Collider Bias Rather Than a Healthy Condition Leads to the Unfavorable Outcome of Sepsis

Abstract

We applaud Alrawashdeh et al1Alrawashdeh M. Klompas M. Simpson S.Q. et al.Prevalence and outcomes of previously healthy adults among patients hospitalized with community-onset sepsis.Chest. 2022; 162: 101-110Abstract Full Text Full Text PDF Scopus (3) Google Scholar in this issue of CHEST for using three large administrative databases to investigate the impact of previously healthy conditions and comorbidities on sepsis outcomes. However, the unexpected finding that previously healthy patients were at greater risk of dying of sepsis than were patients with comorbidities caught our attention. The authors suggested that several characteristics associated with previously healthy patients could explain the counterintuitive finding, including delayed medical attention, more sepsis with unknown sources, late diagnosis of organ dysfunction, or genetic predisposition. This counterintuitive finding may be due to collider bias in the study design, which may lead to misleading results. Despite the dangers of collider bias, the authors failed to adequately discuss them. Collider bias is a form of selection bias that originates from study design as a result of conditioning on a common effect of two exposure variables.2Hernán M.A. Hernández-Díaz S. Robins J.M. A structural approach to selection bias.Epidemiology. 2004; 15: 615-625Crossref PubMed Scopus (1527) Google Scholar In this study, sepsis was defined by Centers for Disease Control and Prevention (CDC) Adult Sepsis Event Surveillance criteria, which requires a blood culture order plus four antibiotic days to define severe infection. A blood culture or four days of antibiotic therapy were prescribed more commonly to patients with comorbidities than to previously healthy patients who presented with signs of sepsis. Consequently, the CDC sepsis criteria actually acted as a common effect of two exposure variables (comorbid vs healthy condition), which is termed a “collider” in a causal diagram due to the collision of two arrows (Fig 1). Selecting patients with sepsis based on CDC criteria is conditioning on the collider and thereby creates a collider bias. In other words, CDC criteria affected the likelihood of an individual being sampled among two groups of interest, thereby inducing spurious associations between these variables in the sample. Many paradoxic findings in the medical literature can be explained by collider bias,3Griffith G.J. Morris T.T. Tudball M.J. et al.Collider bias undermines our understanding of COVID-19 disease risk and severity.Nat Commun. 2020; 11: 5749Crossref PubMed Scopus (233) Google Scholar including a finding that diabetes mellitus risks were lower among obese children compared with children of normal weight.4Stovitz S.D. Banack H.R. Kaufman J.S. Paediatric obesity appears to lower the risk of diabetes if selection bias is ignored.J Epidemiol Community Health. 2018; 72: 302-330Crossref Scopus (6) Google Scholar This was due to the selection of study participants based on the results of glycated hemoglobin (HbA1c) testing, in which HbA1c testing is the common effect of two groups of children with different body weights. Collider bias is difficult to detect at the analysis stage. Neither regression models nor propensity score matching can correct for collider bias; instead, it must be corrected through study design in general. We hope that our analysis will provide useful information for the development of future clinical studies on sepsis. Funding/support: This work was supported by Far Eastern Memorial Hospital, Taiwan (Grant FEMH-2022-C-051). Financial/nonfinancial disclosures: None declared. Role of sponsors: No funding bodies had any role in the study design, data collection and analysis, the decision to publish, or the preparation of the manuscript. Prevalence and Outcomes of Previously Healthy Adults Among Patients Hospitalized With Community-Onset SepsisCHESTVol. 162Issue 1PreviewThe vast majority of patients hospitalized with community-onset sepsis harbor pre-existing comorbidities. However, previously healthy patients may be more likely to die when they seek treatment at the hospital with sepsis compared with patients with comorbidities. These findings underscore the importance of early sepsis recognition and treatment for all patients. Full-Text PDF ResponseCHESTVol. 162Issue 1PreviewWe thank Lee et al for their interest in our study that suggests higher short-term mortality rates in previously healthy individuals hospitalized with community-onset sepsis compared with those with comorbidities and raises collider bias as a potential explanation.1 We agree this is an important consideration and in fact pointed out the same issue through a clinical lens that Lee et al labeled using a statistical term. Specifically, we discussed that providers may be more concerned about possible sepsis in patients with comorbidities and thus may be more likely to order blood cultures and administer antibiotics (thereby triggering Center for Disease Control and Prevention Adult Sepsis Event [ASE] criteria) for less severe illnesses compared with healthy patients. Full-Text PDF