umu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Sensitivity analysis for unobserved confounding of direct and indirect effects using uncertainty intervals
Umeå University, Faculty of Social Sciences, Umeå School of Business and Economics (USBE), Statistics. (Stat4Reg)ORCID iD: 0000-0002-4600-0060
Umeå University, Faculty of Social Sciences, Umeå School of Business and Economics (USBE), Statistics. (Stat4Reg)ORCID iD: 0000-0003-3187-1987
Umeå University, Faculty of Social Sciences, Umeå School of Business and Economics (USBE), Statistics. (Stat4Reg)ORCID iD: 0000-0003-3298-1555
2018 (English)In: Statistics in Medicine, ISSN 0277-6715, E-ISSN 1097-0258, Vol. 37, no 10, p. 1744-1762Article in journal (Refereed) Published
Abstract [en]

To estimate direct and indirect effects of an exposure on an outcome from observed data, strong assumptions about unconfoundedness are required. Since these assumptions cannot be tested using the observed data, a mediation analysis should always be accompanied by a sensitivity analysis of the resulting estimates. In this article, we propose a sensitivity analysis method for parametric estimation of direct and indirect effects when the exposure, mediator, and outcome are all binary. The sensitivity parameters consist of the correlations between the error terms of the exposure, mediator, and outcome models. These correlations are incorporated into the estimation of the model parameters and identification sets are then obtained for the direct and indirect effects for a range of plausible correlation values. We take the sampling variability into account through the construction of uncertainty intervals. The proposed method is able to assess sensitivity to both mediator‐outcome confounding and confounding involving the exposure. To illustrate the method, we apply it to a mediation study based on the data from the Swedish Stroke Register (Riksstroke). An R package that implements the proposed method is available.

Place, publisher, year, edition, pages
John Wiley & Sons, 2018. Vol. 37, no 10, p. 1744-1762
Keywords [en]
direct effects, indirect effects, mediation, sensitivity analysis, sequential ignorability, unmeasured confounding
National Category
Probability Theory and Statistics
Research subject
Statistics
Identifiers
URN: urn:nbn:se:umu:diva-125929DOI: 10.1002/sim.7620ISI: 000429730500011PubMedID: 29462839OAI: oai:DiVA.org:umu-125929DiVA, id: diva2:973915
Note

First published in thesis 2016 in manuscript form.

Available from: 2016-09-23 Created: 2016-09-22 Last updated: 2018-06-07Bibliographically approved
In thesis
1. Statistical methods for register based studies with applications to stroke
Open this publication in new window or tab >>Statistical methods for register based studies with applications to stroke
2016 (English)Doctoral thesis, comprehensive summary (Other academic)
Alternative title[sv]
Statistiska metoder för registerbaserade studier med tillämpningar på stroke
Abstract [en]

This thesis adds to the area of register based research, with a particular focus on health care quality and (in)equality. Contributions are made to the areas of hospital performance benchmarking, mediation analysis, and regression when the outcome variable is limited, with applications related to Riksstroke (the Swedish stroke register).

An important part of quality assurance is to identify, follow up, and understand the mechanisms of inequalities in outcome and/or care between different population groups. The first paper of the thesis uses Riksstroke data to investigate socioeconomic differences in survival during different time periods after stroke. The second paper focuses on differences in performance between hospitals, illustrating the diagnostic properties of a method for benchmarking hospital performance and highlighting the importance of balancing clinical relevance and the statistical evidence level used.

Understanding the mechanisms behind observed differences is a complicated but important issue. In mediation analysis the goal is to investigate the causal mechanisms behind an effect by decomposing it into direct and indirect components. Estimation of direct and indirect effects relies on untestable assumptions and a mediation analysis should be accompanied by an analysis of how sensitive the results are to violations of these assumptions. The third paper proposes a sensitivity analysis method for mediation analysis based on binary probit regression. This is then applied to a mediation study based on Riksstroke data.

Data registration is not always complete and sometimes data on a variable are unavailable above or below some value. This is referred to as censoring or truncation, depending on the extent to which data are missing. The final two papers of the thesis are concerned with the estimation of linear regression models for limited outcome variables. The fourth paper presents a software implementation of three semi-parametric estimators of truncated linear regression models. The fifth paper extends the sensitivity analysis method proposed in the third paper to continuous outcomes and mediators, and situations where the outcome is truncated or censored.

Place, publisher, year, edition, pages
Umeå: Umeå universitet, 2016. p. 30
Series
Statistical studies, ISSN 1100-8989 ; 49
Keywords
Registers, quality of care, socioeconomic status, hospital performance, stroke, mediation, sensitivity analysis, truncation, censoring
National Category
Probability Theory and Statistics
Research subject
Statistics
Identifiers
urn:nbn:se:umu:diva-125953 (URN)978-91-7601-553-7 (ISBN)
Public defence
2016-10-21, Hörsal E, Humanisthuset, Umeå, 09:30 (English)
Opponent
Supervisors
Available from: 2016-09-30 Created: 2016-09-23 Last updated: 2018-06-07Bibliographically approved

Open Access in DiVA

fulltext(932 kB)8 downloads
File information
File name FULLTEXT01.pdfFile size 932 kBChecksum SHA-512
b64059ba94e16cf125172cdf1d873b9e907863e05a1278dd24e54a01088f12245d56ce0dc432b350fd550c724d082512c989bac27f5b8bf0aa8273b1d092cb75
Type fulltextMimetype application/pdf

Other links

Publisher's full textPubMedarXiv

Authority records BETA

Lindmark, Anitade Luna, XavierEriksson, Marie

Search in DiVA

By author/editor
Lindmark, Anitade Luna, XavierEriksson, Marie
By organisation
Statistics
In the same journal
Statistics in Medicine
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 8 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

doi
pubmed
urn-nbn

Altmetric score

doi
pubmed
urn-nbn
Total: 162 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf