title

author

date

output

fitbit analysis for CSEP project

Shelby Sturrock

14/01/2022

html_document

keep_md
true

Fitbit analysis

This script contains preliminary figures and data analysis for Fitbit data.

daily <- read.csv("daily-TOP-CSEP.csv")
str(daily$date)

##  chr [1:3603] "2021-10-19" "2021-10-20" "2021-10-21" "2021-10-22" ...

Make sure that the intervention variable and the weekday variables are formatted as factors where the intervention periods and weekdays are ordered correctly (pre, intervention, post; monday, tuesday, etc.)

  daily$weekday<-factor(daily$weekday,
                        levels=c("Monday","Tuesday","Wednesday","Thursday","Friday","Saturday","Sunday"))
  daily$intervention<-factor(daily$intervention,levels=c("pre","intervention","post"))
  daily$date<-as.Date(daily$date)
  str(daily$weekday)

##  Factor w/ 7 levels "Monday","Tuesday",..: 2 3 4 5 6 7 1 2 3 4 ...

  str(daily$intervention)

##  Factor w/ 3 levels "pre","intervention",..: 1 1 1 1 1 1 2 2 2 2 ...

Create person-day-level for plots. Currently, the "daily" dataframe has one row per person-day-article, so if someone engages with two different articles, they have two rows in the dataframe. I want a maximum of one row per person per day for data exploration.

  dailyForPlots<-daily[!is.na(daily$totalMVPA),] %>% select(-title,-intervention_content,-featured_content,
                                                            -pubdate,-weekofpublish,
                                                            -type,-click,-engage,-clickInt,-engageInt)
  dailyForPlots<-distinct(dailyForPlots)
# tentative: remove particiapnt-days with less than 10 hours of wear time (so 14 hours or more of non-wear time)
  # 14 * 60 = 840 (840 minutes or more of non-wear time)

Remove days where time in bed = 0

  dailyForPlots<-dailyForPlots[dailyForPlots$TotalTimeInBed!=0,]

Remove days with < 10 hours of wear time (14+ hours of non-wear time)

  dailyForPlots<-dailyForPlots[dailyForPlots$nonWearMinutes < 840,]

Explore total minutes of data contributed per particiant day. Should be ~1440 minutes

  dailyForPlots$test<-dailyForPlots$totalMVPA+dailyForPlots$LightlyActiveMinutes+dailyForPlots$SedentaryMinutes+
                      dailyForPlots$TotalMinutesAsleep
  dailyForPlots$test2<-dailyForPlots$totalMVPA+dailyForPlots$LightlyActiveMinutes+dailyForPlots$SedentaryMinutes+
                       dailyForPlots$TotalTimeInBed
  totalMinutesAsleep<-ggplot(dailyForPlots,aes(x=test)) + 
                        geom_bar(fill="#619CFF") + 
                        theme_classic() +
                        labs(x="Total minutes of data\n(MVPA + light + sedentary + time asleep)",
                             y="Frequency",
                             title="Minutes of data per day and\n participant") +
                        theme(axis.title.x = element_text(face="bold"),
                              axis.title.y = element_text(face="bold"),
                              plot.title = element_text(face="bold",
                                                        hjust=0.5))
  totalMinutesInBed<-ggplot(dailyForPlots,aes(x=test2)) + 
                        geom_bar(fill="#619CFF") + 
                        theme_classic() +
                        labs(x="Total minutes of data\n(MVPA + light + sedentary + time in bed)",
                             y="Frequency",
                             title="Minutes of data per day and\n participant") +
                        theme(axis.title.x = element_text(face="bold"),
                              axis.title.y = element_text(face="bold"),
                              plot.title = element_text(face="bold",
                                                        hjust=0.5))
  grid.arrange(totalMinutesAsleep,totalMinutesInBed,ncol=2)

Exploratory Analysis

Engagement and clicks:

Total clicks and engagement per day (summed across all participants)

Total engagements per day, by date

Total clicks and engagements, when summed across participants, appear somewhat cyclical, though the value and the amplitude both appear to diminish over time.

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Clicks and engagement per participant-day

Distribution of clicks and engagement per participant and day

The vast majority of participant-days had 0 clicks and 0 engagement.

Distribution of clicks and engagement per participant (across the entire study period)

Can participants be grouped based on how often they engaged over the study period?

