CoChat
Sign Up
Cited by 1,129Advanced Causal Inference TechniquesStatistical Methods in EpidemiologyMeta-analysis and systematic reviews

The DOI matches its Crossref record; 81 of the 81 references it cites were found in a registry or an index; none of the 93 sources its claims cite is checked for support yet.

Triangulation in aetiological epidemiology

Open access
Lawlor, Debbie A., Tilling, Kate, Davey Smith, George·2017·International Journal of Epidemiology
DOI 10.1093/ije/dyw314PMC PMC5841843
Open this paper in CoChat — free accountOpen PDF

Abstract

Triangulation is the practice of obtaining more reliable answers to research questions through integrating results from several different approaches, where each approach has different key sources of potential bias that are unrelated to each other. With respect to causal questions in aetiological epidemiology, if the results of different approaches all point to the same conclusion, this strengthens confidence in the finding. This is particularly the case when the key sources of bias of some of the approaches would predict that findings would point in opposite directions if they were due to such biases. Where there are inconsistencies, understanding the key sources of bias of each approach can help to identify what further research is required to address the causal question. The aim of this paper is to illustrate how triangulation might be used to improve causal inference in aetiological epidemiology. We propose a minimum set of criteria for use in triangulation in aetiological epidemiology, summarize the key sources of bias of several approaches and describe how these might be integrated within a triangulation framework. We emphasize the importance of being explicit about the expected direction of bias within each approach, whenever this is possible, and seeking to identify approaches that would be expected to bias the true causal effect in different directions. We also note the importance, when comparing results, of taking account of differences in the duration and timing of exposures. We provide three examples to illustrate these points.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This paper defines triangulation in aetiological epidemiology as integrating results from several approaches with different, largely unrelated key sources of bias, and proposes minimum criteria for its valid use. It illustrates the framework with three examples—systolic blood pressure and CHD, maternal gestational glucose and birthweight, and breastfeeding and later BMI—showing how explicit bias direction and exposure timing/duration affect causal interpretation.

Key findings

  • The authors propose five minimum criteria: results from at least two (ideally more) approaches with differing and unrelated key biases are compared; approaches address the same causal question; exposure duration and timing are taken into account; key biases are explicitly acknowledged; and expected bias directions are made explicit where feasible, ideally with some biases in opposite directions.
  • Table 1 summarizes key sources of bias for conventional and additional approaches, including RCTs, multivariable regression, cross-context comparisons, different control groups, natural experiments, within-sibship comparisons, instrumental variable/Mendelian randomization analyses, and negative control studies. The authors argue that applying multiple statistical methods to the same dataset is not triangulation because key biases such as residual confounding and measurement error remain shared.
  • Example 1 found lower systolic blood pressure reduced CHD risk across RCT, prospective cohort, and MR approaches, with magnitudes greatest for MR, intermediate for prospective cohorts, and least for RCTs; after accounting for duration of exposure, results were broadly consistent. The example cites 958,074 adults in 61 prospective studies, 25 RCTs with 109,797 participants, and MR data from 22,223 CHD cases and 64,762 controls.
  • Example 2 found positive associations of maternal fasting glucose with birthweight; IV RCT and MR estimates were both >300 g greater birthweight per 1 mmol/l greater fasting glucose, compared with ∼115 g in fully adjusted multivariable analyses, and the authors concluded more evidence is needed. Example 3 found most approaches, including a large RCT, pointed to no meaningful causal association between breastfeeding and later BMI, while negative control outcomes suggested residual confounding; within-sibship analyses were too small to provide reliable estimates.

Methods

  • This is a methodological/conceptual paper that proposes a definition and criteria for triangulation, summarizes key biases of several epidemiological approaches in Table 1, and illustrates the framework with three examples.
  • The examples draw on published meta-analyses, cohort studies, RCTs, and MR studies, plus some de novo analyses, including additional MR sensitivity analyses for systolic blood pressure and an ALSPAC negative control study for breastfeeding.
  • Comparisons are qualitative: for each example, the authors tabulate key biases and expected directions and consider differences in exposure duration and timing; they do not quantitatively pool results.

Limitations

  • The paper states that it does not consider how to pool data in a triangulation framework to produce a quantified causal effect estimate, nor does it consider effect modification beyond modification by age when considering timing of exposure; triangulation will mostly provide a qualitative assessment.
  • The paper states that increased use of triangulation likely requires multidisciplinary collaboration, more extensive data sharing, and additional resources, and that the illustrative examples were chosen to be “messy” rather than proof of principle.
  • Evident from the text: the examples are selected from the authors’ research interests rather than systematically sampled, some included studies were underpowered (e.g., within-sibship breastfeeding analyses), and judgments about expected bias directions rely on assumptions that may be uncertain.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1093/ije/dyw314

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 of 93 cited sources checked · 0 supported · 0 flagged

Show all 59 claims

  • It may state two separate claims as one. 43
  • Probably not a substantial claim. 52
78 of 78

  • The authors state this as an assumption.

    The key sources of bias of different approaches used in a triangulation are largely unrelated to each other.

    “where each approach has different (and assumed to be largely unrelated) key sources of potential bias.”

  • The authors state this as an assumption.

    In a triangulation using multivariable regression, RCT, and MR, bias from unmeasured confounders, failure to examine effects in randomized groups or subversion of randomization, and population stratification are not related to each other.

