Accessing Apple Health Data

class: center, middle, inverse, title-slide

# Accessing Apple Health Data
## Visualising data measured from activity watches
### Heidi Thornton, PhD
### 2020/05/15

---

background-image: url("Images/Watch.png")
background-size: 30%
background-position: 50% 95%

# Alternative Tracking Method

#### As sports scientists, during this COVID-19 shut, we have limmited ways of determining adherance or completion of training programs (which of course are completed within Govenment guidelines)!
--

As a **guide** to measure the training completed during this period is via activity watches
--

Whilst the reliability and validity of activity watches is not widely available, these measure basic metrics such as total distance, duration, HR, energy burnt etc
--

Here, I will run through how to export this data without an API, basic analysis and manipulation and will demonstrate a few plots using ***ggplot***

---

background-image: url("Images/Steps.png")
background-size: 60%
background-position: 50% 95%

# Accessing Apple Health Data

**Step 1** - Click on apple health app on your phone, on summary page in the top right, click on circle with first name letter (figure on left)
--

**Step 2** - Slide down to ‘Export All Health Data’ (middle figure)
--

**Step 3** - This export will take up to a few minutes, then a page to email/message etc it will pop up. Simply email to yourself and download it

---

background-image: url("Images/Heart.png")
background-size: 15%
background-position: 95% 3%

# Opening Apple Health Data

You can open the zip folder directly using R, however for this project I will extract the file that is within the folder'apple_health_export'.
--

Inside the 'apple_health_export' folder, there will be 2 files. For the purpose of this, you will only need the ‘export’ file
--

I have made a new folder that houses all files where I will set my working directly to this location and load packages. If you don't have these packages installed use install.packages("packagename")

```r
setwd("C:/Users/Heidi.Thornton/Desktop/Apple")
```

```r
library("XML")
library("methods")
library("tidyverse")
library("lubridate")
library("ggplot2")
library("dplyr")
```

---

background-image: url("Images/R.png")
background-size: 20%
background-position: 50% 90%

# Why do this in R?

###R is reproducible - microsoft excel is not...
--

####We could simply export the raw data into excel and manipulate and visualise it there
--

####But...R isn't that difficult and there are lots of resources out there to help learn it
--

### Have the **end game** in mind

---

# Import the Data into R

The file format (XML; Extensible Markup Language) is quite easy to work with in R using the **XML** package
--

I have created a folder where I have put my xml file. If you have multiple files you can use a loop to access them all - I will only use one file though
--

First, we need to make an xml object and view it's contents using summary(xml)
--

```r
xml <- xmlParse(paste("xml/Thornton, Heidi.xml"))
summary(xml)
```

```
## $nameCounts
## 
##        Record       Workout MetadataEntry    ExportDate    HealthData 
##        200264           202            70             1             1 
##            Me 
##             1 
## 
## $numNodes
## [1] 200539
```

---

# View the Data

I'm interested in the workout data which we can open using **xmlAttrsToDataFrame**
--

```r
df_workout <- XML:::xmlAttrsToDataFrame(xml["//Workout"])[c(1:2,4,6,12)]
head(df_workout, n = 5) # View the data
```

```
##            workoutActivityType          duration    totalDistance
## 1 HKWorkoutActivityTypeRunning 42.44751790364583 8.01347998046875
## 2 HKWorkoutActivityTypeCycling 45.69051513671875    11.1736796875
## 3   HKWorkoutActivityTypeOther 45.29886474609375 3.60589990234375
## 4   HKWorkoutActivityTypeOther 58.97060139973959  1.7677099609375
## 5 HKWorkoutActivityTypeRunning 39.55806477864584 4.17927978515625
##   totalEnergyBurned                   endDate
## 1          2041.792 2019-11-23 06:40:35 +1000
## 2          1096.208 2019-11-24 06:56:26 +1000
## 3             836.8 2019-11-24 15:27:41 +1000
## 4           736.384 2019-11-25 05:27:23 +1000
## 5 669.4400000000001 2019-11-25 16:42:42 +1000
```

Here we have the session 'type' and the respective data for each day which needs some cleaning up