Number of clicks and engagement per participant and day over time

It is hard to make out any interesting patterns because there are so many participant-days with 0 engagement and/or clicks (as observed above)

Mean daily clicks and engagement over time

It is easier to see trends in daily clicks and engagement over time using mean clicks and engagement per day (calculated by taking the mean of both countClick (count of clicks per person per day) and countEngage (count of engagement per person per day) across all participants for each day in the study period. Please note, we cannot use median, as the median clicks and median engagement is 0 everyday.

## [1] TRUE

## [1] TRUE

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'
## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Clicks and engagement by weekday in the pre, intervention and post intervention period

I am interested in assessing if daily clicks and/or engagement (at the participant level) varies by day of the week. It looks like clicks and engagement are higher during the work week, particularly during the intervention period, but it is difficult to tell because there are also more Mondays, Tuesday, etc. during this period (since it is much longer).

Mean clicks and engagement, by weekday and period

Mean clicks and engagement is calculated by taking the mean daily click and/or engagement count across all participants, separately for each weekday, in each of the pre-intervention, intervention and post-intervention periods. It appears that the mean clicks and mean engagement by day of the week differ between the pre-intervention, intervention and post-intervention periods, but trends look similar for both clicks and engagements.

Total time use

MVPA

Distribution/frequency of MVPA - with and without including 0s

First, lets assess the distribution of MVPA across participant-days. Similar to the engagement metrics, minutes of MVPA is 0 on the majority of participant days. It is easier to see the distribution if you exclude participant-days with 0 minutes if MVPA (right).

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    0.00   10.00   32.00   43.13   62.00  447.00

Plot of minutes of MVPA per day over time -- overall and comparing engagers and non-engagers

Next, lets assess how daily MVPA varies over time. There is no obvious trend over time, though there is a slight bow to the loess line that appears to correspond (roughly) to the intervention period.

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

When you fit a loess curve to each period, daily MVPA appears to "ramp-up" during the pre-intervention period and remain ~stable throughout the intervention period.

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Interestingly, this "bowing" of the loess curve appears to be driven by participants that engaged with the app content (at least once); a light downward trend is observed among non-engagers (not pictured)

Boxplot of MVPA per day and participant, by week

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

LPA

Distribution/frequency of LPA - with and without including 0s

First, lets assess the distribution of LPA across participant-days. Similar to the engagement metrics, minutes of LPA is 0 on the majority of participant days. It is easier to see the distribution if you exclude participant-days with 0 minutes if LPA (right).

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##     4.0   147.8   197.0   205.5   254.0   547.0

Plot of minutes of LPA per day over time -- overall and comparing engagers and non-engagers

Next, lets assess how daily LPA varies over time. There is no obvious trend over time, though there is a slight bow to the loess line that appears to correspond (roughly) to the intervention period.

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Similar to MVPA, LPA ramped up during the pre-intervention period and was sustained throughout the intervention period.

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Boxplot of LPA per day and participant, by week

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Sedentary Behaviour

Distribution/frequency of SB - with and without including 0s

First, lets assess the distribution of SB across participant-days. Similar to the engagement metrics, minutes of SB is 0 on the majority of participant days. It is easier to see the distribution if you exclude participant-days with 0 minutes if SB (right).

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##     0.0   648.0   735.5   714.2   817.2  1247.0

Plot of minutes of SB per day over time -- overall and comparing engagers and non-engagers

Next, lets assess how daily SB varies over time. There is no obvious trend over time, though there is a slight bow to the loess line that appears to correspond (roughly) to the intervention period.

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Similar to MVPA, SB ramped up during the pre-intervention period and was sustained throughout the intervention period.

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Boxplot of SB per day and participant, by week

## Warning: Ignoring unknown parameters: outliers

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Sleep

Distribution/frequency of Sleep - with and without including 0s

First, lets assess the distribution of Sleep across participant-days. Similar to the engagement metrics, minutes of Sleep is 0 on the majority of participant days. It is easier to see the distribution if you exclude participant-days with 0 minutes if Sleep (right).

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    83.0   395.0   455.0   456.3   517.2   880.0

Plot of minutes of Sleep per day over time -- overall and comparing engagers and non-engagers

Next, lets assess how daily Sleep varies over time. There is no obvious trend over time, though there is a slight bow to the loess line that appears to correspond (roughly) to the intervention period.

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Similar to MVPA, Sleep ramped up during the pre-intervention period and was sustained throughout the intervention period.

