Continuous Glucose Monitoring with Freestyle Libre
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
README_files/figure-markdown_github
figure
librelink
.gitignore
README.Rmd
README.md
cgm.Rproj

README.md

Continuous Glucose Monitoring with Freestyle Libre

Richard Sprague 2019-01-23

See Continous Glucose Monitoring: Start Here

I've been tracking my glucose levels 24 x 7 using a continuous glucose monitor from Abbot Labs called the Freestyle Libre.

View a Shiny version of my current data at https://personalscience.shinyapps.io/librelink/. [Source]

Read (and edit!) my Continuous Glucose Monitoring Hackers Guide for details for how to get started, plus as many resources as I know about other apps and links that you might find useful for beginning your own CGM analysis.

This is a short R script I use for my analysis.


Prerequisites

Besides the working sensor, to run this script you'll need:

  • A registered account on Freestyle's official site: libreview.com
  • Data downloaded from the Libreview site. (I download it and convert to XLSX format in the file "Librelink.xlsx")
  • A separate activity file to register your food, exercise, sleep, and other events. (Another XLSX file I call "Activity.XLSX")

See examples of all my raw data files in the librelink directory.

One you have downloaded the raw Librelink data and created the activity file, you must read the results into two dataframes:

libre_raw : the raw output from a Librelink CSV file. You could just read.csv straight from the CSV if you like.

activity_raw: your file containing the metadata about whatever you'd like to track. The following script assumes you'll have variables for Sleep, Exercise, Food, and a catch-all called Event.

Now clean up the data and then set up a few other useful variables. Be careful about time zones: the raw data comes as UTC time, so you'll need to convert everything to your local time zone if you want the following charts to match.

library(tidyverse)
library(lubridate)
library(ggthemes)

activity_raw$Start <- activity_raw$Start %>% lubridate::parse_date_time(order = "ymd HMS",tz = "US/Pacific")
activity_raw$End <- activity_raw$End %>% lubridate::parse_date_time(order = "ymd HMS", tz = "US/Pacific")


glucose <- libre_raw %>% select(time = "Meter Timestamp", 
                                scan = "Scan Glucose(mg/dL)",
                                hist = "Historic Glucose(mg/dL)",
                                strip = "Strip Glucose(mg/dL)",
                                food = "Notes")

#glucose$time <- readr::parse_datetime(libre_raw$`Meter Timestamp`,locale = locale(tz="US/Pacific"))

glucose$time <- as_datetime(libre_raw$`Meter Timestamp`, tz = "US/Pacific")
# 
glucose$value <- dplyr::if_else(is.na(glucose$scan),glucose$hist,glucose$scan)

# apply correction for faulty 2019-01-08 sensor
#glucose$value <- dplyr::if_else(glucose$time>as_datetime("2019-01-08"),glucose$value+35,glucose$value)

glucose_raw <- glucose

# libre_raw$`Meter Timestamp` %>% lubridate::parse_date_time(order = "ymd HMS",tz = "US/Pacific")

Set up a few convenience functions.

# a handy ggplot object that draws a band through the "healthy" target zones across the width of any graph:
glucose_target_gg <-   geom_rect(aes(xmin=as.POSIXct(-Inf,  origin = "1970-01-01"),
                xmax=as.POSIXct(Inf,  origin= "1970-01-01"),
                ymin=100,ymax=140),
            alpha = 0.01, fill = "#CCCCCC",
            inherit.aes = FALSE)


