Dear Editor:
Gautret and colleagues [1] claim that hydroxychloroquine (HCQ) and azithromycin are efficacious in the treatment of patients infected with SARS-cov-2. However, their conclusion that hydroxychloroquine is associated with viral load reduction is not based on rigorous study design or analysis.
The authors tested whether there is a difference in the rate of negative RT-PCR tests in treated versus untreated patients in each day. This method is not optimal for the longitudinal study design, in which each patient has been repeatedly measured on multiple occasions. The serial measurements introduce the problem of intra-patient correlation in the outcome that can lead to biased estimate of treatment effect and incorrect P values [2]. Using the authors' data, I estimated that the intraclass correlation in the rate of negative tests was 0.37 (95% confidence interval: 0.22 to 0.52), suggesting a substantial within-patient correlation that needs to be accounted for.
The mixed-effects logistic regression model [3] can be used to model the change within each patient, and hence give an unbiased estimate of treatment effect. Using the authors' data in the appendix [1], I conducted a mixed-effects logistic regression analysis accounting for the interaction effects of HCQ and azithromycin and time (Table 1 ). While the rate of negative tests increased with time, the increase was not significantly different between those on HCQ and those not on HCQ (P = 0.355). Interestingly, patients on azithromycin had a significantly greater rate of negative tests compared to those not on azithromycin (P = 0.019; Figure 1 ). On the basis of the results presented, I propose an alternative interpretation that the effect of hydroxychloroquine and azithromycin on the elimination of viral load remains uncertain.
Table 1.
Effects of day, hydroxychloroquine (HCQ), and azithromycin on the rate of negative tests: result of a mixed-effects logistic regression model.
| Parameter | Estimate (Standard error) | P-value |
|---|---|---|
| Day | 0.45 (0.23) | 0.047 |
| HCQ | 1.71 (1.73) | 0.322 |
| Azithromycin | -5.09 (3.43) | 0.138 |
| HCQ by Day | 0.27 (0.30) | 0.355 |
| Azithromycin by Day | 2.72 (1.16) | 0.019 |
Note: Results were obtained from the model log(P / (1 - P)) = Day + HCQ + Azithromycin + HCQ x Day + Azithromycin x Day, where P is the probability of negative test.
Figure 1.
Predictive percent of negative tests by treatment group (HCQ or azithromycin) on each day. The interaction effect between HCQ and day was not statistically significant (P = 0.355), but the interaction effect between azithromycin and day was statistically significant (P = 0.019).
Declaration of Competing Interest
No conflict of interest
Acknowledgments
Funding
No funding
Ethical Approval obtained
NA
References
- 1.Gautret P, Lagierac JC, Parola P, Hoang TV, Meddeb L, Mailhe M. Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial. International Journal of Antimicrobial Agents. 2020;56 doi: 10.1016/j.ijantimicag.2020.105949. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
- 2.Liljequist D, Elfving B, Roaldsen KS. Intraclass correlation – A discussion and demonstration of basic features. PLoS ONE. 2019;14(7) doi: 10.1371/journal.pone.0219854. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Austin PC, Merlod J. Intermediate and advanced topics in multilevel logistic regression analysis. Statist Med. 2017;36:3257–3277. doi: 10.1002/sim.7336. [DOI] [PMC free article] [PubMed] [Google Scholar]