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Boxplot of SB per day and participant, by week

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

High-level: All movement behaviours over time

Plot of all movement behaviours over time

## Before During  After 
##   1192  10020   1796

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Plots

Line Graph - with engagement and clicks overlayed

Any clicks across all three periods

Re-scale clicks and engagement for plots of each of the four movement behaviours

# MVPA
  mean(dailyForPlots$totalMVPA) # 447

## [1] 43.13223

  dailyForPlots$clickMVPA<-dailyForPlots$dayCountClickAny*(mean(dailyForPlots$totalMVPA)/mean(dailyForPlots$dayCountClickAny))
  dailyForPlots$engageMVPA<-dailyForPlots$dayCountEngageAny*(mean(dailyForPlots$totalMVPA)/mean(dailyForPlots$dayCountEngageAny))

# Lightly Active Minutes
  dailyForPlots$clickLPA<-dailyForPlots$dayCountClickAny*
                          (mean(dailyForPlots$LightlyActiveMinutes)/mean(dailyForPlots$dayCountClickAny))
  dailyForPlots$engageLPA<-dailyForPlots$dayCountEngageAny*
                           (mean(dailyForPlots$LightlyActiveMinutes)/mean(dailyForPlots$dayCountEngageAny))
  
# Sedentary Behavior
  dailyForPlots$clickSB<-dailyForPlots$dayCountClickAny*
                          (mean(dailyForPlots$SedentaryMinutes)/mean(dailyForPlots$dayCountClickAny))
  dailyForPlots$engageSB<-dailyForPlots$dayCountEngageAny*
                          (mean(dailyForPlots$SedentaryMinutes)/mean(dailyForPlots$dayCountEngageAny))
  
# Sleep 
  dailyForPlots$clickSleep<-dailyForPlots$dayCountClickAny*
                            (mean(dailyForPlots$TotalTimeInBed)/mean(dailyForPlots$dayCountClickAny))
  dailyForPlots$engageSleep<-dailyForPlots$dayCountEngageAny*
                            (mean(dailyForPlots$TotalTimeInBed)/mean(dailyForPlots$dayCountEngageAny))

Create a time variable (where Oct 18 = 1, Oct 19 = 2, etc.)

dates<-dailyForPlots %>% select(date,intervention) %>% distinct() %>% arrange(date) %>% mutate(nDate=row_number())

MVPA

# Regression model - clicks

# Regression model - MVPA

  
# Figure
  lineIntMVPA<-ggplot(dailyForPlots,aes(x=date,y=totalMVPA))+
              # clicks
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="pre",],
                            method="lm",formula="y~x",aes(x=date,y=clickMVPA),color="black") +
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="intervention",],
                            method="lm",formula="y~x",aes(x=date,y=clickMVPA),color="black") +
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="post",],
                            method="lm",formula="y~x",aes(x=date,y=clickMVPA),color="black") +
              # MVPA
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="pre",],
                            method="lm",formula="y~x",color="#619CFF",fill="#619CFF") +
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="intervention",],
                            method="lm",formula="y~x",color="#619CFF",fill="#619CFF") +
                geom_smooth(data=dailyForPlots[dailyForPlots$intervention=="post",],
                            method="lm",formula="y~x",color="#619CFF",fill="#619CFF") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="MVPA") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyForPlots$totalMVPA)/mean(dailyForPlots$dayCountClickInt)), name = "Clicks"))
  lineIntMVPA

  lineMVPA<-ggplot(dailyForPlots,aes(x=date,y=totalMVPA))+
               #geom_jitter(color="#619CFF",alpha=0.15) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickMVPA),color="black",alpha=0.15) +
               geom_smooth(data=dailyForPlots,aes(x=date,y=clickMVPA),color="black") +
               geom_smooth(color="#619CFF",fill="#619CFF") +
            # lines for clicks
               #geom_smooth(data=dailyForPlots,method="lm",formula="y~x",aes(x=date,y=clickMVPA),color="black") +
            # lines for MVPA
               #geom_smooth(method="lm",formula="y~x",color="#619CFF",fill="#619CFF") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="MVPA") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyForPlots$totalMVPA)/mean(dailyForPlots$dayCountClickAny)), name = "Clicks"))
  lineMVPA