# show glucose levels between start and end times
cgm_display <- function(start=lubridate::now()-lubridate::hours(18),
                        end=now(),
                        activity_df=activity_raw,
                        glucose_df=glucose_raw) {
  ggplot(glucose_df ,aes(x=time,y=value)) + geom_line(size=2, color = "red")+ 
  geom_point(stat = "identity", aes(x=time,y=strip), color = "blue")+
  glucose_target_gg + 
  geom_rect(data=activity_df %>% dplyr::filter(Activity == "Sleep") %>%
              select(xmin = Start,xmax = End) %>% cbind(ymin = -Inf, ymax = Inf),
            aes(xmin=xmin,xmax=xmax,ymin=ymin,ymax=ymax),
            fill="red",
            alpha=0.2,
            inherit.aes = FALSE) +
  geom_rect(data=activity_df %>% dplyr::filter(Activity == "Exercise") %>%
              select(xmin = Start,xmax = End),
            aes(xmin=xmin,xmax=xmax,ymin=-Inf,ymax=Inf),
            fill="blue",
            alpha=0.2,
            inherit.aes = FALSE) +
   geom_vline(xintercept = activity_df %>% 
               dplyr::filter(Activity == "Event" & Comment == "awake") %>% select("Start") %>% unlist(),
             color = "green") +
  geom_vline(xintercept = activity_df %>% 
               dplyr::filter(Activity == "Food") %>% select("Start") %>% unlist(),
             color = "yellow")+
  geom_text(data = activity_df %>%
              dplyr::filter(Activity == "Food") %>% select("Start","Comment") ,
            aes(x=Start,y=50, angle=90, hjust = FALSE,  label = Comment),
            size = 6) +
  labs(title = "Glucose (mg/dL)", subtitle = start) +  theme(plot.title = element_text(size=22))+
    scale_x_datetime(limits = c(start,end),
                     date_labels = "%m/%d %H:%M",
                     timezone = "US/Pacific")
  
}

# returns a dataframe giving all glucose values within "timelength" of a specific activity
food_effect <- function( foodlist = c("Oatmeal","Oatmeal w cinnamon"), activity_df = activity_raw, glucose_df = glucose_raw, timelength = lubridate::hours(2)){
  #food_df <- activity_df %>% dplyr::filter(str_detect(str_to_lower(activity_df$Comment),pattern = foodname))
  food_df <- activity_df %>% dplyr::filter(Comment %in% foodlist)
  food_df$Comment <- paste0(food_df$Comment,rownames(food_df))
  food_df_interval <- interval(food_df$Start,food_df$Start + hours(1))
  food_glucose <- glucose_df %>% dplyr::filter(apply(sapply(glucose_df$time,function(x) x %within% food_df_interval),2,any))
 # food_glucose <- glucose_df %>% dplyr::filter(sapply(glucose_df$time,function(x) x %within% food_df_interval))
  f <- cbind(food_glucose[1,],experiment = "test")
  
  a = NULL
  
  for(i in food_df$Start){
    i_time <- as_datetime(i, tz = "US/Pacific")
    # < rbind(i,a)
    g <- glucose_df %>% dplyr::filter(time %within% interval(i_time - minutes(10), i_time + timelength))
    #print(g)
    p = match(as_datetime(i),food_df$Start)
    f <- rbind(f,cbind(g,experiment = food_df$Comment[p]))
  }
  foods_experiment <- f[-1,]
  foods_experiment
}

View the last couple days of the dataset:

startDate <- now() - days(2) #min(glucose$time)

#cgm_display(startDate,now()-days(6))

cgm_display(startDate,startDate + days(2))

Here's just for a single day. Note that the commented-out lines will let you output to a PDF file if you like.

#pdf("icecream.pdf", width = 11, height = 8.5)
cgm_display(start = min(glucose_raw$time),min(glucose_raw$time)+hours(24))

#dev.off()

The final full day of the dataset:

cgm_display(start = max(glucose_raw$time)-days(1), end = max(glucose_raw$time))

Food types

Here's how I look when eating specific foods:

## Scale for 'x' is already present. Adding another scale for 'x', which
## will replace the existing scale.

Basic Statistics

What is my average glucose level while sleeping?

library(lubridate)

options(scipen = 999)

