Table 3 Considerations relevant to interpretation of the results of gene-association studies for the risk of ARDS

Question Considerations References
What is the prior probability of the gene association? Pathophysiology of ARDS is very complex [1]
Severity of ARDS likely also depends on the inciting cause (e.g., pathogen) and its duration prior to appropriate treatment [15]
Many gene-association studies are the first to examine for an association between the particular genes with the development of ARDS in a hot field of interest [12, 13]
It may be unknown if the SNPs are associated with changes in their respective protein levels or function (e.g., there is no change in the amino acid sequence of the CELF-2 protein) [2]
The rationale for exploration for a gene association is often based on limited prior information (e.g., the current study was done because of the previous finding, in one study with the same cohort, that CFTR gene variants are associated with ARDS in African American children with CAP) [3, 13, 14]
What is the minimum BF observed, or the PPV? The minimum BF for a P = 0.03 and P = 0.01 are 0.095 and 0.036, respectively. This modifies the prior: even if very high (e.g., prior of 0.25 for the alternative hypothesis), the null hypothesis probability is lowered to no less than 22% or 10%, respectively [11]
The PPV, assuming a prior of 0.01, power 0.80, and P values between 0.001 to 0.01, is between 55–90%; assuming a more realistic prior of 0.001, the PPV is 10–50% [14]
Given the expected odds ratios of gene association for complex diseases such as ARDS are <1.5, and the often low number of patients with ARDS, the power of the study was well below 0.5. This lowers the PPV even more [12, 15]
How much bias occurred? Attrition bias: not all the cohort has genotyping done. In this example:
-The number of patients is often lower in more recent times
[15]
Selection bias: flexibility in eligibility criteria due to different definitions of conditions. In this example:
-CAP definition required at least two of tachypnea, dyspnea, or hypoxemia, but in one study the list also included cough or abnormal chest exam (selection bias)
[6, 12, 13, 15]
Analysis and reporting biases: flexibility in definitions of predictor variables, in decisions of which covariates to adjust for, in decisions of outcomes to examine, and in which analyses to report. In this example: [12, 13, 15]
a. Predictor variables definitions: only in the CFTR studies was age categorized as <11 versus ≥11 years; the cohorts were analyzed separately, although in other studies African Americans and Caucasians were combined or only results for African Americans were reported [2, 3, 4, 6, 7]
b. Covariates to adjust for: in the one study where it was examined, a statistically significant association was “lost” if adjusted for bacterial culture positivity; asthma was forced into the multiple regressions in only this most recent study; other known SNPs were not adjusted for (especially those the same group previously found associated with ARDS: IL-1ra A1 allele, SP-B); duration of symptoms prior to enrolment, treatment (e.g., antibiotics), and duration of treatment are not reported or adjusted for [2, 4, 6]
c. Outcomes to determine and report: the outcome was ARDS, but in other studies also included mechanical ventilation, severe sepsis, vasopressor use, and renal or hematologic dysfunction, and could include hospital admission, PICU admission, and duration of hospitalization [3, 47]
1. ARDS acute respiratory distress syndrome, BF Bayes factor, CAP community acquired pneumonia, CELF-2 elav-like family member 2, CFTR cystic fibrosis transmembrane conductance regulator, IL1-ra interleukin 1 receptor antagonist, PPV positive predictive value, SNP single nucleotide polymorphism, SP-B surfactant protein B