Response misclassification in studies on bilateral diseases

Günther, Felix; Brandl, Caroline; Heid, Iris M.; Küchenhoff, Helmut

doi:10.1002/bimj.201900039

Cited by 3 publications

(5 citation statements)

References 10 publications

(11 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our developed approach provides a general framework for bilateral diseases with entity‐specific misclassification that propagates to person‐specific disease misclassification. Our approach also allows for missing classification in one of two entities, which is a second source of bias in association analyses for bilateral diseases as reported previously (Günther et al, 2019). We exemplify our approach on machine learning‐derived AMD compared to manually graded AMD.…”

Section: Resultsmentioning

confidence: 98%

“…In contrast to classical diseases and logistic regression (Carroll et al, 2006; Hausman et al, 1998; Neuhaus, 1999), no method is currently available to adjust for response misclassification in bilateral diseases. As described previously (Günther, Brandl, Heid, & Küchenhoff, 2019), the conceptual challenge is to account for two types of misclassification: (a) entity‐specific misclassification that propagates to an error‐prone person‐specific disease status; and (b) person‐specific misclassification from a missing disease status in one of the two entities. We thus developed an MLA to account for the fact that we are using an error‐prone response

{Y *}_{i} ≔ \max ({Z *}_{1 i}, {Z *}_{2 i})

{Z *}_{1 i}, {Z *}_{2 i} \in true{0,1 true}

, in the association analysis, while the true disease

Y_{i} ≔ normalmax (Z_{1 i}, Z_{2 i})

Z_{1 i}, Z_{2 i} \in {0,1}

is assumed to follow a logistic regression model.…”

Section: Methodsmentioning

confidence: 99%

“…, and uses logistic regression to estimate the association of some covariates X with AMD: the person-specific disease status Y i equals 1, if at least one eye of individual i is classified as AMD, and Y i equals 0, if both eyes are unaffected. As described previously (Günther et al, 2019), such a worse eye disease status can be misclassified because of two reasons: either, because of missing disease information in one of two eyes (in this case disease can be overlooked), or because of error-prone disease status for any of the two eyes. Here, we assume that we observed an error-prone, eye-specific disease status Z Z ( , ) * * i i 1 2 for each of the two eyes of a "main study" participant i and additionally the true disease status in each of the two eyes (Z Z , j j 1 2 ) for a subset of study participants j from the "validation study."…”

Section: Appendix A: Mla To Adjust For Response Misclassification In mentioning

confidence: 99%

“…The general problem of response misclassification when AMD information is missing in one of two eyes and/or the eye-specific classification suffers from misclassification with known classification probabilities has already been evaluated in a previous publication (Günther et al, 2019). There, we also derived the corresponding likelihood contributions for the different scenarios of available outcome data.…”

Section: Appendix A: Mla To Adjust For Response Misclassification In mentioning

confidence: 99%

See 3 more Smart Citations

Chances and challenges of machine learning‐based disease classification in genetic association studies illustrated on age‐related macular degeneration

Guenther

Brandl

Winkler

et al. 2020

Genetic Epidemiology

Self Cite

View full text Add to dashboard Cite

Imaging technology and machine learning algorithms for disease classification set the stage for high‐throughput phenotyping and promising new avenues for genome‐wide association studies (GWAS). Despite emerging algorithms, there has been no successful application in GWAS so far. We establish machine learning‐based phenotyping in genetic association analysis as misclassification problem. To evaluate chances and challenges, we performed a GWAS based on automatically classified age‐related macular degeneration (AMD) in UK Biobank (images from 135,500 eyes; 68,400 persons). We quantified misclassification of automatically derived AMD in internal validation data (4,001 eyes; 2,013 persons) and developed a maximum likelihood approach (MLA) to account for it when estimating genetic association. We demonstrate that our MLA guards against bias and artifacts in simulation studies. By combining a GWAS on automatically derived AMD and our MLA in UK Biobank data, we were able to dissect true association (ARMS2/HTRA1, CFH) from artifacts (near HERC2) and identified eye color as associated with the misclassification. On this example, we provide a proof‐of‐concept that a GWAS using machine learning‐derived disease classification yields relevant results and that misclassification needs to be considered in analysis. These findings generalize to other phenotypes and emphasize the utility of genetic data for understanding misclassification structure of machine learning algorithms.

show abstract

Section: Resultsmentioning

confidence: 98%

{Y *}_{i} ≔ \max ({Z *}_{1 i}, {Z *}_{2 i})

{Z *}_{1 i}, {Z *}_{2 i} \in true{0,1 true}

, in the association analysis, while the true disease

Y_{i} ≔ normalmax (Z_{1 i}, Z_{2 i})

Z_{1 i}, Z_{2 i} \in {0,1}

is assumed to follow a logistic regression model.…”

Section: Methodsmentioning

confidence: 99%

Section: Appendix A: Mla To Adjust For Response Misclassification In mentioning

confidence: 99%

Section: Appendix A: Mla To Adjust For Response Misclassification In mentioning

confidence: 99%

See 2 more Smart Citations

Chances and challenges of machine learning‐based disease classification in genetic association studies illustrated on age‐related macular degeneration

Guenther

Brandl

Winkler

et al. 2020

Genetic Epidemiology

Self Cite

View full text Add to dashboard Cite

show abstract

“…Shorter follow-up intervals would have improved the uncertainty in estimating incidence rates. Missingness of image data for one of the two eyes of a participant can lead to biased estimates 7 35. However, the one-eye missingness was only 4% at baseline and 5% at follow-up and the bias, thus, limited.…”