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Lightly active minutes

# Figure
  lineLPA<-ggplot(dailyForPlots,aes(x=date,y=LightlyActiveMinutes))+
               #geom_jitter(color="#81DA66",alpha=0.25) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickLPA),color="black",alpha=0.25) +
               geom_smooth(data=dailyForPlots,aes(x=date,y=clickLPA),color="black") +
               geom_smooth(color="#81DA66",fill="#81DA66") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Light intensity activity") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyForPlots$LightlyActiveMinutes)/mean(dailyForPlots$dayCountClickAny)), 
                        name = "Clicks"))
  lineLPA

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Sedentary minutes

# Regression model - clicks

# Regression model - Sedentary minutes

# Figure
  lineSB<-ggplot(dailyForPlots,aes(x=date,y=SedentaryMinutes))+
               #geom_jitter(color="#FFC833",alpha=0.25) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickSB),color="black",alpha=0.25) +
               geom_smooth(data=dailyForPlots,aes(x=date,y=clickSB),color="black") +
               geom_smooth(color="#FFC833",fill="#FFC833") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Sedentary time") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyForPlots$SedentaryMinutes)/mean(dailyForPlots$dayCountClickAny)), 
                        name = "Clicks"))
  lineSB

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Sleep minutes

# Regression model - clicks

# Regression model - Sleep

# Figure
  lineSleep<-ggplot(dailyForPlots,aes(x=date,y=TotalTimeInBed))+
               #geom_jitter(color="#FF8333",alpha=0.25) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickSleep),color="black",alpha=0.25) +
               geom_smooth(data=dailyForPlots,aes(x=date,y=clickSleep),color="black") +
               geom_smooth(color="#FF8333",fill="#FF8333") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Time in bed") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyForPlots$TotalTimeInBed)/mean(dailyForPlots$dayCountClickAny)), 
                        name = "Clicks"))
  lineSleep

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

Panel figure

 linesSave<-grid.arrange(lineMVPA,lineLPA,lineSB,lineSleep,ncol=2)

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

 linesSave

## TableGrob (2 x 2) "arrange": 4 grobs
##   z     cells    name           grob
## 1 1 (1-1,1-1) arrange gtable[layout]
## 2 2 (1-1,2-2) arrange gtable[layout]
## 3 3 (2-2,1-1) arrange gtable[layout]
## 4 4 (2-2,2-2) arrange gtable[layout]

Intervention clicks during intervention period

Generate click and engagement variables that are rescaled such that the mean = mean of the respective behaviour for plotting

  dailyInt<-dailyForPlots[dailyForPlots$intervention=="intervention",]

# MVPA
  mean(dailyInt$totalMVPA)

## [1] 44.58323

  dailyInt$clickIntMVPA<-dailyInt$dayCountClickInt*
                               (mean(dailyInt$totalMVPA)/mean(dailyInt$dayCountClickInt))
  dailyInt$engageIntMVPA<-dailyInt$dayCountEngageAny*
                               (mean(dailyInt$totalMVPA)/mean(dailyInt$dayCountEngageAny))

# Lightly Active Minutes
  dailyInt$clickIntLPA<-dailyInt$dayCountClickInt*
                          (mean(dailyInt$LightlyActiveMinutes)/mean(dailyInt$dayCountClickInt))
  dailyInt$engageIntLPA<-dailyInt$dayCountEngageAny*
                           (mean(dailyInt$LightlyActiveMinutes)/mean(dailyInt$dayCountEngageAny))
  
# Sedentary Behavior
  dailyInt$clickIntSB<-dailyInt$dayCountClickInt*
                          (mean(dailyInt$SedentaryMinutes)/mean(dailyInt$dayCountClickInt))
  dailyInt$engageIntSB<-dailyInt$dayCountEngageAny*
                          (mean(dailyInt$SedentaryMinutes)/mean(dailyInt$dayCountEngageAny))
  
# Sleep 
  dailyInt$clickIntSleep<-dailyInt$dayCountClickInt*
                            (mean(dailyInt$TotalTimeInBed)/mean(dailyInt$dayCountClickInt))
  dailyInt$engageIntSleep<-dailyInt$dayCountEngageAny*
                            (mean(dailyInt$TotalTimeInBed)/mean(dailyInt$dayCountEngageAny))

Create a time variable (where Oct 18 = 1, Oct 19 = 2, etc.)

datesInt<-dailyInt %>% select(date,intervention) %>% distinct() %>% arrange(date) %>% mutate(nDate=row_number())
dailyInt<-merge(dailyInt,datesInt,by=c("date"))

