Towards global model generalizability: independent cross-site feature evaluation for patient-level risk prediction models using the OHDSI network.

BACKGROUND: Predictive models show promise in healthcare, but their successful deployment is challenging due to limited generalizability. Current external validation often focuses on model performance with restricted feature use from the original training data, lacking insights into their suitability at external sites. Our study introduces an innovative methodology for evaluating features during both the development phase and the validation, focusing on creating and validating predictive models for post-surgery patient outcomes with improved generalizability. METHODS: Electronic health records (EHRs) from 4 countries (United States, United Kingdom, Finland, and Korea) were mapped to the OMOP Common Data Model (CDM), 2008-2019. Machine learning (ML) models were developed to predict post-surgery prolonged opioid use (POU) risks using data collected 6 months before surgery. Both local and cross-site feature selection methods were applied in the development and external validation datasets. Models were developed using Observational Health Data Sciences and Informatics (OHDSI) tools and validated on separate patient cohorts. RESULTS: Model development included 41 929 patients, 14.6% with POU. The external validation included 31 932 (UK), 23 100 (US), 7295 (Korea), and 3934 (Finland) patients with POU of 44.2%, 22.0%, 15.8%, and 21.8%, respectively. The top-performing model, Lasso logistic regression, achieved an area under the receiver operating characteristic curve (AUROC) of 0.75 during local validation and 0.69 (SD = 0.02) (averaged) in external validation. Models trained with cross-site feature selection significantly outperformed those using only features from the development site through external validation (P < .05). CONCLUSIONS: Using EHRs across four countries mapped to the OMOP CDM, we developed generalizable predictive models for POU. Our approach demonstrates the significant impact of cross-site feature selection in improving model performance, underscoring the importance of incorporating diverse feature sets from various clinical settings to enhance the generalizability and utility of predictive healthcare models.

Distinct and shared genetic architectures of gestational diabetes mellitus and type 2 diabetes.

Gestational diabetes mellitus (GDM) is a common metabolic disorder affecting more than 16 million pregnancies annually worldwide1,2. GDM is related to an increased lifetime risk of type 2 diabetes (T2D)1-3, with over a third of women developing T2D within 15 years of their GDM diagnosis. The diseases are hypothesized to share a genetic predisposition1-7, but few studies have sought to uncover the genetic underpinnings of GDM. Most studies have evaluated the impact of T2D loci only8-10, and the three prior genome-wide association studies of GDM11-13 have identified only five loci, limiting the power to assess to what extent variants or biological pathways are specific to GDM. We conducted the largest genome-wide association study of GDM to date in 12,332 cases and 131,109 parous female controls in the FinnGen study and identified 13 GDM-associated loci, including nine new loci. Genetic features distinct from T2D were identified both at the locus and genomic scale. Our results suggest that the genetics of GDM risk falls into the following two distinct categories: one part conventional T2D polygenic risk and one part predominantly influencing mechanisms disrupted in pregnancy. Loci with GDM-predominant effects map to genes related to islet cells, central glucose homeostasis, steroidogenesis and placental expression.

Mechanisms of thyrotropin receptor-mediated phenotype variability deciphered by gene mutations and M453T-knockin model.

The clinical spectrum of thyrotropin receptor-mediated (TSHR-mediated) diseases varies from loss-of-function mutations causing congenital hypothyroidism to constitutively active mutations (CAMs) leading to nonautoimmune hyperthyroidism (NAH). Variation at the TSHR locus has also been associated with altered lipid and bone metabolism and autoimmune thyroid diseases. However, the extrathyroidal roles of TSHR and the mechanisms underlying phenotypic variability among TSHR-mediated diseases remain unclear. Here we identified and characterized TSHR variants and factors involved in phenotypic variability in different patient cohorts, the FinnGen database, and a mouse model. TSHR CAMs were found in all 16 patients with NAH, with 1 CAM in an unexpected location in the extracellular leucine-rich repeat domain (p.S237N) and another in the transmembrane domain (p.I640V) in 2 families with distinct hyperthyroid phenotypes. In addition, screening of the FinnGen database revealed rare functional variants as well as distinct common noncoding TSHR SNPs significantly associated with thyroid phenotypes, but there was no other significant association between TSHR variants and more than 2,000 nonthyroid disease endpoints. Finally, our TSHR M453T-knockin model revealed that the phenotype was dependent on the mutation's signaling properties and was ameliorated by increased iodine intake. In summary, our data show that TSHR-mediated disease risk can be modified by variants at the TSHR locus both inside and outside the coding region as well as by altered TSHR-signaling and dietary iodine, supporting the need for personalized treatment strategies.

LAVAA: a lightweight association viewer across ailments.

