Bayesian correction for covariate measurement error: A frequentist evaluation and comparison with regression calibration

Bartlett, Jonathan; Keogh, Ruth H.

doi:10.1177/0962280216667764

Cited by 39 publications

(38 citation statements)

References 40 publications

(102 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We’ve proposed a framework which is robust to heteroscedastic variance while maintaining ease of interpretation for use by clinicians and other non-statisticians. As well, the implementation of a Bayesian framework negates the issue of a biased estimate for the ICC, as Bayesian estimates do not rely on closed-form approximations and normal distribution asymptotic theory [ 32 , 42 ].…”

Section: Discussionmentioning

confidence: 99%

“…This is particularly pertinent in models that include variance functions or other terms that introduce parameters in nonlinear forms, place constraints on parameters, or estimate nonlinear functions of parameters such as ICCs. It is known that Bayesian estimates can be more precise than their frequentist counterparts, especially when prior information is informative [ 32 ]. By illustrating our approach on an evaluation of the increasingly popular Observer OPTION 5 tool, we hope to catalyze the adoption of more meaningful and informative computations of ICC across all health measurement scale applications.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Estimation of an inter-rater intra-class correlation coefficient that overcomes common assumption violations in the assessment of health measurement scales

Bobak

Barr

O’Malley

2018

BMC Med Res Methodol

192

123

View full text Add to dashboard Cite

BackgroundIntraclass correlation coefficients (ICC) are recommended for the assessment of the reliability of measurement scales. However, the ICC is subject to a variety of statistical assumptions such as normality and stable variance, which are rarely considered in health applications.MethodsA Bayesian approach using hierarchical regression and variance-function modeling is proposed to estimate the ICC with emphasis on accounting for heterogeneous variances across a measurement scale. As an application, we review the implementation of using an ICC to evaluate the reliability of Observer OPTION5, an instrument which used trained raters to evaluate the level of Shared Decision Making between clinicians and patients. The study used two raters to evaluate recordings of 311 clinical encounters across three studies to evaluate the impact of using a Personal Decision Aid over usual care. We particularly focus on deriving an estimate for the ICC when multiple studies are being considered as part of the data.ResultsThe results demonstrate that ICC varies substantially across studies and patient-physician encounters within studies. Using the new framework we developed, the study-specific ICCs were estimated to be 0.821, 0.295, and 0.644. If the within- and between-encounter variances were assumed to be the same across studies, the estimated within-study ICC was 0.609. If heteroscedasticity is not properly adjusted for, the within-study ICC estimate was inflated to be as high as 0.640. Finally, if the data were pooled across studies without accounting for the variability between studies then ICC estimates were further inflated by approximately 0.02 while formerly allowing for between study variation in the ICC inflated its estimated value by approximately 0.066 to 0.072 depending on the model.ConclusionWe demonstrated that misuse of the ICC statistics under common assumption violations leads to misleading and likely inflated estimates of interrater reliability. A statistical analysis that overcomes these violations by expanding the standard statistical model to account for them leads to estimates that are a better reflection of a measurement scale’s reliability while maintaining ease of interpretation. Bayesian methods are particularly well suited to estimating the expanded statistical model.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Estimation of an inter-rater intra-class correlation coefficient that overcomes common assumption violations in the assessment of health measurement scales

Bobak

Barr

O’Malley

2018

BMC Med Res Methodol

192

123

View full text Add to dashboard Cite

show abstract

“…The three statistical methods, each of which assume a linear relationship between monitored and clinician time, gave similar results. Bartlett [ 19 ] found that the full Bayesian analysis gave more biased results than regression calibration for small effect sizes when the reliability of the imputation model was low. However, regression calibration can underestimate regression coefficients for large effect estimates [ 20 ].…”

Section: Discussionmentioning

confidence: 99%

Accuracy of time to treatment estimates in the CRASH-3 clinical trial: impact on the trial results

