In dayoonkwon/BioAge: Biological Age Calculations Using Several Biomarker Algorithms

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  warning = FALSE,
  message = FALSE
)

Overview

This vignette serves as an example of training biologial aging measures using the NHANES III (1988 - 1994) and projecting into NHANES IV (1999 - 2018) dataset. It also provides documentation for fit parameters contained in the BioAge package. The cleaned NHANES dataset is loaded as the dataset NHANES3 and NHANES4. The original KDM bioage and phenoage values are saved as kdm0 and phenoage0 as part of NHANES dataset.

Installation (via devtools):

You can install the released version of BioAge from (https://github.com/dayoonkwon/BioAge) with:

library(BioAge) #topic of vignette
library(dplyr)

Step 1: train algorithms in NHANES III and project biological aging measures in NHANES IV

I train in the NHANES III and project biological aging measures into the NHANES IV by using the hd_nhanes, kdm_nhanes, and phenoage_nhanes function of the BioAge package.

#specify bioamarkers included in the algorithms 
biomarkers = c("albumin","alp","lymph","mcv","lncreat","lncrp","hba1c","wbc","rdw")
#HD using NHANES (separate training for men and women)
hd = hd_nhanes(biomarkers)
#KDM bioage using NHANES (separate training for men and women)
kdm = kdm_nhanes(biomarkers)
#phenoage uinsg NHANES
phenoage = phenoage_nhanes(biomarkers)

Step 2: compare original KDM bioage and phenoage algorithms with algorithms composed with new biomarker set

The projected data and estimated models are saved as part of the list structure. The dataset can be drawn by typing data. The model can be drawn by typing fit.

#assemble NHANES IV dataset with projected biological aging measures for analysis
data = merge(hd$data, kdm$data) %>% merge(., phenoage$data)

Figure1. Association of biological aging measures with chronological age in NAHNES IV dataset

In the figure below, the graphs titled "KDM Biological Age" and "Levine Phenotypic Age" show measures based on the original biomarker sets published in Levine 2013 J Geron A and Levine et al. 2018 AGING. The remaining graphs shows the new measures computed with the biomarker set specified within this code.

#select biological age variables
agevar = c("kdm0","phenoage0","kdm","phenoage","hd","hd_log")
#prepare labels
label = c("KDM\nBiological Age",
          "Levine\nPhenotypic Age",
          "Modified-KDM\nBiological Age",
          "Modified-Levine\nPhenotypic Age",
          "Homeostatic\nDysregulation",
          "Log\nHomeostatic\nDysregulation")
#plot age vs bioage
plot_ba(data, agevar, label)

Figure2. Correlations among biological aging measures

The figure plots associations among the different biological aging measures. Cells below the diagonal show scatter plots of the measures listed above the cell (x-axis) and to the right (y-axis). Cells above the diagonal show the Pearson correlations for the measures listed below the cell and to the left. For this analysis, KDM Biological Age and Levine Phenotypic Age measures are differenced from chronological age (i.e. plotted values = BA-CA).

#select biological age variables
agevar = c("kdm_advance0","phenoage_advance0","kdm_advance","phenoage_advance","hd","hd_log")
#prepare lables
#values should be formatted for displaying along diagonal of the plot
#names should be used to match variables and order is preserved
label = c(
  "kdm_advance0"="KDM\nBiological Age\nAdvancement",
  "phenoage_advance0"="Levine\nPhenotypic Age\nAdvancement",
  "kdm_advance"="Modified-KDM\nBiological Age\nAdvancement",
  "phenoage_advance"="Modified-Levine\nPhenotypic Age\nAdvancement",
  "hd" = "Homeostatic\nDysregulation",
  "hd_log" = "Log\nHomeostatic\nDysregulation")
#use variable name to define the axis type ("int" or "float")
axis_type = c(
  "kdm_advance0"="float",
  "phenoage_advance0"="float",
  "kdm_advance"="float",
  "phenoage_advance"="flot",
  "hd"="flot",
  "hd_log"="float")
#plot BAA corplot
plot_baa(data,agevar,label,axis_type)

