New biomarker model outmatches conventional risk factors for predicting mortality

A new model using 14 biomarkers may be more accurate at predicting longer-term mortality than a model comprising conventional risk factors, based on the largest metabolomics study to date.

The prognostic model was more accurate at predicting 5- and 10-year mortality across all ages, reported Joris Deelen, PhD, of Leiden (the Netherlands) University Medical Center and colleagues.

“These results suggest that metabolic biomarker profiling could potentially be used to guide patient care, if further validated in relevant clinical settings,” the investigators wrote in Nature Communications.

“There is no consensus on the ultimate set of predictors of longer-term [5-10 years] mortality risk, since the predictive power of the currently used risk factors is limited, especially at higher ages,” the investigators wrote. “However, it is especially this age group and follow-up time window for which a robust tool would aid clinicians in assessing whether treatment is still sensible.”

The current study was a survival meta-analysis of 44,168 individuals from 12 cohorts aged between 18 and 109 years at baseline. First, the investigators looked for associations between 226 metabolic biomarkers and all-cause mortality in the 5,512 people who died during follow-up. This revealed associations between mortality and 136 biomarkers, which increased to 159 biomarkers after adjusting for recently reported all-cause mortality associations with albumin, very low-density lipoprotein (VLDL) particle size, citrate, and glycoprotein acetyls. Because of strong correlations between many of the biomarkers evaluated, the investigators pared the field down to 63 biomarkers, then used a forward-backward procedure to ultimately identify 14 biomarkers independently associated with mortality. Of the four recently described biomarkers, citrate was excluded from the final model because of its minimal contribution to mortality estimates.

The 14 biomarkers were total lipids in chylomicrons and extremely large VLDL cholesterol, total lipids in small HDL cholesterol, mean diameter for VLDL cholesterol particles, ratio of polyunsaturated fatty acids to total fatty acids, glucose, lactate, histidine, isoleucine, leucine, valine, phenylalanine, acetoacetate, albumin, and glycoprotein acetyls.

“The 14 identified biomarkers are involved in various processes, such as lipoprotein and fatty acid metabolism, glycolysis, fluid balance, and inflammation. Although the majority of these biomarkers have been associated with mortality before, this is the first study that shows their independent effect when combined into one model,” the researchers wrote.

Implementation of the new biomarker model led to a score that typically ranged from –2 to 3. A 1-point increase was associated with a 173% increased risk of death (hazard ratio, 2.73; P less than 1 x 10^–132). Analysis of cause-specific mortality revealed that most biomarkers were predictive of multiple causes of death. Some biomarkers were more focused; glucose, for example, was more predictive of cardiovascular-related death than of death because of cancer or nonlocalized infections. Compared with a model incorporating conventional risk factors, the biomarker model more accurately predicted 5- and 10-year mortality, with respective C-statistics of 0.837 versus 0.772 and 0.830 versus 0.790. This superiority was even more pronounced when only individuals aged older than 60 years were included.

The study was funded by Biobanking and BioMolecular resources Research Initiative–Netherlands. The investigators reported additional relationships with Nightingale Health, Novo Nordisk, and Bayer.

SOURCE: Deelen J et al. Nature Comm. 2019 Aug 20. doi: 10.1038/s41467-019-11311-9.

A new model using 14 biomarkers may be more accurate at predicting longer-term mortality than a model comprising conventional risk factors, based on the largest metabolomics study to date.

The prognostic model was more accurate at predicting 5- and 10-year mortality across all ages, reported Joris Deelen, PhD, of Leiden (the Netherlands) University Medical Center and colleagues.

“These results suggest that metabolic biomarker profiling could potentially be used to guide patient care, if further validated in relevant clinical settings,” the investigators wrote in Nature Communications.

“There is no consensus on the ultimate set of predictors of longer-term [5-10 years] mortality risk, since the predictive power of the currently used risk factors is limited, especially at higher ages,” the investigators wrote. “However, it is especially this age group and follow-up time window for which a robust tool would aid clinicians in assessing whether treatment is still sensible.”

The current study was a survival meta-analysis of 44,168 individuals from 12 cohorts aged between 18 and 109 years at baseline. First, the investigators looked for associations between 226 metabolic biomarkers and all-cause mortality in the 5,512 people who died during follow-up. This revealed associations between mortality and 136 biomarkers, which increased to 159 biomarkers after adjusting for recently reported all-cause mortality associations with albumin, very low-density lipoprotein (VLDL) particle size, citrate, and glycoprotein acetyls. Because of strong correlations between many of the biomarkers evaluated, the investigators pared the field down to 63 biomarkers, then used a forward-backward procedure to ultimately identify 14 biomarkers independently associated with mortality. Of the four recently described biomarkers, citrate was excluded from the final model because of its minimal contribution to mortality estimates.