MVPA

# Regression model - clicks
  clicksIntMVPA<-lm(clickIntMVPA~nDate+I(nDate^2),data=dailyInt)
  dailyInt$predClicksMVPA<-predict(clicksIntMVPA,dailyInt)

# Regression model - MVPA
  intMVPA<-lm(totalMVPA~poly(nDate,degree=2,raw=TRUE),data=dailyInt)
  dailyInt$predMVPA<-predict(intMVPA,dailyInt)

# Figure
  lineIntMVPA<-ggplot(dailyInt,aes(x=date,y=totalMVPA))+
               #geom_jitter(color="#619CFF",alpha=0.15) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickMVPA),color="black",alpha=0.15) +
               #geom_smooth(data=dailyInt,aes(x=date,y=clickIntMVPA),color="black") +
               geom_smooth(data=dailyInt,method="lm",formula="y~x",aes(x=date,y=clickIntMVPA),color="black") +
               #geom_line(data=dailyInt,aes(x=date,y=predClicksMVPA),color="black") +
               #geom_smooth(color="#619CFF",fill="#619CFF") +
               geom_smooth(method="lm",formula="y~x",color="#619CFF",fill="#619CFF") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="MVPA") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyInt$totalMVPA)/mean(dailyInt$dayCountClickInt)), name = "Clicks"))
  lineIntMVPA

Lightly active minutes

# Regression model - clicks
  clicksIntLPA<-lm(clickIntLPA~nDate+I(nDate^2),data=dailyInt)
  dailyInt$predClicksLPA<-predict(clicksIntLPA,dailyInt)

# Regression model - LPA
  intLPA<-lm(LightlyActiveMinutes~poly(nDate,degree=2,raw=TRUE),data=dailyInt)
  dailyInt$predLPA<-predict(intLPA,dailyInt)

# Figure
  lineIntLPA<-ggplot(dailyInt,aes(x=date,y=LightlyActiveMinutes))+
               #geom_jitter(color="#619CFF",alpha=0.15) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickLPA),color="black",alpha=0.15) +
               #geom_smooth(data=dailyInt,aes(x=date,y=clickIntLPA),color="black") +
               geom_smooth(data=dailyInt,method="lm",formula="y~x",aes(x=date,y=clickIntLPA),color="black") +
               #geom_line(data=dailyInt,aes(x=date,y=predClicksLPA),color="black") +
               #geom_smooth(color="#619CFF",fill="#619CFF") +
               geom_smooth(method="lm",formula="y~x",color="#81DA66",fill="#81DA66") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Light intensity activity") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyInt$LightlyActiveMinutes)/mean(dailyInt$dayCountClickInt)), name = "Clicks"))
  lineIntLPA

Sedentary minutes

# Regression model - clicks
  clicksIntSB<-lm(clickIntSB~nDate+I(nDate^2),data=dailyInt)
  dailyInt$predClicksSB<-predict(clicksIntSB,dailyInt)

# Regression model - Sedentary
  intSB<-lm(SedentaryMinutes~poly(nDate,degree=2,raw=TRUE),data=dailyInt)
  dailyInt$predSB<-predict(intSB,dailyInt)

# Figure
  lineIntSB<-ggplot(dailyInt,aes(x=date,y=SedentaryMinutes))+
               #geom_jitter(color="#619CFF",alpha=0.15) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickSB),color="black",alpha=0.15) +
               #geom_smooth(data=dailyInt,aes(x=date,y=clickIntSB),color="black") +
               geom_smooth(data=dailyInt,method="lm",formula="y~x",aes(x=date,y=clickIntSB),color="black") +
               #geom_line(data=dailyInt,aes(x=date,y=predClicksSB),color="black") +
               #geom_smooth(color="#619CFF",fill="#619CFF") +
               geom_smooth(method="lm",formula="y~x",color="#FFC833",fill="#FFC833") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Sedentary minutes") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyInt$SedentaryMinutes)/mean(dailyInt$dayCountClickInt)), name = "Clicks"))
  lineIntSB

Sleep minutes

# Regression model - clicks
  clicksIntSleep<-lm(clickIntSleep~nDate+I(nDate^2),data=dailyInt)
  dailyInt$predClicksSleep<-predict(clicksIntSleep,dailyInt)

# Regression model - Sedentary
  intSleep<-lm(TotalTimeInBed~poly(nDate,degree=2,raw=TRUE),data=dailyInt)
  dailyInt$predSleep<-predict(intSleep,dailyInt)

