Monday, September 21, 2020

Romer on testing

As part of an email conversation about testing, Paul Romer sent the following message. He so beautifully encapsulated the case for testing, I asked for permission to post his email. Here it is. 

Here is a short summary of the case for testing.

1. A program of "test and isolate" will reduce the effective reproduction number, R.

2. A combined policy of (i) more "test and isolate" which reduces R and (ii) more social interaction and more economic activity which increases R can be designed so that the net effect on R is zero.

3. The ratio of the cost of the additional testing to the additional economic activity that this combined policy will allow offers one way to estimate of the "rate or return" to spending on tests. My rough estimate is that this rate of return lies in the range of 10x to 100x so there is no doubt that test and isolate would be cost effective. To reach the higher end of the range, the cost of the test would have to be relatively low, say  $10.

4. The combined plan under #2 will lead to more total cases. If the main measure of policy success or failure were deaths, an increase in the number of cases would not matter. Under the current circumstances, an increase in the number of cases is likely to be interpreted as a sign of a policy failure. This increases the political cost to the administration of increasing the number of tests. A second-best solution that avoids this cost might be to use at home tests and encourage people to self-isolate. This way, the results from the tests need not generate any new confirmed cases.

[JC Comment: one reluctance that the president or governors may have is that more testing naturally produces more measured cases, and the media don't seem all that good about recognizing this fact.] 

Details on Targeting, Timing, and Compliance:

- Under the program in #3, the benefit created when more infectious people are isolated is received by unknown others who are free to resume normal activities. This is a classic case of an external effect. As a result, it makes sense for the government to pay for the tests and perhaps even to pay for "supported isolation" to increase the compliance rate. Because the fraction of the population that is infected is relatively small and because the required period of isolation is short, it would be relatively inexpensive to pay the few people who are in isolation, for example by making up any lost wages. Because transmission in the household is likely, it would make sense to offer a choice of isolation in a hotel or isolating the entire family at home. However, implementing this would require some way to confirm that someone is infectious, which precludes its use in the at-home approach noted under #4 above.

- For purposes of calculating the rate of return in the combined program described in #3, it is useful to consider a thought experiment of testing people at random. But in any practical program, the efficient way to use more tests is to start by targeting populations that have high ex ante probability of being infected. This could be done by concentrating the tests in high prevalence geographical regions, in high exposure populations, or on people identified by contact tracing. I am skeptical that contact tracing is the cost effective way to identify a large number of people who have a higher ex ante probability of being infected.

- For reducing R, what matters is the average number of infectious-person-days in isolation per test. This depends on (a) the number of true positives that are isolated and (b) when in the course of their infection they are isolated. The way to increase (a) is to target populations with a high ex ante probability of infection. The way to increase (b) is to use tests with a shorter time from sample to result.

- The choice between centralized lab testing and POC tests depends in part on an easily quantified tradeoff between a reduction in the sample-to-result time of most POC tests and a reduction in their sensitivity. But in the early months of any program for expanding the number of tests, the most important differentiator is likely to be the supply response. Many people are convinced that there is a large amount of lab capacity on university campuses that could rapidly be mobilized so that this path probably offers the lowest-cost path of expansion until manufacturing capacity increases for the POC or at home tests.

- There is a synergy between the frequency of testing in a population and the use of pooling to increase lab capacity. As the frequency increases, the frequency of positives will go down so that pooling becomes more cost effective.

-  A large fraction of the total cost of a test comes from the discomfort experienced by the person who gives the sample and the time it takes for a healthcare professional to collect the sample. On both grounds, saliva samples will almost surely have the lowest cost.

- To reduce the cost from isolating false positives, any initial positives could be retested. Because the number of positive results will be a small fraction of the number of tests, retesting adds only a small amount to the cost of the program.

- As long as any true positives are isolated, the net effect of the combined program described in #2 will be to increase the total amount of social interaction by people who are not infectious, even if there are some false positives.

I hope this is helpful. 

Paul. 

 

12 comments:

  1. I will say this as I have been all along. POC tests are not less sensitive than PCR lab tests for infectiousness (even if they are that way for infection). You can be infected and not infectious. This is a critical distinction. You are infectious if your viral load is high and quick antigen tests pick this up.

    Thus, they are MORE informative than PCR tests. There is no trade-off as claimed here. They dominate PCR tests on all fronts.

