Biomarker de-Mendelization: principles, potentials and limitations of a strategy to improve biomarker prediction by reducing the component of variance explained by genotype

Abstract

In observational studies, the Mendelian randomization approach can be used to circumvent confounding, bias and reverse causation, and to assess a potential causal association between a biomarker and risk of disease. If, on the other hand, a substantial component of variance of a non-causal biomarker is explained by genotype, then genotype could potentially attenuate the observational association and the strength of the prediction.In order to reduce the component of variance explained by genotype, an approach that can be seen as the inverse of Mendelian randomization -biomarker de-Mendelizationappears plausible.Plasma YKL-40 is a good candidate for demonstrating principles of biomarker de-Mendelization because it is a non-causal biomarker with a substantial component of variance explained by genotype. This approach is an attempt to improve the observational association and the strength of a predictive biomarker; it is explicitly not aimed at detection of causal effects.We studied 21 161 individuals form the Danish general population with measurements of YKL-40 concentration and rs4950928 genotype. Four different methods for biomarker de-Mendelization are explored for alcoholic liver cirrhosis and lung cancer.De-Mendelization methods only improved predictive ability slighly. We observed an interaction between genotype and markers of developing disease with respect to YKL-40 concentration.Even when genotype explains 14% of the variance in a non-causal biomarker, we found no useful empirical improvement in risk prediction by biomarker de-Mendelization. This could reflect the predictive interaction between genotype and disease development being removed which counterbalanced any beneficial properties of the method in this situation. The CGPS is a prospective cohort study initiated in 2003 with ongoing enrollment and emphasis on participants aged 20-100. Measurements of both plasma YKL-40 concentrations and CHI3L1 rs4950928 genotype were available in 13 180 individuals, examined 2003 through 2011. Ethics The studies were conducted according to the Declaration of Helsinki and approved by Herlev Hospital and a Danish scientific ethical committee (No. 100.2039/91 and 01-144/01). All participants gave written informed consent. Endpoints Diagnoses of alcoholic liver cirrhosis and lung cancer were defined according to the World Health Organization International Classification of Diseases; the 8th and 10th edition for alcoholic liver cirrhosis (ICD8: 571.09; ICD10: K70.3) and the 7th and 10th edition for lung cancer (ICD7: 162-164; 462; 562; 962-964. ICD10: C33-34; C37-38). Diagnoses were obtained from the national Danish Patient Registry, the national Danish Causes of Death Registry, and the national Danish Cancer Registry, with a very high degree of completeness [19]. Date of death was obtained from the national Danish Civil RegistrationSystem. We did not lose track of any participant. YKL-40 analysisPlasma concentrations of YKL-40 were determined in duplicates in samples frozen for 12-15 years at -80°C in the CCHS (N=7981) ...

Can Mendelian Randomization Shift into Reverse Gear?

“…the association arises from reverse causality), measuring the trait is of importance as it gives a dynamic 'read out' of disease status. Finally, when a biomarker is non-causal in disease, including genotype in the model may increase the predictive utility of the biomarker, through the principle of so-called "biomarker de-Mendelization", by maximizing the non-genetic variation in the phenotype (15).…”

Genetically Adjusted PSA Levels for Prostate Cancer Screening

“…Leveraging genetic profiles to personalize clinical biomarkers enables the translation of GWAS discoveries into clinical practice. This concept has been referred to as “de-Mendelization,” since it is essentially Mendelian randomization in reverse – instead of relying on genetically-predicted biomarker values to investigate causal relationships, subtracting the component of variance attributed to genetic factors for non-causal predictive biomarkers can maximize the residual disease-related signal and yield appreciable improvement in disease prediction 44,45 . While doing so has been alluded to in previous work on PSA genetics 24,46 and other biomarkers 44,47,48 , the value of this approach for detecting clinically meaningful disease and reducing unnecessary diagnostic testing has not been demonstrated prior to this study.…”