Study population and setting
The investigators sought to measure an association between hydroxychloroquine (HCQ) use and in-hospital mortality in a retrospective cohort of consecutive adult patients admitted with COVID-19 to a 6-hospital health system in Michigan. Patients were included if they were admitted for at least 48 hours and diagnosed with SARS-CoV-2 from March 10 to May 2, 2020. COVID-19 treatments were protocol driven. The following independent variables were abstracted: age, gender, race, preexisting medical conditions, mSOFA (modified sequential organ failure assessment) on admission (25% were missing the mSOFA), O2 saturation on admission, receipt of glucocorticoids, receipt of IL-6 pathway agents, ICU admission, and mechanical ventilation. The investigators categorized patients into four treatment groups for analysis: (1) HCQ, (2) HCQ plus azithromycin, (3) azithromycin alone, and (4) neither HCQ nor azithromycin. They used multivariable Cox proportional hazards regression models to calculate hazard ratios adjusted using the abstracted data. The investigators also used the available data to match 1:1 HCQ and no HCQ patients for repeat Cox proportional hazards modeling. During the study period there were 2,948 COVID-19 admissions; 286 (10%) were excluded due to not being discharged, transfer, or leaving against medical advice, leaving 2541 patients for analysis. The median length of hospitalization was 6 days (4-10), and median follow-up time was 28 days. The majority of patients received HCQ (1202 (47%) HCQ alone and 783 (31%) HCQ plus azithromycin). A total of 409 (16%) patients received neither medication. The median time from admission to HCQ initiation was 1 day (interquartile range 1-2).
Summary of Main Findings
In-hospital mortality in this cohort was 18.1%. Among patients receiving HCQ alone, the mortality was 13.5%, for HCQ plus azithromycin it was 20%, for azithromycin alone it was 22%, and for neither drug it was 26%. There were substantial differences between the groups, with the non-HCQ group generally having a greater proportion of characteristics associated with higher mortality. These included older age (median 68 years compared to 63 years among HCQ only), a lower proportion of Black race (Black race was associated with lower mortality in this cohort), and a substantial imbalance in glucocorticoid use: 78% among patients receiving HCQ compared to 35% among those who received neither medication received a glucocorticoid. In addition, only 8% of those non-HCQ received mechanical ventilation compared with 13% among HCQ alone and 29% among HCQ plus azithromycin. Based on the Kaplan-Meier curves, although not explicitly described in the manuscript, it appears that approximately 30% of the deaths occurred in the first 3 days after admission for those not receiving HCQ while few deaths occurred during this time period among those who received HCQ.
This study includes a large sample size, systematically collected data, and a heterogeneous population.
This study had substantial limitations. For example, the inclusion of time prior to HCQ use for estimating survival benefit for HCQ rather than using time dependent variables would favor the HCQ arms because people receiving HCQ had to survive long enough to receive it (which was >2 days for 25% of the HCQ patients). Also, although the study team used multivariable regression methods and equate the groups on a set of observed characteristics, there is a strong potential for additional, unobserved confounders. Perhaps most importantly, unmeasured patient factors likely impacted the decision to treat with HCQ. Very few details are provided regarding how treatment decisions were made expect that a standardized protocol was used. The use of a standardized protocol suggests that patient who did not receive HCQ differed in important ways from those who did. In addition, those who received HCQ had to survive in order to receive HCQ. In particular, the high mortality, low number of ICU admissions, and high prevalence of multiple factors associated with higher mortality seen in the non-HCQ group all suggest that HCQ was preferentially prescribed to less sick patients. It is plausible that decisions were made for palliative care for some individuals, leading to neither admission to the ICU nor use of HCQ, and thus further contributing to an apparent beneficial effect of HCQ.
This study reinforces the association between age and mortality and uses a relatively large dataset of a diverse set of COVID-19 patients.
This review was posted on: 15 July 2020