---

# Plotting Data

Now lets start with my running sessions which we need to filter
--

We will use %>% (pipes) from the **tidyverse** package as it is much quicker rather than making new data frames and will plot the data using **ggplot2**

```r
df_workout %>%
  
# Change data types (i.e. distance to m not km, numeric) 
    mutate(workoutActivityType = as.character(workoutActivityType),
        totalDistance = as.numeric(as.character(totalDistance))*1000, # convert to m not km
        duration = as.numeric(duration), # Change to numeric 
        endDate = as.Date(endDate)) %>%  # change date from factor format
# Only running sessions- depending on watch the name may differ
      filter(workoutActivityType == "HKWorkoutActivityTypeRunning") %>%
      filter(endDate >= "2020-03-23") %>% # only after shut down

# Create ggplot    
ggplot(aes(x= endDate, y = totalDistance)) +
    geom_bar(stat="identity", fill='#5ab4ac')+
      labs(title = "Running Sessions") +  # title
      ylab(bquote(bold("Total distance (m)"))) + # y axis title
      xlab(bquote(bold("Date")))+ # x axis title
      scale_x_date(date_breaks = "7 days") + 
        theme_minimal()  # remove some theme formatting
```
  
  
---

# Daily Running Sessions

Now we have our first plot, showing running volume (m) by day, including days after the COVID shut down. Not exactly periodised, but it's better than nothing....
--

---

# Weekly Running Volume

We need to manipulate the data a bit more to get the weekly running volume
--

```r
df_workout %>%
   mutate(workoutActivityType = as.character(workoutActivityType),
    totalDistance = as.numeric(as.character(totalDistance))*1000,
    endDate = as.Date(endDate),
    week = isoweek(ymd(endDate))) %>% # Add week column
      filter(workoutActivityType == "HKWorkoutActivityTypeRunning") %>%
      filter(endDate >= "2020-03-23")%>%
          group_by(week)%>% # summarise by week (starts week in 1st jan)
ggplot(aes(x=week, y=totalDistance)) + 
  geom_bar(stat="identity", fill='#5ab4ac')+labs(title = "Running Sessions")+
  ylab(bquote(bold("Total distance (m)"))) +
  xlab(bquote(bold("Week")))+
  theme_minimal()
```

---

# Energy Consumption

I want to know energy consumption by **activity type** which we need to rename
--

```r
df_workout %>%
   mutate(workoutActivityType = as.character(workoutActivityType),
      totalEnergyBurned = as.numeric(as.character(totalEnergyBurned)),
      endDate = as.Date(endDate), week = isoweek(ymd(endDate)), # Add week column
      Type = str_sub(workoutActivityType, 22)) %>% # new column- text after 22nd character
        filter(endDate >= "2020-03-23" & 
          Type %in% c('Running','Other')) %>% # Filter out cycling
            group_by(week) %>%
  
ggplot(aes(x=week, y=totalEnergyBurned, fill=Type)) + 
     geom_bar(stat="identity", position=position_dodge())+
       facet_wrap(~Type)+
       labs(title = "Energy Consumption by Session Type")+
       ylab(bquote(bold("Energy Consumption (kj)"))) +
       xlab(bquote(bold("Week")))+
       scale_x_continuous( breaks = seq(13, 19, by = 1)) +
         theme_minimal() + 
         theme(legend.position = "none")  
```
---

# Energy Consumption Plot

Now we can see weekly energy consumption by **activity type**. 'Other' sessions include weights or walking  
--

---

# General Activity Data

Lets move on from workout data and import 'Record' data. This one will take a fair while to load

```r
df_record <- XML:::xmlAttrsToDataFrame(xml["//Record"]) [c(1,6,8)]
```

```r
# See data types available in record
  df_record %>% 
    mutate(Type = str_sub(type, 25)) %>% # Include text after the 25th character
    select(Type) %>% distinct
```

```
##                      Type
## 1            DietaryWater
## 2                  Height
## 3                BodyMass
## 4               HeartRate
## 5               StepCount
## 6  DistanceWalkingRunning
## 7       BasalEnergyBurned
## 8      ActiveEnergyBurned
## 9          FlightsClimbed
## 10       RestingHeartRate
## 11 HeadphoneAudioExposure
## 12          SleepAnalysis
```
--

