In personalscience/taster: Personal Science CGM Analysis Web App

knitr::opts_chunk$set(echo = TRUE)
knitr::opts_chunk$set(fig.height=12)
library(cgmr)
library(tastermonial)
devtools::load_all(path="/Users/sprague/dev/psi/tasterdb")
library(tidyverse)
library(lubridate)
library(knitr)
library(kableExtra)
library(showtext)
font_add_google("Montserrat")
showtext_auto()

psi_theme <-  function() {
  theme(text = element_text(family = "Montserrat",
                                         face = "bold", size = 15),
                     axis.text.x = element_text(size = 15, angle = 90, hjust = 1),
                     legend.title = element_blank())
} 

conn_args <- config::get("dataconnection")
con <- DBI::dbConnect(
  drv = conn_args$driver,
  user = conn_args$user,
  host = conn_args$host,
  port = conn_args$port,
  dbname = conn_args$dbname,
  password = conn_args$password
)

I plotted iAUC for all the brands to get an idea of the variability between testers.

GLUCOSE_RECORDS <- tbl(con, "glucose_records")  %>% filter(time > "2021-06-01") %>% collect()
NOTES_RECORDS <- tbl(con, "notes_records") %>% filter(Start > "2021-06-01") %>% collect()

foodname <- "Munk Pack"
start_limit = 100
# return all users who ate foodname
food_results <- cgmr::food_times_df(glucose_df = GLUCOSE_RECORDS,
                                    notes_df = NOTES_RECORDS,
                              prefixLength = 0,
                              timeLength = 150,
                              foodname=foodname) 



AUC_for_food <- function(foodname) {
  cgmr::auc_for_food(foodname, glucose_records = GLUCOSE_RECORDS,
                                    notes_records = NOTES_RECORDS)
}

munk_pack = AUC_for_food("Munk Pack") %>% select(meal, ave, sd, AUC, iAUC) 

build_all_AUC <- function(s_list) {
  df <- cgmr::df_for_all_auc(s_list, glucose_records = GLUCOSE_RECORDS,
                                    notes_records = NOTES_RECORDS)
  return(df %>% mutate(user_id = meal %>% stringr::str_extract("^([:digit:])+(?=-)") %>% as.numeric()))
}


all_foods <- tibble(foodname = food_list_db() ,
                    iAUC = purrr::map_dbl(food_list_db(), function(x) {AUC_for_food(x) %>% pull(iAUC) %>% mean()}),
                    n = purrr::map_dbl(food_list_db(), function(x) {AUC_for_food(x) %>% nrow()}))

auc_df <- build_all_AUC(food_list_db()) 
auc_df

all_foods %>% arrange(iAUC) %>% ggplot(aes(x=reorder(foodname, -iAUC),y=iAUC)) + 
  geom_col() + 
 annotate(geom="text", x=all_foods$foodname, y = all_foods$iAUC+ 300 , label = all_foods$n) + 
  labs(title = "Tastermonial Product Test Results", x = "", y = "Average iAUC Among All Testers") + 
  psi_theme()

Boxplot: range of data for each food

In this boxplot graph, the white rectangle represents the middle 50% of the iAUC testers found for that product. The line in the middle is the median.

auc_df %>% pivot_longer(cols=ave:iAUC, names_to = "param") %>% 
  filter(param == "iAUC") %>% 
  ggplot(aes(x=foodname, y=value)) + 
  geom_boxplot()+ 
 # geom_point( aes(x=foodname,y=value), color = "red") + 
  #geom_point( data = auc_df %>% filter(user_id == 1013), aes(x=foodname, y = iAUC), color = "blue", size = 3) + 
  #geom_facet(rows=vars(user_id)) + 
  labs(title = "Tastermonial Product Test Results", x = "", y = "iAUC Among All Testers") + 
  psi_theme()

Boxplot (including names)

auc_dfs <- auc_df %>% mutate(username = purrr::map_chr(user_id, tasterdb::name_for_user_id))

auc_dfs %>% filter(foodname != "other") %>%  pivot_longer(cols=ave:iAUC, names_to = "param") %>%
  filter(param == "iAUC") %>%
  ggplot(aes(x=foodname, y=value)) +
  geom_boxplot()+
   geom_point( aes(x=foodname,y=value), color = "red") +
   #geom_point( aes(x=foodname, y = value, color = username), size = 3) +
  geom_text(aes(x = foodname, y = value, label = username), size = 2, check_overlap = TRUE) +
  #facet_grid(rows=vars(user_id)) +
  labs(title = "Tastermonial Product Test Results", x = "", y = "iAUC Among All Testers") +
  psi_theme()

Lots of caveats: this data may be incomplete and some of my data may be incorrect due to mistaken interpretation of food start eating times)

Why is Kind Bar so low?

This is a good way to tease out possible problems in the data. For example, Kind bar looks suspiciously low iAUC, which indicates there may have been a data collection or processing issue.

kind_bar <- AUC_for_food("Kind Bar Dark Chocolate") %>% select(meal, ave, sd, AUC, iAUC) 
kind_bar %>% arrange(desc(iAUC)) %>% kable(digits = 3) %>% kableExtra::kable_styling(bootstrap_options = c("striped", "hover", "condensed"))