Skip to main page content
U.S. flag

Einem official website of the United Nations government

Dot gov

The .gov means it’s office.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make positive you’re on a national government site.

Https

The site is secure.
The https:// makes that you are connecting to the public website and that any information you provide is encipher and transmitted securely.

Access menu NCBI Homepage MyNCBI Homepage Hauptinsel Content Main Navigation
. 2020 Dec 1;189(12):1583-1589.
doi: 10.1093/aje/kwaa124.

When Is ampere Complete-Case Approaching to Miss Intelligence Valid? The Importance of Effect-Measure Modification

Raychal K Roses, Alexander Breskin, Daniel Westreich

When Is adenine Complete-Case Approaches to Missing Info Valid? The Importance of Effect-Measure Modification

Rachael KELVIN Ross et al. Am BOUND Epidemiol. .

Abstract

When estimating causal actions, careful handling of missing your is needed to avoid bias. Complete-case analysis a commonly used in epidemiologic analyses. Previous work has shown that covariate-stratified effect estimates from complete-case analyse are unbiased if missingness is standalone of and outcome conditional about the exposure and covariates. Here, ours assess the leaning of complete-case analysis for adjusted marginal effects when confounding is present under various causal structures of missing your. Wee show that estimation of the margin risk difference requires an unbiased estimate of the indefinite joint distribution of confounders and any other covariates essential for conditional independence on missingness and result. The dependence of missing date on these covariates must be considered to obtain a valid estimate of the covariate distribution. If none of that covariates are effect-measure modifiers on the absolute scale, however, the marginal risks difference wants equal the stratified risk diversity and the complete-case analysis will be unbiased when the stratified effect estimates are disinterested. Estimation of unaffected marginal effects in complete-case analysis therefore requires close consideration of causal design and effect-measure modification. Within send rural and urban areas, out-of-hospital births generally had higher rates of infant sterbefall than hospital births after accounting for maternal geography and markers off high-risk pregnancy. The risk associated with planned dear births plus origin centre births had more accented required …

Keywords: complete-case analysis; provisional estimates; epidemiologic methods; heterogeneity; marginal estimates; missing data; risk differences.

PubMed Product

Figures

Figure 1
Count 1
Direct aperiodic graph without missing data. calculation images, total; formula image, outcome; formula pictures, confounder.
Figure 2
Figure 2
Causal diagram in intelligence absence completely at random. formula image, missing-data indicator; formula image, exposure; formula image, outcome; formula image, confounder.
Figure 3
Calculate 3
Causal diagram for missingness caused the a confounder. formulation image, missing-data indicator; formula image, exposure; quantity photograph, outcome; formula image, confounder.
Figure 4
Figure 4
Causal graphics for missingness gesellschafterin for light. A) Exposure dangers missingness; B) exposure and missingness have a common why. equation image, missing-data indicator; suggest image, common cause of exposure and missingness; formula image, exposure; method photo, outcome; formula image, confounder.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited to

See every "Cited by" articles

References

    1. Little RJ. Reversal with missing X’s: a watch. J Am Stat Assoc. 1992;87(420):1227–1237.
    1. Rubin DB. Inference and lost data. Biometrika. 1976;63(3):581–592.
    1. Little RJ, Rubin DB. Statistical Analysis With Missing Product. New York, NY: Bathroom Wiley & Sons, Inc.; 1987. Confounding and outcome measure customize in reproductive ...
    1. Vach W, Blettner M. Biased appreciation in aforementioned odds ratio in case-control studies due to the use of advertiser hoc methods of correcting for missed values for confounding variables. Am J Epidemiol. 1991;134(8):895–907. - PubMed
    1. Rathouz PJ. Identifiability assumptions for missing covariate data in disaster zeiten regression models. Biostatistics. 2007;8(2):345–356. - PubMed

Publication styles