Where in this design would substantive theories of postulated therapeutic effects enter into consideration? Take for example the several theories advanced in [1,pp5–6]:
Findings from previous studies have suggested that chloroquine and hydroxychloroquine may inhibit the coronavirus through a series of steps. Firstly, the drugs can change the pH at the surface of the cell membrane and thus, inhibit the fusion of the virus to the cell membrane. It can also inhibit nucleic acid replication, glycosylation of viral proteins, virus assembly, new virus particle transport, virus release and other process to achieve its antiviral effects.
… and then later on pp.15–16:
In some patients it has been reported that their immune response to the SARS-CoV-2 virus results in the increase of cytokines IL-6 and IL-10. This may progress to a cytokine storm, followed by multi-organ failure and potentially death. Both hydroxychloroquine and chloroquine have immunomodulatory effects and can suppress the increase of immune factors[29, 30]. Bearing this in mind, it is possible that early treatment with either of the drugs may help prevent the progression of the disease to a critical, life-threatening state.
How would continuous learning proceed about the several mechanisms of action posited here? I have to think that incorporating more information via multivariate, time-series outcomes would be essential. As I read it, the design in its present form seems organized around a single, ordinal outcome measure as its central feature.
- Yao X, Ye F, Zhang M, et al. In Vitro Antiviral Activity and Projection of Optimized Dosing Design of Hydroxychloroquine for the Treatment of Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Clinical Infectious Diseases. March 2020:ciaa237. doi:10.1093/cid/ciaa237
A March 23 paper  came to my attention this morning, which illustrates some of the opportunities for obtaining quantitative time-series data on the course of illness. N=23 patients were followed serially, with a mean of 7.5 respiratory specimens on which reverse transcriptase quantitative PCR (RT-qPCR) was done to quantify SARS-CoV-2 RNA. Moreover, serology for IgG and IgM antibodies to the virus surface spike receptor binding domain (RBD) and internal nucleoprotein (NP) were used to (semi?)quantitate antibody responses. The paper claimed, e.g., that initial viral load was high initially, and declined steadily. This suggests an opportunity to model the slope of log10 copies/mL RNA—a chance to estimate rather than hypothesis-test, which should please some of you! Furthermore, the onset of antibody response offers up a time-to-event outcome that might be worth modeling. (But on that latter point, please note that the IgG response and its timing may have a subtler—spline-able?—relation to outcomes as discussed at the top left-hand column on p.9 including with correlation to macaque experimental studies. This thread by virologist Akiko Iwasaki has a nice further explanation of the point made there.)
- To KK-W, Tsang OT-Y, Leung W-S, et al. Temporal profiles of viral load in posterior oropharyngeal saliva samples and serum antibody responses during infection by SARS-CoV-2: an observational cohort study. The Lancet Infectious Diseases. March 2020:S1473309920301961. doi:10.1016/S1473-3099(20)30196-1