Mansukhani

Frimley

Shakur-Still

et al. 2020

Trials

View full text Add to dashboard Cite

Background Early treatment with tranexamic acid may reduce deaths after traumatic brain injury (TBI). In mild and moderate TBI, there is a time to treatment interaction, with early treatment being most beneficial. Time to treatment was recorded by clinicians and is subject to error. Using monitoring data from the CRASH-3 trial, we examine the impact of errors in time to treatment on estimated treatment effects. Methods The CRASH-3 trial was a randomised trial of the effect of tranexamic acid on death and vascular occlusive events in 12,737 TBI patients. This analysis includes the 8107 patients with a Glasgow coma scale score of 9 to 15 since previous analyses showed that these patients benefit most from early treatment. Clinician-recorded time to treatment was checked against ambulance and hospital records for 1368/12,737 (11%) patients. Patients who died were preferentially selected for monitoring and we monitored 36% of head injury deaths. We describe measurement errors using Bland-Altman graphs. We model the effect of tranexamic acid on head injury death using logistic regression with a time-treatment interaction term. We use regression calibration, multiple imputation and Bayesian analysis to estimate the impact of time to treatment errors. Results Clinicians rounded times to the nearest half or full hour in 66% of cases. Monitored times were also rounded and were identical to clinician times in 63% of patients. Times were underestimated by an average of 9 min (95% CI − 85, 66). There was more variability between clinician-recorded and monitored times in low- and middle-income countries than in high-income countries. The treatment effect estimate at 1 h was greater for monitored times OR = 0.61 (95% CI 0.47, 0.81) than for clinician-recorded times OR = 0.63 (95% CI 0.48, 0.83). All three adjustment methods gave similar time to treatment interactions. For Bayesian methods, the treatment effect at 1 h was OR = 0.58 (95% CI 0.43, 0.78). Using monitored times increased the time-treatment interaction term from 1.15 (95% CI 1.03, 1.27) to 1.16 (95% CI 1.05, 1.28). Conclusions Accurate estimation of time from injury to treatment is challenging, particularly in low resource settings. Adjustment for known errors in time to treatment had minimal impact on the trial results. Trial registration ClinicalTrials.gov NCT01402882 . Registered on 25 July 2011

show abstract

“…Recently, a number of methods have been proposed to account for shared error components through the integration of multiple realizations of exposure estimates in risk estimation [ 13 – 15 ]. In our view, the Bayesian hierarchical approach is another promising framework in this context [ 26 , 33 ]. It is arguably the most flexible approach to account for exposure uncertainty and corrected parameter estimates can be obtained by Markov Chain Monte Carlo sampling.…”

Section: Discussionmentioning

confidence: 99%

Shared and unshared exposure measurement error in occupational cohort studies and their effects on statistical inference in proportional hazards models

Hoffmann¹,

Laurier²,

Rage³

et al. 2018

PLoS ONE

View full text Add to dashboard Cite

Exposure measurement error represents one of the most important sources of uncertainty in epidemiology. When exposure uncertainty is not or only poorly accounted for, it can lead to biased risk estimates and a distortion of the shape of the exposure-response relationship. In occupational cohort studies, the time-dependent nature of exposure and changes in the method of exposure assessment may create complex error structures. When a method of group-level exposure assessment is used, individual worker practices and the imprecision of the instrument used to measure the average exposure for a group of workers may give rise to errors that are shared between workers, within workers or both. In contrast to unshared measurement error, the effects of shared errors remain largely unknown. Moreover, exposure uncertainty and magnitude of exposure are typically highest for the earliest years of exposure. We conduct a simulation study based on exposure data of the French cohort of uranium miners to compare the effects of shared and unshared exposure uncertainty on risk estimation and on the shape of the exposure-response curve in proportional hazards models. Our results indicate that uncertainty components shared within workers cause more bias in risk estimation and a more severe attenuation of the exposure-response relationship than unshared exposure uncertainty or exposure uncertainty shared between individuals. These findings underline the importance of careful characterisation and modeling of exposure uncertainty in observational studies.

show abstract

Bayesian correction for covariate measurement error: A frequentist evaluation and comparison with regression calibration

Cited by 39 publications

References 40 publications

Estimation of an inter-rater intra-class correlation coefficient that overcomes common assumption violations in the assessment of health measurement scales

Estimation of an inter-rater intra-class correlation coefficient that overcomes common assumption violations in the assessment of health measurement scales

Accuracy of time to treatment estimates in the CRASH-3 clinical trial: impact on the trial results

Shared and unshared exposure measurement error in occupational cohort studies and their effects on statistical inference in proportional hazards models

Contact Info

Product

Resources

About