    ReplyDelete
  2. Since the prevalence of infection in the general population is .5%, a positive result with the Roche or Cepheid tests that have a likelihood ratio of 14 in an asymptomatic person will indicate actual infection in only 7% of cases. If you "isolate" these people, you will be isolating a population that is 93% normal.

    ReplyDelete
  3. Compliance is a problem. In some places, eg, Korea, it is socially acceptable to require isolated to load an app on their phones, and keep on and with them at all times. If the app stops communicating or is off location the cops roll up. The government delivers a large box of standard food each week and that's all ok.

    I don't see that could work in the US where the cultural mythologies prioritise the individual and demonise the government. And, a significant proportion of people apparently think the disease is a fabrication or at least a vehicle of government intrusion. Obviously, things in general could work a lot better if people were a bit more rational, but good luck with that.

    ReplyDelete
    Replies
    1. You personally wouldn't use it? If you had a party, you wouldn't get some for your guests?

      Delete
  4. yes, some people in the US have this really annoying habit of believing that their individual freedom is worth more than my freedom... It is sick. I am glad it is only a handful of them.

    ReplyDelete
  5. One issue keeps not getting raised----the age distribution of deaths. CDC releases this---keeping in mind their official description is "people who die WITH Covid". For example---about 60% over age 75---about 3% under 45. Is this taken into account in the Romer study? Why have we never seen (I have not, anyway) an analysis on how to optimize policy based on this highly unusual and predictive "death by age". If anything, we seem to have almost "Ezekiel Emmanuelized" our implementation---with Nursing homes dominating the deaths of the aged.

    I do not understand why this topic is not front and center

    ReplyDelete
  6. One issue keeps not getting raised----the age distribution of deaths. CDC releases this---keeping in mind their official description is "people who die WITH Covid". For example---about 60% over age 75---about 3% under 45. Is this taken into account in the Romer study? Why have we never seen (I have not, anyway) an analysis on how to optimize policy based on this highly unusual and predictive "death by age". If anything, we seem to have almost "Ezekiel Emmanuelized" our implementation---with Nursing homes dominating the deaths of the aged.

    I do not understand why this topic is not front and center

    ReplyDelete
    Replies
    1. Because it implies you are trying to go herd with the under 45s, which will not work. If you aren't going to test and isolate for two weeks everyone who comes up positive we aren't going to get the R numbers down by enough to have the virus die off. That puts EVERYONE at continued risk.

      I keep hearing this zombie idea everywhere, including my school district's board meetings. "It's OK if the kids get it, they'll be fine." Yes but completely misses the point that we are trying to extinguish the virus altogether. It's doable if we get the R numbers down. Low R is why we don't have the original SARS, Ebola, Small Pox, Plague and numerous others still raging around today.

      Delete
  7. Into a Paul Romer note Cochrane interjects: one reluctance that the president or governors may have is that more testing naturally produces more measured cases, and the media don't seem all that good about recognizing this fact.

    It is interesting to consider why Romer did not make this point himself, and why Cochrane felt he needed to.

    I'd argue the answer lies partly in deep differences of temperament and empathy. Paul Romer is truly gifted in taking on controversial topics in a diplomatic, non-inflammatory, non-polemical way. Cochrane (and Krugman) not so much.

    Perhaps "how would Paul Romer make this point?" is a question everyone should ask, prior to hitting the 'post' button.

    ReplyDelete
    Replies
    1. Sadly the truth is nowhere near as nefarious, or interesting. Paul emphasized the interjection to me in later follow up email, especially that politicians may not want testing because it drives up measured cases. I had not understood the point when first reading his letter. If I didn't see the point he was making, I thought perhaps I should help my readers to do so as well.

      Delete
    2. I though he did make this point in #4 bullet:

      "Under the current circumstances, an increase in the number of cases is likely to be interpreted as a sign of a policy failure. This increases the political cost to the administration of increasing the number of tests."

      Cochrane just used plainer language.

      I would argue a big part of the reason for the political costs is precisely because of the President's (and yes, the President, not the Admin) dismissive, inept response to the pandemic. All attempts at nuance about testing numbers and so on are going to get lost in the face of that.

      Delete
  8. John--please revisit your national VAT proposal/pieces in light of the comical outrage over Trump's taxes in the NYT!

    ReplyDelete

Comments are welcome. Keep it short, polite, and on topic.

Thanks to a few abusers I am now moderating comments. I welcome thoughtful disagreement. I will block comments with insulting or abusive language. I'm also blocking totally inane comments. Try to make some sense. I am much more likely to allow critical comments if you have the honesty and courage to use your real name.