Section: Discussionmentioning

confidence: 99%

Incidence, progression and risk factors of age-related macular degeneration in 35–95-year-old individuals from three jointly designed German cohort studies

et al. 2022

Self Cite

View full text Add to dashboard Cite

ObjectiveTo estimate age-related macular degeneration (AMD) incidence/progression across a wide age range.Methods and analysisAMD at baseline and follow-up (colour fundus imaging, Three Continent AMD Consortium Severity Scale, 3CACSS, clinical classification, CC) was assessed for 1513 individuals aged 35–95 years at baseline from three jointly designed population-based cohorts in Germany: Kooperative Gesundheitsforschung in der Region Augsburg (KORA-Fit, KORA-FF4) and Altersbezogene Untersuchungen zur Gesundheit der Universität Regensburg (AugUR) with 18-year, 14-year or 3-year follow-up, respectively. Baseline assessment included lifestyle, metabolic and genetic markers. We derived cumulative estimates, rates and risk factor association for: (1) incident early AMD, (2) incident late AMD among no AMD at baseline (definition 1), (3) incident late AMD among no/early AMD at baseline (definition 2), (4) progression from early to late AMD.ResultsIncidence/progression increased by age, except progression in 70+-year old. We observed 35–55-year-old with 3CACSS-based early AMD who progressed to late AMD. Predominant risk factor for incident late AMD definition 2 was early AMD followed by genetics and smoking. When separating incident late AMD definition 1 from progression (instead of combined as incident late AMD definition 2), estimates help judge an individual’s risk based on age and (3CACSS) early AMD status: for example, for a 65-year old, 3-year late AMD risk with no or early AMD is 0.5% or 7%, 3-year early AMD risk is 3%; for an 85-year old, these numbers are 0.5%, 21%, 12%, respectively. For CC-based ‘early/intermediate’ AMD, incidence was higher, but progression was lower.ConclusionWe provide a practical guide for AMD risk for ophthalmology practice and healthcare management and document a late AMD risk for individuals aged <55 years.

show abstract

Chances and challenges of machine learning based disease classification in genetic association studies illustrated on age-related macular degeneration

Günther

Brandl

Winkler

et al. 2019

Preprint

View full text Add to dashboard Cite

18Imaging technology and machine learning algorithms for disease classification set the 19 stage for high-throughput phenotyping and promising new avenues for genome-wide 20 association studies (GWAS). Despite emerging algorithms, there has been no successful 21 application in GWAS so far. We established machine learning based disease classification 22 in genetic association analysis as a misclassification problem. To evaluate chances and 23 challenges, we performed a GWAS based on automated classification of age-related 24 macular degeneration (AMD) in UK Biobank (images from 135,500 eyes; 68,400 persons). 25We quantified misclassification of automatically derived AMD in internal validation data 26 (images from 4,001 eyes; 2,013 persons) and developed a maximum likelihood approach 27 (MLA) to account for it when estimating genetic association. We demonstrate that our MLA 28 guards against bias and artefacts in simulation studies. By combining a GWAS on 29 automatically derived AMD classification and our MLA in UK Biobank data, we were able 30 to dissect true association (ARMS2/HTRA1, CFH) from artefacts (near HERC2) and to 31 identify eye color as relevant source of misclassification. On this example of AMD, we are 32 able to provide a proof-of-concept that a GWAS using machine learning derived disease 33 classification yields relevant results and that misclassification needs to be considered in 34 the analysis. These findings generalize to other phenotypes and also emphasize the utility 35 of genetic data for understanding misclassification structure of machine learning 36 algorithms. 37 one setting might be misinterpreted in another, which can hamper transferability of trained 53 models; a topic discussed as dataset shift or domain shift 2-4 . Most predictions of deep learning 54 algorithms for image-based disease classification will be error-prone and the structure of 55 misclassification will generally be unknown. When using automated disease classification as 56 outcome for association analyses and GWAS, the underlying response misclassification is 57 usually unaccounted for, giving rise to biased effect estimates and potentially false-positive 58 associations 5-7 . Extent and structure of the misclassification process can be assessed by internal 59 validation data, i.e. a subset of participants with both automated and gold standard classification, 60 which can also be utilized to account for response misclassification in statistical models 7,8 . 61 At present, it is unclear whether machine learning based disease classification is of any 62 utility for association analyses, particularly for detecting disease signals in GWAS. We thus set 63 out to evaluate machine learning derived disease classification in GWAS on the example of age-64 related macular degeneration (AMD) and we developed a statistical approach accounting for the 65 implied response misclassification. AMD is an ideal role model, as a common disease 66 ascertained via imaging of the central retina 9 and with particularly strong known geneti...

show abstract

Response misclassification in studies on bilateral diseases

Cited by 3 publications

References 10 publications

Chances and challenges of machine learning‐based disease classification in genetic association studies illustrated on age‐related macular degeneration

Chances and challenges of machine learning‐based disease classification in genetic association studies illustrated on age‐related macular degeneration

Incidence, progression and risk factors of age-related macular degeneration in 35–95-year-old individuals from three jointly designed German cohort studies

Chances and challenges of machine learning based disease classification in genetic association studies illustrated on age-related macular degeneration

Contact Info

Product

Resources

About