# Figure
  lineIntSleep<-ggplot(dailyInt,aes(x=date,y=TotalTimeInBed))+
               #geom_jitter(color="#619CFF",alpha=0.15) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickSleep),color="black",alpha=0.15) +
               #geom_smooth(data=dailyInt,aes(x=date,y=clickIntSleep),color="black") +
               geom_smooth(data=dailyInt,method="lm",formula="y~x",aes(x=date,y=clickIntSleep),color="black") +
               #geom_line(data=dailyInt,aes(x=date,y=predClicksSleep),color="black") +
               #geom_smooth(color="#619CFF",fill="#619CFF") +
               geom_smooth(method="lm",formula="y~x",color="#FF8333",fill="#FF8333") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Time in bed") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyInt$TotalTimeInBed)/mean(dailyInt$dayCountClickInt)), name = "Clicks"))
  lineIntSleep

# Regression model - clicks

# Regression model - Sleep 

# Figure
  lineIntSleep<-ggplot(dailyInt,aes(x=date,y=TotalTimeInBed))+
               #geom_jitter(color="#FF8333",alpha=0.25) + 
               #geom_jitter(data=dailyForPlots,aes(x=date,y=clickSleep),color="black",alpha=0.25) +
               geom_smooth(data=dailyInt,aes(x=date,y=clickIntSleep),color="black") +
               geom_smooth(color="#FF8333",fill="#FF8333") +
               theme_classic()+
               labs(x="Date",
                    y="Minutes",
                    title="Time in bed") +
               theme(axis.title.x = element_text(face="bold"),
                     axis.title.y = element_text(face="bold"),
                     plot.title = element_text(face="bold",
                                               hjust=0.5))+
  scale_y_continuous(
    sec.axis = sec_axis(trans=~./(mean(dailyInt$TotalTimeInBed)/mean(dailyInt$dayCountClickInt)), 
                        name = "Clicks"))

Panel figure

 lineIntSave<-grid.arrange(lineIntMVPA,lineIntLPA,lineIntSB,lineIntSleep,ncol=2)

## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'
## `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

 lineIntSave

## TableGrob (2 x 2) "arrange": 4 grobs
##   z     cells    name           grob
## 1 1 (1-1,1-1) arrange gtable[layout]
## 2 2 (1-1,2-2) arrange gtable[layout]
## 3 3 (2-2,1-1) arrange gtable[layout]
## 4 4 (2-2,2-2) arrange gtable[layout]

Box plot - per period - one panel per behaviour

  boxplotMVPAperiod<-ggplot(dailyLong[dailyLong$behaviour=="MVPA",],
                            aes(x=date,y=minutes,fill=period),group=weekStart)+
                             geom_boxplot() +
                             scale_fill_manual(values=c("#FFC833","#81DA66","#619CFF"))+
                             theme_classic() +
                            labs(x="Date",
                                 y="Minutes MVPA",
                                 fill="Intervention Period",
                                 title="MVPA")+
                            theme(axis.text.x = element_text(angle=90,vjust = 0.5, hjust=1),
                                  axis.title.x = element_text(face="bold"),
                                  axis.title.y = element_text(face="bold"),
                                  plot.title = element_text(face="bold",hjust=0.5),
                                  legend.title = element_text(face="bold"),
                                  strip.text = element_text(face="bold")) 
  boxplotMVPAperiod

Box plot - per week - one panel per behaviour

Models

Exploratory: Linear regression

(a) only independent variable is intervention period

Compared to the pre-intervention period, daily MVPA was significantly higher in the intervention period and significantly lower in the post-intervention period

dailyForPlots <- dailyForPlots %>% 
  mutate(intervention=factor(intervention)) %>% 
  mutate(intervention=fct_relevel(intervention,c("pre","intervention","post"))) %>%
 arrange(intervention)

  linearInt<-lm(totalMVPA~intervention,data=dailyForPlots)
  tab_model(linearInt)

	totalMVPA
Predictors	Estimates	CI	p
(Intercept)	45.69	40.57 – 50.82	<0.001
intervention [intervention]	-1.11	-6.53 – 4.31	0.688
intervention [post]	-12.36	-18.97 – -5.75	<0.001
Observations	3252
R² / R² adjusted	0.008 / 0.007

