RR:C19 Evidence Scale rating by reviewer:
Potentially informative. The main claims made are not strongly justified by the methods and data, but may yield some insight. The results and conclusions of the study may resemble those from the hypothetical ideal study, but there is substantial room for doubt. Decision-makers should consider this evidence only with a thorough understanding of its weaknesses, alongside other evidence and theory. Decision-makers should not consider this actionable, unless the weaknesses are clearly understood and there is other theory and evidence to further support it.
This paper provides an analytical study of different contact tracing protocols, including overall quarantine duration and a "test-and-release" approach, and computes the "utility" of each by taking the ratio of the proportion of infections prevented and the number of days spent in quarantine. The authors also provide an excellent online webapp to allow users to further explore their findings.
While this study was nicely executed, in my view it relies heavily on assumptions that are hard to defend. These are described in more detail below. However, the main issues are: (1) the assumption of a much lower RT-PCR test sensitivity rate than what some (although not all) studies have reported; (2) the attempt to define a "utility" function and thus determine an "optimal" quarantine duration, without properly accounting for the utility disparity between preventing infections (extremely high utility) and preventing additional days spent in quarantine (smaller utility); and (3) ignoring relative prevalence rates, which should come into the utility calculation and which invalidates the returning traveler analysis.
p. 2, "the low sensitivity of PCR"—the authors rely very heavily on Kucirka et al. on this point, but other studies have found vastly different results. Consider Richardson et al. (10.1001/jama.2020.6775) for example, who found only 3% of patients had a negative first test and positive repeat test, implying a sensitivity of 97%. A review of these studies (including many of low quality) was conducted by Arevalo-Rodriguez et al. (10.1101/2020.04.16.20066787).
p. 4, At risk of complicating an already excellent figure, it would be helpful if Fig. 1B could show the latent period (duration between infection and infectiousness) and the pre-symptomatic period, which are major additional points of confusion. Note that the generation time is sometimes used to refer to the interval between the onset of infectiousness, not infections themselves (https://nccid.ca/publications/glossary-terms-infectious-disease-modelling-proposal-consistent-language/).
p. 7, The utility metric defined is indeed "one possible metric." However, it is not necessarily an appropriate one, as the authors briefly note in the discussion (p. 16). A single transmission prevented is worth far more than one day in quarantine saved. It is difficult to determine an "optimal" value since the two quantities (infections/deaths on one hand and time spent in quarantine) are so incomparable. In practice, the optimal ratio would depend on the current prevalence: if elimination has been achieved (e.g. as it has been in New Zealand), then it is imperative to prevent as close to 100% of transmissions as possible; in other locations (e.g. the US), the utility of individual-level quarantine should be compared against broader societal mobility restrictions and lockdowns. It is also not correct to say (p. 15) that quarantine durations longer than 10 days have "no extra benefit."
p. 7, It does not seem reasonable to assume that all symptomatic individuals immediately self-quarantine with 100% efficacy—rather, symptomatic, asymptomatic, and pre-symptomatic transmission are generally thought to comprise roughly equal proportions of all transmissions.
p. 8, The assumption of a 10-day upper limit of quarantine effectiveness is highly dependent on the distribution in Fig. 2B. A sensitivity analysis exploring the impact of heavier-tailed distributions might be worthwhile.
p. 10, Antigen tests have lower sensitivity and specificity than PCR, but still very high, on the order of 97%: see e.g. https://www.fda.gov/media/141570/download.
p. 12, The analysis of returning travelers does not properly take into account the risk of infection. A traveler who has spent one day in a high-prevalence location has a lower risk of being infected than a traveler who has spent three days. Importantly, it seems the point has been missed that the risk of infection on the final day of travel is constant, and hence travel duration is unimportant. For example, consider returning from a location where the traveler was exposed to a 1% per day risk of infection (which is much higher than is realistic). For a 1-day trip, their total risk of infection is 1% and their risk of being infected on the last day is 1%. For a 10-day trip, their total risk of infection is 1-(1-0.01)^10 = 9.6%, and their risk of being infected on the last day is 0.91%—almost identical to the 1.0% risk of being infected on a 1-day trip.