key: cord-0841559-dzt9zwvz
authors: Keller, Kevin; Sussman, Jeremy
title: Chronologic Bias, Confounding by Indication, and COVID-19 Care
date: 2021-07-07
journal: Chest
DOI: 10.1016/j.chest.2021.01.087
sha: 7c8a446af308540bc3676dc21087628c989f4d99
doc_id: 841559
cord_uid: dzt9zwvz

nan

To the Editor:

In the battle against COVID-19, scientists all over the world are doing their best to fight. From January 24, 2020, when the SARS-CoV-2 cases were first reported, 1 to today (February 16, 2021), more than 100,000 related articles have been published. These scientific discoveries have enabled us to better understand our enemies.

In the research article published in CHEST, 2 

deserve praise for their study, which to date is the highest quality evidence to evaluate the use of ivermectin in patients with this disease.

Propensity score matching, like other adjustment techniques, can only account for between-group differences that are included in the propensity score itself. 2 One possible variable that the authors themselves raise in their discussion, but did not adjust for, is "timing bias" or chronologic bias. The authors state "more of the control group was enrolled in the first weeks of the study." If care changed in other ways at the same time ivermectin became the norm in the authors' hospital, then the outcomes could be ascribed falsely to ivermectin. Nationally available data have shown declining in-hospital mortality rates during this time period. 3 Unlike most design flaws, chronologic bias could be tested for simply by adding date of admission to the propensity score. If this makes matching impossible, then chronologic bias becomes likely. We hope the authors consider this analysis.

Further, the unusually common administration of ivermectin to admitted patients during this timeframe consecutively, particularly later in the study, suggests that ivermectin was effectively the standard of care at these sites and implies that patients who did not receive it may have differed systematically in other, unmeasured ways. This is a form of confounding by indication, is more statistically intractable, and may have also led to misleading results during the early period of the pandemic with anticoagulation 4 and hydroxychloroquine 5 for hospitalized patients with COVID-19.

To the Editor:

We appreciate the thoughtful comments of Drs Keller and Sussman. Per their suggestion, we reran the propensity match, adding admission date to the variables for propensity scoring performed in the original article (age, sex, pulmonary condition, hypertension, HIV, severe pulmonary presentation, exposure to corticosteroids, race, WBC count, absolute lymphocyte count, and the need for mechanical ventilation prior to or on the day of study entry). 1 We are in agreement with their second point regarding unmeasured cofounders that are, by definition, not something that can be corrected for without a randomized design. We believe our findings remain very compelling and support the continued need for a welldesigned randomized study. chestjournal.org

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Use of ivermectin is associated with lower mortality in hospitalized patients with Coronavirus disease 2019: the ivermectin in COVID nineteen study

Using propensity score methods to create target populations in observational clinical research

Variation in US hospital mortality rates for patients admitted with COVID-19 during the first 6 months of the pandemic

Association of treatment dose anticoagulation with in-hospital survival among hospitalized patients with COVID-19

Treatment with hydroxychloroquine, azithromycin, and combination in patients hospitalized with COVID-19