Commentary

Swanson, Sonja A; Hernán, Miguel A.

doi:10.1097/ede.0b013e31828d0590

Cited by 153 publications

(95 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…If investigators discover lack of identification in an applied context, they should consider not providing a point estimate for the causal effect and simply reporting the test for association between the IV and the outcome to establish a causal effect of the exposure on the outcome (which should be reported in any case) (36). If they consider using an alternative estimation technique, such as a fully parametric method, it should be made clear that this makes use of stronger untestable assumptions in order to obtain identification.…”

Section: Discussionmentioning

confidence: 99%

Lack of Identification in Semiparametric Instrumental Variable Models With Binary Outcomes

Burgess¹,

Granell²,

Palmer³

et al. 2014

American Journal of Epidemiology

View full text Add to dashboard Cite

A parameter in a statistical model is identified if its value can be uniquely determined from the distribution of the observable data. We consider the context of an instrumental variable analysis with a binary outcome for estimating a causal risk ratio. The semiparametric generalized method of moments and structural mean model frameworks use estimating equations for parameter estimation. In this paper, we demonstrate that lack of identification can occur in either of these frameworks, especially if the instrument is weak. In particular, the estimating equations may have no solution or multiple solutions. We investigate the relationship between the strength of the instrument and the proportion of simulated data sets for which there is a unique solution of the estimating equations. We see that this proportion does not appear to depend greatly on the sample size, particularly for weak instruments (ρ2 ≤ 0.01). Poor identification was observed in a considerable proportion of simulated data sets for instruments explaining up to 10% of the variance in the exposure with sample sizes up to 1 million. In an applied example considering the causal effect of body mass index (weight (kg)/height (m)2) on the probability of early menarche, estimates and standard errors from an automated optimization routine were misleading.

show abstract

Section: Discussionmentioning

confidence: 99%

Lack of Identification in Semiparametric Instrumental Variable Models With Binary Outcomes

Burgess¹,

Granell²,

Palmer³

et al. 2014

American Journal of Epidemiology

View full text Add to dashboard Cite

show abstract

“…The Instrumental Variable Checklist 51 (Supplementary Table 3, available as Supplementary Data at IJE online) was used for MR analysis and the instrumental variable flow chart considered throughout analyses and reporting 52 …”

Section: Methodsmentioning

confidence: 99%

Is smoking heaviness causally associated with alcohol use? A Mendelian randomization study in four European cohorts

Taylor

Rode

Bjørngaard

et al. 2018

International Journal of Epidemiology

View full text Add to dashboard Cite

BackgroundObservational studies have shown that tobacco and alcohol use co-occur, but it is not clear whether this relationship is causal.MethodsUsing data from the Avon Longitudinal Study of Parents and Children (ALSPAC) and UK Biobank, we used observational methods to test the hypothesis that smoking heaviness increases alcohol consumption. Mendelian randomization (MR) analyses were then used to test the causal relationship between smoking heaviness and alcohol consumption using 55 967 smokers from four European studies [ALSPAC, The Nord-Trøndelag Health Study (HUNT), the Copenhagen General Population Study (CGPS) and UK Biobank]. MR analyses used rs1051730/rs16969968 as a genetic proxy for smoking heaviness.ResultsObservational results provided evidence of an association between cigarettes per day and weekly alcohol consumption (increase in units of alcohol per additional cigarette smoked per day = 0.10, 95% confidence interval (CI) 0.05 to 0.15, P ≤ 0.001 in ALSPAC; and 0.48, 95% CI 0.45 to 0.52, P ≤ 0.001 in UK Biobank). However, there was little evidence for an association between rs1051730/rs16969968 and units of alcohol consumed per week across ALSPAC, HUNT, CGPS and UK Biobank (standard deviation increase in units of alcohol per additional copy of the risk allele = –0.004, 95% CI –0.023 to 0.016, P=0.708, I2 = 51.9%). We had 99% and 88% power to detect a change of 0.03 and 0.02 standard deviation units of alcohol per additional copy of the risk allele, respectively.ConclusionsPreviously reported associations between smoking and alcohol are unlikely to be causal, and may be the result of confounding and/or reverse causation. This has implications for public health research and intervention research.

show abstract

“…A strong version of the homogeneity assumption is that the causal effect of the exposure on the outcome has the same magnitude in all individuals [8]. A weaker version is that there is no additive effect modification by the instrumental variable at different values of the exposure [23].…”

Section: Additional Assumptions For the Estimation Of A Causal Effectmentioning

confidence: 99%

“…This is analogous to performing an intention-to-treat analysis in a randomized trial [7]. The motivation for this is that the claim of a causal effect requires fewer assumptions than the estimation of a causal effect [8,9], and the magnitude of the causal estimate is of secondary importance, as the quantitative effect of intervening on the exposure in practice is likely to differ from the causal estimand of the instrumental variable analysis [10]. For example, the effect of reducing low-density lipoprotein cholesterol (LDL-c) on coronary heart disease (CHD) risk by taking statin drugs depends on the choice of statin, the dosage (amount and frequency), the duration of treatment, the patient group, and so on [11].…”

Section: Introductionmentioning

confidence: 99%

Predicting the Direction of Causal Effect Based on an Instrumental Variable Analysis: A Cautionary Tale

Burgess

Small

2016

Journal of Causal Inference

View full text Add to dashboard Cite

An instrumental variable can be used to test the causal null hypothesis that an exposure has no causal effect on the outcome, by assessing the association between the instrumental variable and the outcome. Under additional assumptions, an instrumental variable can be used to estimate the magnitude of causal effect of the exposure on the outcome. In this paper, we investigate whether these additional assumptions are necessary in order to predict the direction of the causal effect, based on the direction of association between the instrumental variable and the outcome, or equivalently based on the standard (Wald) instrumental variable estimate. We demonstrate by counterexample that if these additional assumptions (such as monotonicity of the instrument-exposure association) are not satisfied, then the instrumental variable-outcome association can be in the opposite direction to the causal effect for all individuals in the population. Although such scenarios are unlikely, in most cases, a definite conclusion about the direction of causal effect requires similar assumptions to those required to estimate a causal effect.

show abstract

Commentary

Cited by 153 publications

References 21 publications

Lack of Identification in Semiparametric Instrumental Variable Models With Binary Outcomes

Lack of Identification in Semiparametric Instrumental Variable Models With Binary Outcomes

Is smoking heaviness causally associated with alcohol use? A Mendelian randomization study in four European cohorts

Predicting the Direction of Causal Effect Based on an Instrumental Variable Analysis: A Cautionary Tale

Contact Info

Product

Resources

About