Methodological considerations for identifying multiple plasma proteins associated with all-cause mortality in a population-based prospective cohort
Author | Drake, Isabel |
Author | Hindy, George |
Author | Almgren, Peter |
Author | Engström, Gunnar |
Author | Nilsson, Jan |
Author | Melander, Olle |
Author | Orho-Melander, Marju |
Available date | 2023-08-29T10:21:47Z |
Publication Date | 2021 |
Publication Name | Scientific Reports |
Resource | Scopus |
ISSN | 20452322 |
Abstract | Novel methods to characterize the plasma proteome has made it possible to examine a wide range of proteins in large longitudinal cohort studies, but the complexity of the human proteome makes it difficult to identify robust protein-disease associations. Nevertheless, identification of individuals at high risk of early mortality is a central issue in clinical decision making and novel biomarkers may be useful to improve risk stratification. With adjustment for established risk factors, we examined the associations between 138 plasma proteins measured using two proximity extension assays and long-term risk of all-cause mortality in 3,918 participants of the population-based Malmö Diet and Cancer Study. To examine the reproducibility of protein-mortality associations we used a two-step random-split approach to simulate a discovery and replication cohort and conducted analyses using four different methods: Cox regression, stepwise Cox regression, Lasso-Cox regression, and random survival forest (RSF). In the total study population, we identified eight proteins that associated with all-cause mortality after adjustment for established risk factors and with Bonferroni correction for multiple testing. In the two-step analyses, the number of proteins selected for model inclusion in both random samples ranged from 6 to 21 depending on the method used. However, only three proteins were consistently included in both samples across all four methods (growth/differentiation factor-15 (GDF-15), N-terminal pro-B-type natriuretic peptide, and epididymal secretory protein E4). Using the total study population, the C-statistic for a model including established risk factors was 0.7222 and increased to 0.7284 with inclusion of the most predictive protein (GDF-15; P < 0.0001). All multiple protein models showed additional improvement in the C-statistic compared to the single protein model (all P < 0.0001). We identified several plasma proteins associated with increased risk of all-cause mortality independently of established risk factors. Further investigation into the putatively causal role of these proteins for longevity is needed. In addition, the examined methods for identifying multiple proteins showed tendencies for overfitting by including several putatively false positive findings. Thus, the reproducibility of findings using such approaches may be limited. |
Sponsor | The authors would like to thank all participants of the Malmö Diet and Cancer Study and the Malmö Diet and Cancer Study Steering Committee. The Clinical Biomarker Facility at SciLifeLab, Sweden, is acknowledged for providing assistance in protein analyses. This study was supported by the Påhlsson Foundation (I. Drake), the Crafoord Foundation (I. Drake), the Swedish Society for Medical Research (I. Drake), the Swedish Heart and Lung Foundation (M. Orho-Melander), the Swedish Research Council (M. Orho-Melander), the European Research Council (Consolidator grant nr 649021, M. Orho-Melander), the Novo Nordic Foundation (M. Orho-Melander), the Swedish Diabetes Foundation (M. Orho-Melander), the Region Skåne (ALF for M. Orho-Melander), the Linneus Foundation for the Lund University Diabetes Center, the Swedish Foundation for Strategic Research for IRC15-0067 and by equipment grants from the Knut and Alice Wallenberg Foundation. We further acknowledge support from Lund University Infrastructure grant “Malmö population-based cohorts” (STYR 2019/2046). |
Language | en |
Publisher | Nature Research |
Subject | Biomarkers Cancer Cardiovascular diseases Computational models Epidemiology Machine learning Outcomes research Predictive markers Proteomic analysis |
Type | Article |
Issue Number | 1 |
Volume Number | 11 |
Files in this item
This item appears in the following Collection(s)
-
Medicine Research [1418 items ]