspects that they have the virus should be able to access a test quickly, efficiently, and at zero or minimal cost. That test should yield rapid results, overnight at worst, near-instantaneous at best.

4. RT-PCR or antigen testing is what we want for personal health testing, because one can detect infection by these methods sooner than by antibody test. There may be personal health reasons to test for antibodies, though until we more more about protection those seem limited.

5. Second, we have *Testing for Surveillance*. To set health policy and adjust to changing circumstances, epidemiologists and public health officials need to how prevalent the virus is within a community, and how prevalence is changing over time.

6. Here it would be ideal to have a mix of sampling schemes. It would be good to know what fraction of symptomatic individuals have the virus, but we also want to know about asymptomatic carriers. Random sampling from the population at large is difficult but would be valuable.

7. Both RT-PCR and antibody testing are useful for surveillance. The former helps us track prevalence (number concurrently infected) and the latter incidence (total number infected since the start of the pandemic). Both are important to know for public health planning.

8. The third role of testing is *Testing for Mitigation*. Here the idea is to use frequent testing to identify those who are pre-symptomatic, asymptomatic, or only mildly symptomatic, so that we can ask them to self-isolate and thereby prevent them from transmitting the virus.

9. If we assume that people with strong symptoms will self-isolate anyway, testing for mitigation should target those who are not currently showing symptoms or who are showing mild symptoms. Individuals with high contact rates, e.g. front-line works, are particularly important.

10. Testing for mitigation should involve RT-PCR or antigen testing, because these will allow the earliest detection and thus allow us to remove those individuals who could be spreading disease from public interactions.

11. Some of the confusion around testing—how much do we need, and who should be getting these tests—arises from confusion about which mode we are discussing.

12. For example, testing for personal health requires those with strong symptoms be tested, whereas some models of testing for mitigation allocate tests only to those without strong symptoms—those with strong symptoms are presumed infected and asked to self-isolate regardless.

13. The necessary cadences of testing vary considerably as well. Testing for surveillance, it might be sufficient to test a small fraction of the population every day, and use the power of statistical statistical inference to draw conclusions about the population as a whole.

14. Testing for mitigation requires that a much larger fraction of the population be tested. The benefit of such testing is proportional to the number of infectious-days that can be eliminated by asking those without strong symptoms to self isolate.

15. If you test everyone in the community once a month, you will head off only a small fraction of the infectious days. If you test every two days, you'll catch patients early in the course of disease and ask people to isolate before they have been infectious in public for long.

16. Arguments about sensitivity versus speed and cost also often stem from misunderstandings about the purpose of testing. From a personal health perspective, having a test with a 50% false negative rate is absolutely unacceptable.

17. From a surveillance standpoint, 50% sensitivity would be ok so long as we can control for it.

Or when testing for mitigation, you just want to remove infectious people from the population. Better to test 100,000 people with 50% sensitivity than 50,000 with 70% sensitivity.

18. I've found it useful to be very clear about which role I am thinking of. Of course there is often overlap. Data from personal health testing is often used for surveillance purposes, for example. But I encourage people to be clear about aims when planning how to test.