Motivation: Biobank scale genetic associations results over thousands of traits can be difficult to visualize and navigate. Results: We have created LAVAA, a visualization web-application to generate genetic volcano plots for simultaneously considering the P-value, effect size, case counts, trait class and fine-mapping posterior probability at a single-nucleotide polymorphism (SNP) across a range of traits from a large set of genome-wide association study. We find that user interaction with association results in LAVAA can enrich and enhance the biological interpretation of individual loci. Availability and implementation: LAVAA is available as a stand-alone web service (https://geneviz.aalto.fi/LAVAA/) and will be available in future releases of the finngen.fi website starting with release 10 in late 2023.

Screening for Mutations in Isolated Central Hypothyroidism Reveals a Novel Mutation in Insulin Receptor Substrate 4.

Background: Central hypothyroidism (CeH) is a rare condition affecting approximately 1:16 000- 100 000 individuals. Congenital forms can harm normal development if not detected and treated promptly. Clinical and biochemical diagnosis, especially of isolated CeH, can be challenging. Cases are not usually detected in neonatal screening, which, in most countries, is focused on detection of the more prevalent primary hypothyroidism. Until now, five genetic causes for isolated CeH have been identified. Here we aimed to identify the genetic cause in two brothers with impaired growth diagnosed with CeH at the age of 5 years. We further evaluated the candidate gene variants in a large genetic database. Methods: Clinical and biochemical characterization together with targeted next-generation sequencing (NGS) was used to identify the genetic cause in a family of two brothers presenting with CeH. Screening of insulin receptor substrate 4 (IRS4) variants was carried out in the FinnGen database. Results: A novel monoallelic frameshift mutation c.1712_1713insT, p.Gly572Trp fs*32 in the X-linked IRS4 gene was identified by NGS analysis in both affected males and confirmed using Sanger sequencing. Their mother was an unaffected carrier. In addition to the declined growth at presentation, central hypothyroidism and blunted TRH test, no other phenotypic alterations were found. Diagnostic tests included head MRI, thyroid imaging, bone age, and laboratory tests for thyroid autoantibodies, glucose, insulin and glycosylated hemoglobin levels. Examination of the IRS4 locus in FinnGen (R5) database revealed the strongest associations to a rare Finnish haplotype associated with thyroid disorders (p = 1.3e-7) and hypothyroidism (p = 8.3e-7). Conclusions: Here, we identified a novel frameshift mutation in an X-linked IRS4 gene in two brothers with isolated CeH. Furthermore, we demonstrate an association of IRS4 gene locus to a general thyroid disease risk in the FinnGen database. Our findings confirm the role of IRS4 in isolated central hypothyroidism.

An expanded analysis framework for multivariate GWAS connects inflammatory biomarkers to functional variants and disease.

Multivariate methods are known to increase the statistical power to detect associations in the case of shared genetic basis between phenotypes. They have, however, lacked essential analytic tools to follow-up and understand the biology underlying these associations. We developed a novel computational workflow for multivariate GWAS follow-up analyses, including fine-mapping and identification of the subset of traits driving associations (driver traits). Many follow-up tools require univariate regression coefficients which are lacking from multivariate results. Our method overcomes this problem by using Canonical Correlation Analysis to turn each multivariate association into its optimal univariate Linear Combination Phenotype (LCP). This enables an LCP-GWAS, which in turn generates the statistics required for follow-up analyses. We implemented our method on 12 highly correlated inflammatory biomarkers in a Finnish population-based study. Altogether, we identified 11 associations, four of which (F5, ABO, C1orf140 and PDGFRB) were not detected by biomarker-specific analyses. Fine-mapping identified 19 signals within the 11 loci and driver trait analysis determined the traits contributing to the associations. A phenome-wide association study on the 19 representative variants from the signals in 176,899 individuals from the FinnGen study revealed 53 disease associations (p < 1 × 10-4). Several reported pQTLs in the 11 loci provided orthogonal evidence for the biologically relevant functions of the representative variants. Our novel multivariate analysis workflow provides a powerful addition to standard univariate GWAS analyses by enabling multivariate GWAS follow-up and thus promoting the advancement of powerful multivariate methods in genomics.

Phenome-wide association studies across large population cohorts support drug target validation.

Phenome-wide association studies (PheWAS) have been proposed as a possible aid in drug development through elucidating mechanisms of action, identifying alternative indications, or predicting adverse drug events (ADEs). Here, we select 25 single nucleotide polymorphisms (SNPs) linked through genome-wide association studies (GWAS) to 19 candidate drug targets for common disease indications. We interrogate these SNPs by PheWAS in four large cohorts with extensive health information (23andMe, UK Biobank, FINRISK, CHOP) for association with 1683 binary endpoints in up to 697,815 individuals and conduct meta-analyses for 145 mapped disease endpoints. Our analyses replicate 75% of known GWAS associations (P < 0.05) and identify nine study-wide significant novel associations (of 71 with FDR < 0.1). We describe associations that may predict ADEs, e.g., acne, high cholesterol, gout, and gallstones with rs738409 (p.I148M) in PNPLA3 and asthma with rs1990760 (p.T946A) in IFIH1. Our results demonstrate PheWAS as a powerful addition to the toolkit for drug discovery.