    “we would assume in a triangulation study using these approaches that bias from unmeasured confounders, failure to examine effects in randomized groups or subversion of the randomization, and population stratification in an MR study would not be related to each other.”

  • The authors state this as an assumption.

    The negative control exposure does not have a causal effect on the real outcome.

    “The negative control exposure does not have a causal effect on the real outcome.”

  • The authors state this as an assumption.

    The key sources of bias, including specific confounders, misclassification bias and other biases, are the same for the real and negative control exposures.

    “The key sources of bias, including specific confounders, misclassification bias and other biases, are the same for the real and negative control exposures.”

  • The authors state this as an assumption.

    The IV is not associated with confounders of the exposure-outcome association.

    “IV is not associated with confounders of exposure-outcome association.”

  • The authors state this as an assumption.

    The IV is not related to the outcome other than via its association with the exposure (the exclusion restriction criteria).

    “IV is not related to the outcome other than via its association with the exposure (the exclusion restriction criteria)”

  • The authors state this as an assumption.

    The populations being compared in natural experiments are similar with the exception of the naturally or quasi-randomized exposure.

    “The populations being compared are similar with the exception of the naturally or quasi-randomized exposure”

  • The authors state this as an assumption.

    There is little or no individual-level confounding in within sibling comparisons.

    “There is little or no individual-level confounding.”

  • The authors state this as an assumption.

    Different results between populations in cross-context comparisons are due to different confounding structures and not due to true differences in causal effects between populations.

    “Different results between populations are due to different confounding structures and not due to true differences in causal effects between populations.”

  • The authors state this as an assumption.

    Similar results between populations cannot be explained by confounding, given the differences between the populations/contexts in their confounding structures.

    “Similar results between populations cannot be explained by confounding, given the differences between the populations/contexts in their confounding structures.”

  • The authors state this as an assumption.

    There are no other (than confounding) sources of bias that could explain similar or different results between the two populations.

    “There are no other (than confounding) sources of bias that could explain similar or different results between the two populations”

  • The authors state this as an assumption.

    The different sources of bias for the different control groups are different and would produce different results.

    “The different sources of bias for the different control groups are different and would produce different results”

  • The authors state this as an assumption.

    Confounding structures differ between the different contexts/populations in cross-context comparisons.

    “under the a priori assumption that confounding structures differ between the different contexts/populations.21”

  • The paper relies on this without support.

    For many exposures, greater duration of exposure results in greater risk.

    “For many exposures, greater duration of exposure will result in greater risk.”

  • The authors state this as an assumption.

    In prospective cohort studies, exposure duration is from baseline assessment to end of follow-up.

    “(assuming, as in Ference et al.,53 that exposure duration in a prospective cohort is from baseline assessment to end of follow-up)”
15 of 15

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Blood-pressure discussion points to the glucose table

    Minor

    The cited Table 3 covers maternal glucose and birthweight, not the systolic blood pressure and coronary heart disease results discussed in this sentence.

  • Pregnancy exposure is described in months instead of weeks

    Minor

    The same discussion says the cohort measurements represent exposure during the last 12-14 weeks of gestation, but later calls that interval the last 12-14 months of pregnancy.

2 of 2

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. Vanderweele TJ. Explanation in Causal Inference. New York, NY: Oxford University Press, 2015.
  2. Rothman KJ Greenland S Lash TL. Modern Epidemiology. 3d edn. Philadelphia, PA: Lippincott Williams & Wilkins, 2008.

Cited by

Sorted by citations

See all 1,129

Related

See all

Metadata

Authors
Lawlor, D. A., Tilling, K., & Davey Smith, G.
Affiliations
University of Bristol (GB)
Corresponding author
Lawlor, Debbie A.
ORCID
Venue
International Journal of Epidemiology
Publisher
Oxford University Press (OUP)
Published
2017-01-20
Volume
45
Issue
6
Pages
dyw314
ISSN
  • 0300-5771 (ISSN-L)
  • 1464-3685
DOI
10.1093/ije/dyw314
PMC
PMC5841843
OpenAlex
W2579779633
Semantic Scholar
8ed819773b0e480d384cecb6fa7b55c5eb1c55de
Cited by
1,129
References
81
Citations by year
2026: 71 · 2025: 177 · 2024: 153 · 2023: 154 · 2022: 131 · 2021: 124 · 2020: 124 · 2019: 107 · 2018: 60 · 2017: 24
FWCI
65.73
Type
journal article
Language
en
Open access
Locations
  • International Journal of Epidemiology (publishedVersion)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Explore Bristol Research (publishedVersion)
Funders
  • Harvard University
  • National Institute for Health and Care Research (NF-SI-0611-10196)
  • European Commission (2007–2013, FP7 2007-2013, FP7/2007, 669545)
  • University of Bristol (MC_UU_12013/1, MC_UU_12013/9, MC_UU_12013/5)
  • Universiteit Leiden
  • Medical Research Council (MC_UU_12013_5, MC_UU_12013, MCUU12013/1, MC_PC_15018, MC_UU_12013/9)
  • H2020 European Research Council (669545)
Topics
  • Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
  • Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods in Epidemiology
  • Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › Meta-analysis and systematic reviews
Keywords
triangulation, causal inference, bias, confounding, causal effect estimation, exposure duration, exposure timing, research design comparison
MeSH terms
Data Collection, Epidemiology, Humans, Research Design, Bias, Causality, Mendelian Randomization Analysis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review