Limitations and Misinterpretations of E-Values for Sensitivity Analyses of Observational Studies

Ioannidis, John P. A.; Tan, Yuan; Blum, Manuel R.

doi:10.7326/m18-2159

Cited by 73 publications

(72 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Determining a potential unmeasured confounder is therefore not straightforward, nor is finding the relevant literature with which to assess the degree of potential confounding. Furthermore, the E-value is not without its critics ( Ioannidis et al, 2019 ). Another challenging aspect of the tool is the assessment of small study bias in the presence of heterogeneity ( Peters et al, 2010 ).…”

Section: Discussionmentioning

confidence: 99%

“…Instead of taking a certain value of the relative risk as the cut-off point, it is reasonable to judge whether confounding could have easily influenced the pooled effect size found in the meta-analysis. To this end, the application of the E-value approach is helpful ( VanderWeele and Ding, 2017 , Haneuse et al, 2019 , Ioannidis et al, 2019 , VanderWeele et al, 2019 ). This statistic is based on an assessment of how easily unmeasured confounders could explain away the relationship found between the exposure and the health outcome.…”

Section: By: the Who Global Air Quality Guidelines Working Group On Cmentioning

confidence: 99%

See 1 more Smart Citation

Long-term exposure to NO2 and O3 and all-cause and respiratory mortality: A systematic review and meta-analysis

Huangfu

Atkinson

2020

Environment International

317

155

View full text Add to dashboard Cite

Highlights 46 cohort studies assessed long-term concentrations of NO2 and O3 and mortality. Meta-analysis of 24 studies found increased risk of death associated with NO2. Weak associations were observed for peak period O3 and mortality. High levels of heterogeneity were observed. Certainty in NO2 associations with mortality was generally low/moderate.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: By: the Who Global Air Quality Guidelines Working Group On Cmentioning

confidence: 99%

Long-term exposure to NO2 and O3 and all-cause and respiratory mortality: A systematic review and meta-analysis

Huangfu

Atkinson

2020

Environment International

317

155

View full text Add to dashboard Cite

show abstract

“…rather, it represents a hypothetical effect assessed in a sensitivity analysis. 51 The same RR will always generate the same E-value, regardless of the study question. Therefore, in addition to estimating E-values, we also conducted a sensitivity analysis for the metaanalysis to assess the magnitude of a bias factor and the minimum confounding strength for both the exposure and the outcome that would be required to explain the association.…”

Section: Limitations Of the Datamentioning

confidence: 99%

Acetaminophen use during pregnancy and the risk of attention deficit hyperactivity disorder: A causal association or bias?

Masarwa

Platt

Filion

2020

Paediatric Perinatal Epid

View full text Add to dashboard Cite

Background:The association between acetaminophen use during pregnancy and the development of attention deficit hyperactivity disorder (ADHD) in the offspring may be due to bias. Objectives:The primary objective was to assess the role of potential unmeasured confounding in the estimation of the association between acetaminophen use during pregnancy and the risk of ADHD, through bias analysis. The secondary objective was to assess the roles of selection bias and exposure misclassification. Data sources:We searched MEDLINE, Embase, Scopus, and the Cochrane Library up to December 2018. Study selection and data extraction:We included observational studies examining the association between acetaminophen use during pregnancy and the risk of ADHD. Synthesis:We meta-analysed data across studies, using random-effects model. We conducted a bias analysis to studies that did not adjust for important confounders, to explore systematic errors related to unmeasured confounding, selection bias, and exposure misclassification. Results:The search resulted in seven studies included in our meta-analysis. When adjusted estimates were pooled across all studies, the risk ratio (RR) for ADHD was 1.35 (95% confidence interval [CI] 1.25, 1.46; I 2 = 48%). Sensitivity analysis for unmeasured confounding in this meta-analysis showed that a confounder of 1.69 on the RR scale would reduce to 10% the proportion of studies with a true effect size of RR >1.10. Unmeasured confounding bias analysis decreased the point estimate in five of the seven studies and increased in two studies, suggesting that the observed association could be confounded by parental ADHD. Unadjusted and bias-corrected risk ratios (bcRRs) were: RR = 1.34, bcRR = 1.13; RR = 1.51, bcRR = 1.17; RR = 1.63, bcRR = 1.38; RR = 1.44, bcRR = 1.17; RR = 1.16, bcRR = 1.18; RR = 1.25, bcRR = 1.05;and RR = 0.99, bcRR = 1.18. Conclusions:Bias analysis suggests that the previously reported association between acetaminophen use during pregnancy and an increased risk of ADHD in the offspring may be due to unmeasured confounding. Our ability to conclude a causal association is limited. K E Y W O R D Sacetaminophen, ADHD, bias, confounding, pregnancy

show abstract

“…Confounding is a major problem in observational research. Methods, such as computation of E-values (minimum strength of association that an unmeasured confounder would need to have with both treatment and outcome to fully explain away a specific treatment-outcome association, conditional on the measured covariates), have been proposed but have caveats 12. Despite the adjustment of the analysis to various available factors that might influence prescribing patterns, we could not take all of them into account.…”

Section: Discussionmentioning

confidence: 99%

Association between gifts from pharmaceutical companies to French general practitioners and their drug prescribing patterns in 2016: retrospective study using the French Transparency in Healthcare and National Health Data System databases

Goupil

Balusson

Naudet

et al. 2019

BMJ

View full text Add to dashboard Cite

ObjectiveTo evaluate the association between gifts from pharmaceutical companies to French general practitioners (GPs) and their drug prescribing patterns.DesignRetrospective study using data from two French databases (National Health Data System, managed by the French National Health Insurance system, and Transparency in Healthcare).SettingPrimary care, France.Participants41 257 GPs who in 2016 worked exclusively in the private sector and had at least five registered patients. The GPs were divided into six groups according to the monetary value of the received gifts reported by pharmaceutical, medical device, and other health related companies in the Transparency in Healthcare database.Main outcome measuresThe main outcome measures were the amount reimbursed by the French National Health Insurance for drug prescriptions per visit (to the practice or at home) and 11 drug prescription efficiency indicators used by the National Health Insurance to calculate the performance related financial incentives of the doctors. Doctor and patient characteristics were used as adjustment variables. The significance threshold was 0.001 for statistical analyses.ResultsThe amount reimbursed by the National Health Insurance for drug prescriptions per visit was lower in the GP group with no gifts reported in the Transparency in Healthcare database in 2016 and since its launch in 2013 (no gift group) compared with the GP groups with at least one gift in 2016 (−€5.33 (99.9% confidence interval −€6.99 to −€3.66) compared with the GP group with gifts valued at €1000 or more reported in 2016) (P<0.001). The no gift group also more frequently prescribed generic antibiotics (2.17%, 1.47% to 2.88% compared with the ≥€1000 group), antihypertensives (4.24%, 3.72% to 4.77% compared with the ≥€1000 group), and statins (12.14%, 11.03% to 13.26% compared with the ≥€1000 group) than GPs with at least one gift between 2013 and 2016 (P<0.001). The no gift group also prescribed fewer benzodiazepines for more than 12 weeks (−0.68%, −1.13% to −0.23% compared with the €240-€999 group) and vasodilators (−0.15%, −0.28% to −0.03% compared with the ≥€1000 group) than GPs with gifts valued at €240 or more reported in 2016, and more angiotensin converting enzyme (ACE) inhibitors compared with all ACE and sartan prescriptions (1.67%, 0.62% to 2.71%) compared with GPs with gifts valued at €1000 or more reported in 2016 (P<0.001). Differences were not significant for the prescription of aspirin and generic antidepressants and generic proton pump inhibitors.ConclusionThe findings suggest that French GPs who do not receive gifts from pharmaceutical companies have better drug prescription efficiency indicators and less costly drug prescriptions than GPs who receive gifts. This observational study is susceptible to residual confounding and therefore no causal relation can be concluded.Trial registrationOSF register OSF.IO/8M3QR.

show abstract

Limitations and Misinterpretations of E-Values for Sensitivity Analyses of Observational Studies

Cited by 73 publications

References 21 publications

Long-term exposure to NO2 and O3 and all-cause and respiratory mortality: A systematic review and meta-analysis

Long-term exposure to NO2 and O3 and all-cause and respiratory mortality: A systematic review and meta-analysis

Acetaminophen use during pregnancy and the risk of attention deficit hyperactivity disorder: A causal association or bias?

Association between gifts from pharmaceutical companies to French general practitioners and their drug prescribing patterns in 2016: retrospective study using the French Transparency in Healthcare and National Health Data System databases

Contact Info

Product

Resources

About