The 14 biomarkers were total lipids in chylomicrons and extremely large VLDL cholesterol, total lipids in small HDL cholesterol, mean diameter for VLDL cholesterol particles, ratio of polyunsaturated fatty acids to total fatty acids, glucose, lactate, histidine, isoleucine, leucine, valine, phenylalanine, acetoacetate, albumin, and glycoprotein acetyls.

“The 14 identified biomarkers are involved in various processes, such as lipoprotein and fatty acid metabolism, glycolysis, fluid balance, and inflammation. Although the majority of these biomarkers have been associated with mortality before, this is the first study that shows their independent effect when combined into one model,” the researchers wrote.

Implementation of the new biomarker model led to a score that typically ranged from –2 to 3. A 1-point increase was associated with a 173% increased risk of death (hazard ratio, 2.73; P less than 1 x 10^–132). Analysis of cause-specific mortality revealed that most biomarkers were predictive of multiple causes of death. Some biomarkers were more focused; glucose, for example, was more predictive of cardiovascular-related death than of death because of cancer or nonlocalized infections. Compared with a model incorporating conventional risk factors, the biomarker model more accurately predicted 5- and 10-year mortality, with respective C-statistics of 0.837 versus 0.772 and 0.830 versus 0.790. This superiority was even more pronounced when only individuals aged older than 60 years were included.

The study was funded by Biobanking and BioMolecular resources Research Initiative–Netherlands. The investigators reported additional relationships with Nightingale Health, Novo Nordisk, and Bayer.

SOURCE: Deelen J et al. Nature Comm. 2019 Aug 20. doi: 10.1038/s41467-019-11311-9.

A new model using 14 biomarkers may be more accurate at predicting longer-term mortality than a model comprising conventional risk factors, based on the largest metabolomics study to date.

The prognostic model was more accurate at predicting 5- and 10-year mortality across all ages, reported Joris Deelen, PhD, of Leiden (the Netherlands) University Medical Center and colleagues.

“These results suggest that metabolic biomarker profiling could potentially be used to guide patient care, if further validated in relevant clinical settings,” the investigators wrote in Nature Communications.

“There is no consensus on the ultimate set of predictors of longer-term [5-10 years] mortality risk, since the predictive power of the currently used risk factors is limited, especially at higher ages,” the investigators wrote. “However, it is especially this age group and follow-up time window for which a robust tool would aid clinicians in assessing whether treatment is still sensible.”

The current study was a survival meta-analysis of 44,168 individuals from 12 cohorts aged between 18 and 109 years at baseline. First, the investigators looked for associations between 226 metabolic biomarkers and all-cause mortality in the 5,512 people who died during follow-up. This revealed associations between mortality and 136 biomarkers, which increased to 159 biomarkers after adjusting for recently reported all-cause mortality associations with albumin, very low-density lipoprotein (VLDL) particle size, citrate, and glycoprotein acetyls. Because of strong correlations between many of the biomarkers evaluated, the investigators pared the field down to 63 biomarkers, then used a forward-backward procedure to ultimately identify 14 biomarkers independently associated with mortality. Of the four recently described biomarkers, citrate was excluded from the final model because of its minimal contribution to mortality estimates.

The 14 biomarkers were total lipids in chylomicrons and extremely large VLDL cholesterol, total lipids in small HDL cholesterol, mean diameter for VLDL cholesterol particles, ratio of polyunsaturated fatty acids to total fatty acids, glucose, lactate, histidine, isoleucine, leucine, valine, phenylalanine, acetoacetate, albumin, and glycoprotein acetyls.

“The 14 identified biomarkers are involved in various processes, such as lipoprotein and fatty acid metabolism, glycolysis, fluid balance, and inflammation. Although the majority of these biomarkers have been associated with mortality before, this is the first study that shows their independent effect when combined into one model,” the researchers wrote.

Implementation of the new biomarker model led to a score that typically ranged from –2 to 3. A 1-point increase was associated with a 173% increased risk of death (hazard ratio, 2.73; P less than 1 x 10^–132). Analysis of cause-specific mortality revealed that most biomarkers were predictive of multiple causes of death. Some biomarkers were more focused; glucose, for example, was more predictive of cardiovascular-related death than of death because of cancer or nonlocalized infections. Compared with a model incorporating conventional risk factors, the biomarker model more accurately predicted 5- and 10-year mortality, with respective C-statistics of 0.837 versus 0.772 and 0.830 versus 0.790. This superiority was even more pronounced when only individuals aged older than 60 years were included.

The study was funded by Biobanking and BioMolecular resources Research Initiative–Netherlands. The investigators reported additional relationships with Nightingale Health, Novo Nordisk, and Bayer.

SOURCE: Deelen J et al. Nature Comm. 2019 Aug 20. doi: 10.1038/s41467-019-11311-9.