Results:
Inclusion and exclusion: Of the 2 799 patients hospitalised during the study period, we excluded 440 for the reasons described above. Of the 2 359 patients included, we selected those with a first PCR test result obtained within 48 hours of admission (1 294) and those who had a PCR within the first ten days of care (1 276) (Figure 1 ). Of them, 747 were PCR negative within 30 days of follow-up and 529 were censored at the date of their last positive PCR during follow-up (Table 1 ).
Comparison of treatment with HCQ versus no HCQ: The population analysed included 776 people who received hydroxychloroquine (HCQ) and 500 who did not receive hydroxychloroquine. Patients in the HCQ-treated group were significantly younger than those in the group not treated with HCQ, they had a longer time from symptom onset to treatment onset, and a lower baseline viral load (Table 2 ). It should be noted (see above) that these three factors were likely to affect viral clearance in favour to treatment. In the crude analysis, the time from treatment onset to viral clearance was significantly lower in the HCQ group than in the untreated group (Log-rank testp <.001) (Table 1 , Figure 2 ). For example, on D0+5 days, 50.0% (95%CI [46.0%-53.9%]) of patients in the HCQ group were still positive, compared to 63.2% (95%CI [58.4%-67.6%]) of patients in the non-HCQ group. At D0+10 days, 23.0% (95%CI [19.0%-27.1%]) of patients in the HCQ group were still positive, compared to 33.4% (95%CI [27.8%-39.2%]) in the non-HCQ group (Table 1 ). Overall, the probability of viral clearance was significantly higher in the group treated with HCQ (Hazard ratio 95% CI 1.39 [1.20–1.61], p <.0001).
When adjusted for age, initial viral load and time from symptom onset to treatment onset, which were potential confounding factors, the adjusted hazard ratio of viral clearance for the HCQ group remained statistically significant (Hazard ratio 95% CI 1.18 [1.01-1.38], p =.037), suggesting that the HCQ treatment had a significant impact upon the probability of viral clearance within 30 days of the onset of treatment (Table 3 ). We noted a decrease in the probability of negativization as age increases (Hazard ratios =1, 0.90, 0.72, 0.59, and 0.50 for patients aged <50, 50-59, 60-69, 70-79 and >79 respectively). When the CT of the first PCR increases by one unit, the probability of negativization increases by 12% (Hazard ratio=1.12) i.e. the lower the initial viral load, the greater the probability of a negative result. Finally, an increased likelihood of negativization was observed when the time to treatment was longer (longer time associated with a lower viral load at the time of treatment).
When treatment with both HCQ and AZT was compared to treatment with AZT alone, similar results were obtained, suggesting that the essential element in lowering the viral load is treatment with HCQ (supplementary data, Figure 1S, Table 1S, Table 2S ). The sensitivity analysis, using a competitive risk approach (death was considered a competing event), yielded similar results (supplementary data, Figure 2S, Table 3S, Table 4S ).