Table 1. Associations of biological aging measures with mortality

table_surv(data, agevar, label)

Table 2. Associations of biological aging measures with healthspan-related characteristics

The linear regression models and sample sizes in “Table 2” and “Table 3” below are saved as part of the list structure. Regression model can be drawn by typing table. Sample size can be drawn by typing n.

table2 = table_health(data,agevar,outcome = c("health","adl","lnwalk","grip_scaled"), label)
#pull table
table2$table
#pull sample sizes
table2$n

Table 3. Associations of socioeconomic circumstances measures with measures of biological aging

table3 = table_ses(data,agevar,exposure = c("edu","annual_income","poverty_ratio"), label)
#pull table
table3$table
#pull sample sizes
table3$n

Step 3: Project biological aging measures onto new data

In this example, the projection dataset is the 2016 Venous Blood Study dataset from the US Health and Retirement Study (data are not included in the package but are available from the HRS). For this analysis, HRS data were previously cleaned and units of measure and variable names were harmonized to match the NHANES data included with the package. All algorithms were trained using the NHANES III data and projected into the HRS using the hd_calc, kdm_calc, and phenoage_calc functions of the BioAge package.

#The HRS dataset is loaded from my local drive that has previously been downloaded and cleaned
newdata = HRS %>%
  select(hhidpn, sex, age, raracem, albumin, alp, lymphpct, mcv, creat, lncreat, crp, lncrp, hba1c, wbc, rdw, glucose) %>%
  rename(sampleID = hhidpn,
         gender = sex,
         race = raracem,
         lymph = lymphpct) %>%
  mutate(gender = ifelse(gender == "Women", 2,
                         ifelse(gender == "Men", 1, NA)),
         albumin_gL = albumin * 10,
         creat_umol = creat * 88.4017,
         lncreat_umol = log(creat_umol),
         crp = crp / 10,
         lncrp = log(crp),
         glucose_mmol = glucose*0.0555) %>%
  group_by(gender) %>%
  mutate_at(vars(albumin:creat,crp,hba1c:glucose),
            list(~ifelse((. > (mean(., na.rm = TRUE) + 5 * sd(., na.rm = TRUE)))|
                           (. < (mean(., na.rm = TRUE) - 5 * sd(., na.rm = TRUE))), NA, .))) %>%
  ungroup() %>%
  mutate(lncreat = ifelse(is.na(creat), NA, lncreat),
         lncrp = ifelse(is.na(crp), NA, lncrp),
         albumin_gL = ifelse(is.na(albumin), NA, albumin_gL),
         creat_umol = ifelse(is.na(creat), NA, creat_umol),
         lncreat_umol = ifelse(is.na(creat), NA, lncreat_umol),
         glucose_mmol = ifelse(is.na(glucose), NA, glucose_mmol))

Projecting HD into the HRS using NHANES III

For HD, the constructed variable is based on a malhanobis distance statistic, which is theoretically the distance between observations and a hypothetically healthy, young cohort. In this example, I train separately for men and women who are between the ages of 20 and 30 and not pregnant, and have observe biomarker data within clinically accpetable distributions. For clinical guidelines, I relied upon the ranges reported by the Mayo Clinic website.

