In bjb40/bioage: Klemera-Doubal biological age algorithm.

knitr::opts_chunk$set(warning = FALSE, message=FALSE)

Overview

This vignette serves as an example of training biological aging parameters using the NHANES dataset. It also provides documentation for fit parameters contained in the bioage package.

Data

For these data, we use NHANES datasets from 1994 to 2018. We train in the full NHANES, but project bioage into 50% of the nhanes3 sample. NHANES is a dataset which has been produce by

These data were merged and loaded as the dataset "nhanes". Here are the depenencies:

library(dplyr)
library(bioage) #beta version currently -- topic of vignette

#This codeblock loads data from my local machine that has previously been downloaded and cleane. The Original data is not available.
library(haven)
nhanes = read_dta('C:/Users/bjb40/Box Sync/Bartlett_BioAge/Data/nhanes/Stata/NHANESforBA.dta')

##limit by excluding pregnant woman and constraining the ages
nhanes = nhanes %>%
  filter(pregnant==0,
         age>=30, age<=75) %>%
  mutate(seqn=ifelse(samp=='nhanes3',paste0('33333',seqn),seqn)) #liminate duplicate sequence numbers

Training Selections and Biomarkers

We train on biological separately for men and women, using all individuals between 30 and 75 years old, excluding pregnant women. We use the year of coding as a covariate (in a dummy variable series).

For biomarkers we truncate the biomarkers by eliminating any observations that fall outside of Dan edited; need his input. We also limit the training on certain biomarkers as described in the table below.

should probably id units for the biomarkers

|Variable|Description|Additional Training Limitations, if Any| |:-------|:----------|:---------------------------| |seqn|Unique individual identifier|| |samp |Dummy varialbe series for NHANES sample year; reference is "nhanes3"|| |age|chronological age| |sex|gender (1=female; 2=male)|| |Biomarkers|| |albumin|Albumin|limited to ages 45-75| |lnalp|Alkaline Phosphate (log)|| |bun|Blood Urea Nitrogen|| |cmv|CMV||
|lncreat|Creatinine (log)|| |lncrp|CRP (log)|| |hba1c|Glycalated Hemoglobin (%)|| |lymphpct|Lymphocite Percent|limited to ages 45-75| |lnwbc|White Blood Cell Count (log)|| |lnua|Uric Acid (log)|| |sbp|Systolic Blood Pressure|| |mcv|Mean Corpuscular Volume|| |totchol|Total Cholesterol|limited to ages 30-55|

We prepare the holdout sample by extracting 1/2 of the individuals from nhanes3.

#pull list of ids and randomly sample 1/2 of them
ids = nhanes$seqn[nhanes$samp=='nhanes3']
holdout.ids = sample(ids,size=round(length(ids)/2),replace=FALSE)
#generate holdout dataset
holdout = filter(nhanes,seqn %in% holdout.ids)
train = filter(nhanes,!seqn %in% holdout.ids)
rm(ids,holdout.ids,nhanes)

We esitmate biologial age parameters separately for men and women by using the "kdm_calc" function of the bioage package (type "?kdm_calc" for help information).

#set sample control to nahnes3
#this centers estimates for years coinciding with nhanes3
train$samp = factor(train$samp)
train$samp = relevel(train$samp,ref='nhanes3')

#generate biological age estimates

female.est = kdm_calc(data=train %>% filter(sex==1),
                        agevar='age',
                        controls='samp',
                        biomarkers=c('albumin','lnalp','bun','lncreat',
                                     'lncrp','lymphpct','mcv','lnwbc',
                                     'lnua','sbp','hba1c','totchol','cmv'),
                      filter=list(albumin='age>=45',
                                  lymphpct='age>=45',
                                  totchol='age>=30 & age<=55'))

male.est = kdm_calc(data=train %>% filter(sex==2),
                        agevar='age',
                        controls='samp',
                        biomarkers=c('albumin','lnalp','bun','lncreat',
                                     'lncrp','lymphpct','mcv','lnwbc',
                                     'lnua','sbp','hba1c','totchol','cmv'),
                      filter=list(albumin='age>=45',
                                  lymphpct='age>=45',
                                  totchol='age>=30 & age<=55'))

The estimates above are saved as part of the data structure (although the data projected is set to NULL). These estimates can be drawn by typing data(male.est) or data(female.est) as the case may be.

Using these estimates, biological ages are projecte into the holdout sample by running kdm_calc again, but supplying a fit argument. In this case, the filter argument is meaningless, because we are not training parameters, so it is excluded.

female.project = kdm_calc(data=holdout %>% filter(sex==1),
                        agevar='age',
                        controls='samp',
                        biomarkers=c('albumin','lnalp','bun','lncreat',
                                     'lncrp','lymphpct','mcv','lnwbc',
                                     'lnua','sbp','hba1c','totchol','cmv'),
                        fit=female.est$fit)

male.project = kdm_calc(data=holdout %>% filter(sex==2),
                        agevar='age',
                        controls='samp',
                        biomarkers=c('albumin','lnalp','bun','lncreat',
                                     'lncrp','lymphpct','mcv','lnwbc',
                                     'lnua','sbp','hba1c','totchol','cmv'),
                        fit=male.est$fit)

#######
#pull the full dataset using the extract data command; 
#type ?extract_data for more information

holdout.withba = rbind(extract_data(female.project),
                       extract_data(male.project))

Using the projected data, we can use the newly created variable `bioage however we want. Such as identifying a scatterplot and correlation:

library(ggplot2)

ggplot(holdout.withba,
       aes(x=bioage,y=age,color=factor(sex),shape=factor(sex))) +
  geom_point(alpha=0.2) +
  ylim(20,90) + xlim(20,90) +
  geom_smooth(method='lm') +
  theme_classic()

For those wishing to use other statistical programs for analyzing biological age, I suggest the package haven from the tidyverse. For example, the following code could be used to read a training and projection dataset from stata, project biological age, and then resave to stata. (This is pseudo-code; replace "file" with the appropriate quoted string).

library(haven)

#estimate kdm parameters
training=read_dta(train.file)
train.est = kdm_calc(data=training,
                      agevar='age',
                      biomarkers=c('bm1','bm2')
                      )

#use kdm parameters
projection=read_dta(projection.file)
projection.results = kdm_calc(data=projection,
                                agevar='age',
                                biomarkers=c('bm1','bm2'),
                                fit=train.est$fit)

#save to stata for further processing
write_dta(new.projection.file)

bjb40/bioage documentation built on May 20, 2019, 3:05 p.m.