# all sleep intervals, using the lubridate function lubridate::interval
# sleep_intervals <- interval(activity_raw %>% dplyr::filter(Activity == "Sleep") %>% select(Start)
#                             %>% unlist() %>% as_datetime(),
#                             activity_raw %>% dplyr::filter(Activity == "Sleep") %>% select(End) 
#                             %>% unlist() %>% as_datetime())

activity_intervals <- function(activity_raw_df, activity_name){
  interval(activity_raw_df %>% dplyr::filter(Activity == activity_name) %>% select(Start)
                            %>% unlist() %>% as_datetime(),
                            activity_raw_df %>% dplyr::filter(Activity ==activity_name) %>% select(End) 
                            %>% unlist() %>% as_datetime())
}



glucose %>% filter(apply(sapply(glucose$time,
                                function(x) x %within% activity_intervals(activity_raw,"Sleep")),2,any)) %>% select(value) %>% 
  DescTools::Desc(main = "Glucose Values While Sleeping")
## ------------------------------------------------------------------------- 
## Describe . (tbl_df, tbl, data.frame):
## 
## data.frame:  869 obs. of  1 variables
## 
##   Nr  ColName  Class    NAs  Levels
##   1   value    numeric  .          
## 
## 
## ------------------------------------------------------------------------- 
## Glucose Values While Sleeping
## 
##   length       n    NAs  unique     0s   mean  meanCI
##      869     869      0      81      0  80.02   79.11
##           100.0%   0.0%           0.0%          80.94
##                                                      
##      .05     .10    .25  median    .75    .90     .95
##    55.00   64.00  73.00   81.00  87.00  96.00  102.00
##                                                      
##    range      sd  vcoef     mad    IQR   skew    kurt
##    92.00   13.75   0.17   10.38  14.00  -0.05    1.20
##                                                      
## lowest : 40.0 (3), 41.0 (4), 42.0 (3), 43.0 (2), 44.0
## highest: 120.0, 121.0, 125.0, 131.0, 132.0 (2)

glucose %>% filter(apply(sapply(glucose$time,
                                function(x) !(x %within% activity_intervals(activity_raw,"Sleep"))),
                         2,
                         any)) %>% select(value) %>% 
  DescTools::Desc(main = "Glucose Values While Awake")
## ------------------------------------------------------------------------- 
## Describe . (tbl_df, tbl, data.frame):
## 
## data.frame:  3179 obs. of  1 variables
## 
##   Nr  ColName  Class    NAs        Levels
##   1   value    numeric  17 (0.5%)        
## 
## 
## ------------------------------------------------------------------------- 
## Glucose Values While Awake
## 
##   length      n    NAs  unique      0s    mean  meanCI
##    3'179  3'162     17     120       0   90.93   90.27
##           99.5%   0.5%            0.0%           91.59
##                                                       
##      .05    .10    .25  median     .75     .90     .95
##    64.00  70.00  79.00   89.00  101.00  116.00  126.00
##                                                       
##    range     sd  vcoef     mad     IQR    skew    kurt
##   151.00  18.84   0.21   16.31   22.00    0.62    1.17
##                                                       
## lowest : 40.0 (3), 41.0 (4), 42.0 (3), 43.0 (2), 44.0 (5)
## highest: 166.0 (2), 172.0, 187.0, 188.0, 191.0

glucose %>% filter(apply(sapply(glucose$time,
                                function(x) x %within% activity_intervals(activity_raw,"Exercise")),2,any)) %>% select(value) %>% 
  DescTools::Desc(main = "Glucose Values While Exercising")
## ------------------------------------------------------------------------- 
## Describe . (tbl_df, tbl, data.frame):
## 
## data.frame:  34 obs. of  1 variables
## 
##   Nr  ColName  Class    NAs       Levels
##   1   value    numeric  1 (2.9%)        
## 
## 
## ------------------------------------------------------------------------- 
## Glucose Values While Exercising
## 
##   length      n    NAs  unique      0s    mean  meanCI
##       34     33      1      28       0   91.76   85.33
##           97.1%   2.9%            0.0%           98.18
##                                                       
##      .05    .10    .25  median     .75     .90     .95
##    69.60  71.20  78.00   87.00  100.00  114.80  120.80
##                                                       
##    range     sd  vcoef     mad     IQR    skew    kurt
##    77.00  18.12   0.20   19.27   22.00    0.77    0.13
##                                                       
## lowest : 67.0, 69.0, 70.0, 71.0, 72.0
## highest: 110.0, 116.0, 120.0, 122.0, 144.0