If you want to view the full dataset, you can use **head(df_record)**

---

# Steps Data Manipulation

This one isn't exactly useful for athletes - this is more for my own interest of my activity (or lack of) during the COVID shut down
I am replicating a plot created [online](https://taraskaduk.com/2019/03/23/apple-health/), demonstrating step count  
--

```r
df_record %>%
mutate(Type = str_remove(type, "HKQuantityTypeIdentifier"), # Rename type 
       value = as.numeric(as.character(value)),
       Date = as.Date.character(startDate),weekday = wday(Date), # Day of week
       hour = hour(startDate)) %>% # Need to use the factor date
   filter(Type == 'StepCount' & Date >= "2020-03-23") %>%
     group_by(Date, weekday, hour) %>%  # Summarise by date, weekday and hour
       summarise(value = sum(value)) %>% # Sum steps over ^^
     group_by(weekday, hour) %>% # Now summarise by weekday and hour  
       summarise(value = mean(value)) %>% # Take mean steps over ^^
   filter(between(hour,6,21)) %>% # Filtering to include between 6am - 9pm
 
 ggplot(aes(x=hour, y=weekday,  fill=value)) + 
     geom_tile(col = 'grey40') + 
     scale_fill_continuous(labels = scales::comma, low='grey95',high ='#008FD5') +
      scale_x_continuous(breaks = c(6,9,12,15,18), 
         label = c("6 AM","9 AM", "Midday", "3PM", "6 PM")) +
     scale_y_reverse(breaks = c(1,2,3,4,5,6,7),
      label =c("Sunday","Monday","Tuesday","Wednesday","Thursday","Friday","Saturday"))+
     labs(title = "Step Count Heatmap") +
     ylab(bquote(bold("Weekday"))) + xlab(bquote(bold("Hour")))+
     guides(fill=FALSE)+
     coord_equal()+
   theme_minimal()
```

---

# Steps by Hour by Day Heatmap

Not a lot of activity at the moment....

---

# Heart Rate

One last one - we will plot HR across 2 days. I have added a colour scale for low (green) and high (red) HR
--

.pull-left[

```r
df_record %>%
 mutate(Type = str_remove(type, "HKQuantityTypeIdentifier"),
 value = as.numeric(as.character(value)), 
 startDate = as_datetime(startDate),
 Date = as.Date.character(startDate)) %>% 
 filter(Type == 'HeartRate') %>%
 filter(Date >= as.Date("2020-04-03") 
 & Date <= as.Date("2020-04-04")) %>%
 
ggplot(aes(x=startDate, y=value,colour=value))+
 geom_line(size=1) +
 scale_color_gradient(low="green", high="red")+
 labs(title = "Heart Rate") +
 ylab(bquote(bold("Heart Rate"))) + 
 xlab(bquote(bold("Date/Time")))+
 theme_minimal()
```

]

.pull-right[
![](Presentation_files/figure-html/HR Graph-1.png)
]

---

# Access More Apple Health Info

There are other data types available from Apple Health

```apple
df_record <- XML:::xmlAttrsToDataFrame(xml["//Record"])
df_activity <- XML:::xmlAttrsToDataFrame(xml["//ActivitySummary"])
df_workout <- XML:::xmlAttrsToDataFrame(xml["//Workout"])
df_clinical <- XML:::xmlAttrsToDataFrame(xml["//ClinicalRecord"])
df_location <- XML:::xmlAttrsToDataFrame(xml["//Location"])
```

For more information on analysing Apple Health data, check out;

####["Analyze and visualize your iPhone's Health app data in R"](https://taraskaduk.com/2019/03/23/apple-health/)

####["Explore your Apple Watch heart rate data in R"](https://jeffjjohnston.github.io/rstudio/rmarkdown/2016/04/28/explore-your-apple-watch-heart-rate-data.html)

---

# Thanks for looking! 😊

#### If you want to learn more about R, there is some awesome work out there from fellow Aussies! Click on the names to view

####[Alice Sweeting](http://sportstatisticsrsweet.rbind.io/)

####[Mitch Henderson](https://www.mitchhenderson.org/)

####[Jacquie Tran](https://www.jacquietran.com/)

<a href="mailto:heidi.thornton@goldcoastfc.com.au">
.white[<heidi.thornton@goldcoastfc.com.au]
</a>

<a href="http://twitter.com/heidithornton09">
.white[<@heidithornton09]
</a>