Boulware study says HCQ doesn't work. Boulware is smart, but I get a certain biased vibe from him. Not sure, though.
After high-risk or moderate-risk exposure to Covid-19, hydroxychloroquine did not prevent illness compatible with Covid-19 or confirmed infection when used as postexposure prophylaxis within 4 days after exposure.
nejm.org
But these scientists say that Boulware's conclusions do not follow from his data.
- In the final results, Boulware included patients who never started taking HCQ in order. The effect is to mask HCQ's positive benefit. And Boulware wrote up his analysis to obscure that fact.
- Another statistical game Boulware plays is comparing randomized adherent group with non-randomized non-adherent group. Again, the effect is to mask HCQ's positive benefit.
Patients who never started treatment kept in analysis
Claudia N. Paiva, Daniel V. TauskUFRJ, USP7 September 2020
We read the article intitled “Hydroxychloroquine in nonhospitalized adults with COVID-19”, by Skipper et al, with much interest. We disagree with some of the authors’ conclusions, which we believe could not be taken from the data shown.
(1) In suppl table 3, concerning medication adherence, the sum of all categories in HCQ group is: 100% of the tablets, 157; 75-99% of the tablets, 8; <75% of the tablets, 16; never started medication, 22, so = 157+8+16+22= 203. In placebo group, the sum of these categories is 166+3+12+13=194. In suppl table 2, concerning the analysis of symptoms severity according to medication adherence, table states that in HCQ group, 165 patients took >75% of the tablets (15-19) and 38 <75% (1-14), 165+38= 203. In placebo group, 169+25= 194. Comparing these two tables, it becomes clear that in suppl table 2 if category <75% was in fact 1-14 tablets, we should be missing 22 patients in group HCQ and 13 patients in placebo group, who never started medication (0 tablets). Since we are not missing them, in both tables group HCQ has 203 patients and group placebo has 194 patients, I assume the author kept patients who never started medication in HCQ and placebo groups. It is not a question of writing 0-14 instead of 1-14 tablets; it is about excluding patients who did not start treatment from ITT analysis. The “1-14 tablets” induces the reader to assume that authors performed the usual analysis, excluding patients who did not start treatment. We recall that even in intention-to-treat analysis, if no treatment was applied at all, the usual procedure is to exclude these patients from the full analysis set. clinfo.eu
(2) We acknowledge that ignoring those patients who stopped the medicine because of side effects introduces biases. However, it does not apply to this situation, since these 22 patients who never started HCQ treatment never experienced any side effects and there were also 13 patients who never started placebo; they did not start treatment for reasons not related to group allocation. While no biases might arise from excluding patients who never started treatment allocated to both HCQ and placebo groups, keeping them in the full analysis set might mask any positive HCQ effects. Masking the positive effects of a drug during a pandemic is a harmful choice.
(3) If a medication is found to work under optimal conditions in the trial setting, the adherence to this medication in the field will be far better. We recall that the uncertainty about the efficacy of a drug being tested in the trial setting may mean that trial participants are less likely to persevere with unpleasant side effects of treatment than those who have been assured of the efficacy of their treatment (1). Yet the authors seem to disregard their own HCQ positive results found when patients taking >75% of the tablets are considered, a significant 19.5% relative benefit in symptom severity (two-tailed p-value=0.022).
(4) Authors “caution against over-interpretation of this result”, a 19.5% relative benefit in symptom severity, arguing that “this [adherent HCQ] group did not improve any more than the non-adherent hydroxychloroquine group nor the non-adherent placebo group”. However, unlike comparison between treated and untreated group, the comparison between adherent and non-adherent group is invalid as the latter groups are not selected by randomization and are subject to “non-adherence bias”, i.e., non-adherent patients might have not taken the drug for instance because they were healthier than adherent patients. This is corroborated by the data in the table which shows that the small non-adherent group did overall much better than the larger adherent group.
(5) We understand that the proper way to analyze the data is by doing both an analysis of the intention-to-treat population (excluding, as usual, patients who took zero tablets) and a per protocol analysis of the patients with 75% adherence to the treatment.
The consequence of these considerations is that conclusions such as the one in the abstract and discussion, “hydroxychloroquine did not substantially reduce symptom severity in outpatients with early, mild COVID-19”, cannot be assumed when a possible positive effect might arise from the rightful exclusion of patients who never started treatment and there is a clear 19.5% effect of hydroxychloroquine over placebo when patients who took the tablets are compared, i.e., per-protocol analysis.
acpjournals.org
Tom |
