Comparison of methods to construct a genetic risk score for prediction of rheumatoid arthritis in the population-based Nord-Trøndelag Health Study, Norway

S Rostami; M Hoff; M A Brown; K Hveem; V Videm

doi:10.1093/rheumatology/kez638

Comparison of methods to construct a genetic risk score for prediction of rheumatoid arthritis in the population-based Nord-Trøndelag Health Study, Norway

Rheumatology (Oxford). 2020 Jul 1;59(7):1743-1751. doi: 10.1093/rheumatology/kez638.

Authors

S Rostami¹, M Hoff^{2

3

4}, M A Brown⁵, K Hveem⁶, V Videm^{1

7}

Affiliations

¹ Department of Clinical and Molecular Medicine, NTNU - Norwegian University of Science and Technology, Trondheim, Norway.
² Department of Rheumatology, St. Olavs University Hospital, Trondheim, Norway.
³ Department of Neuromedicine and Movement Science, NTNU - Norwegian University of Science and Technology, Trondheim, Norway.
⁴ Department of Public Health and Nursing, NTNU - Norwegian University of Science and Technology, Trondheim, Norway.
⁵ Institute of Health and Biomedical Innovation, Translational Research Institute, Princess Alexandra Hospital, Queensland University of Technology, Brisbane, Queensland, Australia.
⁶ KG Jebsen Center for Genetic Epidemiology, NTNU - Norwegian University of Science and Technology, Trondheim, Norway.
⁷ Department of Immunology and Transfusion Medicine, St. Olavs University Hospital, Trondheim, Norway.

PMID: 31930319
DOI: 10.1093/rheumatology/kez638

Abstract

Objectives: To evaluate selection methods among published single-nucleotide polymorphisms (SNPs) associated with RA to construct predictive genetic risk scores (GRSs) in a population-based setting.

Methods: The Nord-Trøndelag Health (HUNT) Study is a prospective cohort study among the whole adult population of northern Trøndelag, Norway. Participants in HUNT2 (1995-1997) and HUNT3 (2006-2008) were included (489 RA cases, 61 584 controls). The initial SNP selection from relevant genome-wide studies included 269 SNPs from 30 studies. Following different selection criteria, SNPs were weighted by published odds ratios. The sum of each person's carriage of all weighted susceptibility variants was calculated for each GRS.

Results: The best-fitting risk score included 27 SNPs [weighted genetic risk score 27 (wGRS27)] and was identified using P-value selection criterion ≤5 × 10-8, the largest possible SNP selection without high linkage disequilibrium (r2 < 0.8), and lasso regression to select for positive coefficients. In a logistic regression model adjusted for gender, age and ever smoking, wGRS27 was associated with RA [odds ratio 1.86 (95% CI 1.71, 2.04) for each s.d. increase, P < 0.001]. The AUC was 0.76 (95% CI 0.74, 0.78). The positive and negative predictive values were 1.6% and 99.7%, respectively, and the positive predictive value was not improved in sensitivity analyses subselecting participants to illustrate settings with increased RA prevalences. Other schemes selected more SNPs but resulted in GRSs with lower predictive ability.

Conclusion: Constructing a wGRS based on a smaller selection of informative SNPs improved predictive ability. Even with a relatively high AUC, the low PPV illustrates that there was a large overlap in risk variants among RA patients and controls, precluding clinical usefulness.

Keywords: epidemiology; genetics; rheumatoid arthritis.

Publication types

Comparative Study
Research Support, Non-U.S. Gov't

MeSH terms

Adult
Aged
Arthritis, Rheumatoid / diagnosis
Arthritis, Rheumatoid / epidemiology*
Arthritis, Rheumatoid / genetics*
Female
Genetic Predisposition to Disease*
Humans
Male
Middle Aged
Norway / epidemiology
Polymorphism, Single Nucleotide
Prevalence
Prospective Studies
Risk Assessment