Heterogenous associations of polygenic indices of 35 traits with mortality
Curation statements for this article:-
Curated by eLife
eLife Assessment
This valuable study reports convincing evidence about associations between 35 polygenic indices (PGIs) for social, behavioral, and psychological traits, along with some non-fatal health conditions (e.g., BMI) and all-cause mortality in data from Finnish population-based surveys and a twin cohort linked with administrative registers. PGIs for education, depression, alcohol use, smoking, BMI, and self-rated health showed the strongest associations with all-cause mortality, on the order of ~10% increment in risk per PGI standard deviation. Effect sizes from twin-difference analyses tended to be slightly larger than the effect sizes from population cohorts, opposite the pattern generally observed when testing PGI associations with their target phenotypes and supporting robustness of findings to confounding by population stratification.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
Background
Polygenic indices (PGIs) of various traits abound, but the knowledge remains limited on how they predict wide-ranging health indicators, including the risk of death. We investigated the associations between mortality and 35 different PGIs related to social, behavioural and psychological traits, and typically non-fatal health conditions.
Methods
Data consist of Finnish adults from population-representative genetically informed epidemiological surveys (Finrisk 1992–2012, Health2000/2011, FinHealth 2017), linked to administrative registers (N: 40 097, 5948 deaths). Within-sibship analysis was complemented with dizygotic twins from Finnish twin study cohorts (N: 10 174, 2116 deaths). We estimated Cox proportional hazards models with mortality follow up 1995– 2019.
Results
PGIs most strongly predictive of all-cause mortality were ever smoking (hazard ratio [HR]=1.12, 95% confidence interval [95%CI] 1.09;1.14 per one standard deviation larger PGI), self-rated health (HR=0.90, 95%CI 0.88;0.93), body mass index (HR=1.10, 95%CI 1.07;1.12), educational attainment (HR=0.91, 95%CI 0.89;0.94, depressive symptoms (HR=1.07, 95%CI 1.04;1.10), and alcohol drinks per week (HR=1.06, 95%CI 1.04;1.09). Within-sibship estimates were approximately consistent with the population analysis. The investigated PGIs were typically more predictive for external than for natural causes of death. PGIs were more strongly associated with death occurring at younger ages, while among those who survived to age 80, the PGI–mortality associations were negligible.
Conclusions
PGIs related to the best-established mortality risk phenotypes had the strongest associations with mortality. They offer moderate additional prediction even when mutually adjusting with their phenotype. Within-sibship analysis indicated no evidence for inflation of PGI-mortality associations by population phenomena.
Article activity feed
-
eLife Assessment
This valuable study reports convincing evidence about associations between 35 polygenic indices (PGIs) for social, behavioral, and psychological traits, along with some non-fatal health conditions (e.g., BMI) and all-cause mortality in data from Finnish population-based surveys and a twin cohort linked with administrative registers. PGIs for education, depression, alcohol use, smoking, BMI, and self-rated health showed the strongest associations with all-cause mortality, on the order of ~10% increment in risk per PGI standard deviation. Effect sizes from twin-difference analyses tended to be slightly larger than the effect sizes from population cohorts, opposite the pattern generally observed when testing PGI associations with their target phenotypes and supporting robustness of findings to confounding by population …
eLife Assessment
This valuable study reports convincing evidence about associations between 35 polygenic indices (PGIs) for social, behavioral, and psychological traits, along with some non-fatal health conditions (e.g., BMI) and all-cause mortality in data from Finnish population-based surveys and a twin cohort linked with administrative registers. PGIs for education, depression, alcohol use, smoking, BMI, and self-rated health showed the strongest associations with all-cause mortality, on the order of ~10% increment in risk per PGI standard deviation. Effect sizes from twin-difference analyses tended to be slightly larger than the effect sizes from population cohorts, opposite the pattern generally observed when testing PGI associations with their target phenotypes and supporting robustness of findings to confounding by population stratification.
-
Reviewer #1 (Public review):
Lahtinen et al. evaluated the association between polygenic scores and mortality. This question has been intensely studied (Sakaue 2020 Nature Medicine, Jukarainen 2022 Nature Medicine, Argentieri 2025 Nature Medicine), where most studies use PRS as an instrument to attribute death to different causes. The presented study focuses on polygenic scores of non-fatal outcomes and separates the cause of death into "external" and "internal". The majority of the results are descriptive, and the data doesn't have the power to distinguish effect sizes of the interesting comparisons: (1) differences between external vs. internal (2) differences between PGI effect and measured phenotype. I have two main comments:
(1) The authors should clarify whether the p-value reported in the text will remain significant after …
Reviewer #1 (Public review):
Lahtinen et al. evaluated the association between polygenic scores and mortality. This question has been intensely studied (Sakaue 2020 Nature Medicine, Jukarainen 2022 Nature Medicine, Argentieri 2025 Nature Medicine), where most studies use PRS as an instrument to attribute death to different causes. The presented study focuses on polygenic scores of non-fatal outcomes and separates the cause of death into "external" and "internal". The majority of the results are descriptive, and the data doesn't have the power to distinguish effect sizes of the interesting comparisons: (1) differences between external vs. internal (2) differences between PGI effect and measured phenotype. I have two main comments:
(1) The authors should clarify whether the p-value reported in the text will remain significant after multiple testing adjustment. Some of the large effects might be significant; for example, Figure 2C (note that the small prediction accuracy of PGI in older age groups has been extensively studied, see Jiang, Holmes, and McVean, 2021, PLoS Genetics).
(2) The authors might check if PGI+Phenotype has improved performance over Phenotype only. This is similar to Model 2 in Table 1, but slightly different.
-
Reviewer #2 (Public review):
Summary:
This study provides a comprehensive evaluation of the association between polygenic indices (PGIs) for 35 lifestyle and behavioral traits and all-cause mortality, using data from Finnish population- and family-based cohorts. The analysis was stratified by sex, cause of death (natural vs. external), age at death, and participants' educational attainment. Additional analyses focused on the six most predictive PGIs, examining their independent associations after mutual adjustment and adjustment for corresponding directly measured baseline risk factors.
Strengths:
Large sample size with long-term follow-up.
Use of both population- and family-based analytical approaches to evaluate associations.
Weaknesses:
It is unclear whether the PGIs used for each trait represent the most current or optimal versions …
Reviewer #2 (Public review):
Summary:
This study provides a comprehensive evaluation of the association between polygenic indices (PGIs) for 35 lifestyle and behavioral traits and all-cause mortality, using data from Finnish population- and family-based cohorts. The analysis was stratified by sex, cause of death (natural vs. external), age at death, and participants' educational attainment. Additional analyses focused on the six most predictive PGIs, examining their independent associations after mutual adjustment and adjustment for corresponding directly measured baseline risk factors.
Strengths:
Large sample size with long-term follow-up.
Use of both population- and family-based analytical approaches to evaluate associations.
Weaknesses:
It is unclear whether the PGIs used for each trait represent the most current or optimal versions based on the latest GWAS data.
If the Finnish data used in this study also contributed to the development of some of the PGIs, there is a risk of overestimating their associations with mortality due to overfitting or "double-dipping." Similar inflation of effect sizes has been observed in studies using the UK Biobank, which is widely used for PGI construction.
-