#projecting HD into the HRS using NHANES III (seperate training for gender)
hd_fem = hd_calc(data = newdata %>%
                   filter(gender == 2),
                 reference = NHANES3 %>%
                   filter(gender == 2 & age >= 20 & age <= 30 & pregnant == 0) %>%
                   mutate(albumin = ifelse(albumin >= 3.5 & albumin <= 5, albumin, NA),
                          alp = ifelse(alp >= 37 & alp <= 98, alp, NA),
                          lymph = ifelse(lymph >= 20 & lymph <= 40, lymph, NA),
                          mcv = ifelse(mcv >= 78 & mcv <= 101, mcv, NA),
                          creat = ifelse(creat >= 0.6 & creat <= 1.1, creat, NA),
                          lncreat = ifelse(is.na(creat), NA, lncreat),
                          crp = ifelse(crp < 2, crp, NA),
                          lncrp = ifelse(is.na(crp), NA, lncrp),
                          hba1c = ifelse(hba1c >= 4 & hba1c <= 5.6, hba1c, NA),
                          wbc = ifelse(wbc >= 4.5 & wbc <= 11, wbc, NA),
                          rdw = ifelse(rdw >= 11.5 & rdw <= 14.5, rdw, NA)),
                 biomarkers)
hd_male = hd_calc(data = newdata %>%
                   filter(gender == 1),
                 reference = NHANES3 %>%
                   filter(gender == 1 & age >= 20 & age <= 30 & pregnant == 0) %>%
                   mutate(albumin = ifelse(albumin >= 3.5 & albumin <= 5, albumin, NA),
                          alp = ifelse(alp >= 45 & alp <= 115, alp, NA),
                          lymph = ifelse(lymph >= 20 & lymph <= 40, lymph, NA),
                          mcv = ifelse(mcv >= 82 & mcv <= 102, mcv, NA),
                          creat = ifelse(creat >= 0.8 & creat <= 1.3, creat, NA),
                          lncreat = ifelse(is.na(creat), NA, lncreat),
                          crp = ifelse(crp < 2, crp, NA),
                          lncrp = ifelse(is.na(crp), NA, lncrp),
                          hba1c = ifelse(hba1c >= 4 & hba1c <= 5.6, hba1c, NA),
                          wbc = ifelse(wbc >= 4.5 & wbc <= 11, wbc, NA),
                          rdw = ifelse(rdw >= 11.5 & rdw <= 14.5, rdw, NA)),
                 biomarkers)
#pull the HD dataset
hd_data = rbind(hd_fem$data, hd_male$data)

Projecting KDM bioage into the HRS using NHANES III

Having estimated biological aging models using NHANES III in "Step 1", I can project KDM bioage and phenoage into the HRS data by running kdm_calc and phenoage_calc and supplying a fit argument.

#projecting KDM bioage into the HRS using NHANES III (seperate training for gender)
kdm_fem = kdm_calc(data = newdata %>%
                     filter (gender ==2),
                   biomarkers,
                   fit = kdm$fit$female,
                   s_ba2 = kdm$fit$female$s_b2)
kdm_male = kdm_calc(data = newdata %>%
                     filter (gender ==1),
                   biomarkers,
                   fit = kdm$fit$male,
                   s_ba2 = kdm$fit$male$s_b2)
#pull the KDM dataset
kdm_data = rbind(kdm_fem$data, kdm_male$data)

Projecting phenoage into the HRS using NHANES III

phenoage_hrs = phenoage_calc(data = newdata %>%
                           mutate(albumin = albumin_gL,
                                  lncreat = lncreat_umol),
                         biomarkers,
                         fit = phenoage$fit,
                         orig = TRUE) #this calculate original phenoage 
phenoage_data = phenoage_hrs$data
#pull the full dataset
newdata = left_join(newdata, hd_data[, c("sampleID", "hd", "hd_log")], by = "sampleID") %>%
  left_join(., kdm_data[, c("sampleID", "kdm", "kdm_advance")], by = "sampleID") %>%
  left_join(., phenoage_data[, c("sampleID","phenoage0","phenoage_advance0",
                                 "phenoage","phenoage_advance")], by = "sampleID") 

Summary statistics of calculated biological aging measures for the HRS

summary(newdata %>% select(phenoage0, kdm, phenoage, hd, hd_log))

dayoonkwon/BioAge documentation built on Oct. 11, 2023, 1:33 a.m.