(b) only independent variable is a binary indicator for Mon/Wed

MVPA was not significantly different on Monday and Wednesday compared to other days of the week (without accounting for any other variables, including intervention period and/or repeated measures within participants)

  linearMW<-lm(totalMVPA~mon_wed,data=dailyForPlots)
  summary(linearMW)

## 
## Call:
## lm(formula = totalMVPA ~ mon_wed, data = dailyForPlots)
## 
## Residuals:
##    Min     1Q Median     3Q    Max 
## -44.21 -32.72 -10.72  18.28 404.28 
## 
## Coefficients:
##             Estimate Std. Error t value Pr(>|t|)    
## (Intercept)  42.7192     0.9328  45.798   <2e-16 ***
## mon_wedyes    1.4957     1.7751   0.843    0.399    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 45.26 on 3250 degrees of freedom
## Multiple R-squared:  0.0002184,	Adjusted R-squared:  -8.92e-05 
## F-statistic:  0.71 on 1 and 3250 DF,  p-value: 0.3995

(c) only independent variable is whether the participant ever engaged (clicked or engaged) with app content

Participants who never enagaged with app content recorded an average of 13 minutes less MVPA per day (statistically significant) compared to those who engaged at least once (note: only a small number of participants (n=7) never engaged)

  linearEverEngage<-lm(totalMVPA~ever_engage,data=dailyForPlots)
  tab_model(linearEverEngage)

	totalMVPA
Predictors	Estimates	CI	p
(Intercept)	43.73	42.10 – 45.36	<0.001
ever engage [Never engaged or clicked]	-6.61	-12.04 – -1.18	0.017
Observations	3252
R² / R² adjusted	0.002 / 0.001

(d) only independent variable is dayCountClickAnys

Every one click was associated with a 2 minute increase in MVPA on that date (statistically signiciant)

  linearCountClick<-lm(totalMVPA~dayCountClickAny,data=dailyForPlots)
  tab_model(linearCountClick)

	totalMVPA
Predictors	Estimates	CI	p
(Intercept)	42.90	41.30 – 44.50	<0.001
dayCountClickAny	1.09	-0.66 – 2.85	0.223
Observations	3252
R² / R² adjusted	0.000 / 0.000

(e) only independent variable is countEngage

Every one engagement was associated with a <1.5 minute increase in MVPA on that date; however, the relationship was not statistically significant

  linearCountEngage<-lm(totalMVPA~dayCountEngageAny,data=dailyForPlots)
  tab_model(linearCountEngage)

	totalMVPA
Predictors	Estimates	CI	p
(Intercept)	43.01	41.42 – 44.60	<0.001
dayCountEngageAny	0.83	-1.53 – 3.19	0.490
Observations	3252
R² / R² adjusted	0.000 / -0.000

(f) intervention, countClicks, engagers, and univeristy

Every one click was associated with a 1.7 minute increase in MVPA on that date (statistically signiciant). There is also a significant intervention effect of 6.4 more daily minutes of MVPA on average over the intervention period compared to the pre intervention period. There is a interaction between the intervention and university. Both universities increased activity but compared to Queens UBC had less activity overall, but the intervention had a bigger effect. Last, people who never engaged had significantly less activity.

We can't do the interaction between ever_engage and intervention while including clicks in the model because those variables are collinear. Need to think about stratifying the model by engagers probably.

  linearClicksInt<-lm(totalMVPA ~ dayCountClickAny + intervention + ever_engage + intervention*University, data=dailyForPlots)
  tab_model(linearClicksInt)

	totalMVPA
Predictors	Estimates	CI	p
(Intercept)	53.87	46.79 – 60.96	<0.001
dayCountClickAny	0.44	-1.30 – 2.19	0.617
intervention [intervention]	-2.74	-10.21 – 4.72	0.471
intervention [post]	-16.31	-25.52 – -7.09	0.001
ever engage [Never engaged or clicked]	-8.60	-14.01 – -3.19	0.002
University [UBC]	-15.36	-25.53 – -5.19	0.003
intervention [intervention] * University [UBC]	3.62	-7.13 – 14.37	0.509
intervention [post] * University [UBC]	8.21	-4.91 – 21.32	0.220
Observations	3252
R² / R² adjusted	0.026 / 0.024

Files

fitbit-analysis.md

Latest commit

History

fitbit-analysis.md

File metadata and controls

Fitbit analysis

Exploratory Analysis

Engagement and clicks:

Total clicks and engagement per day (summed across all participants)

Total engagements per day, by date

Clicks and engagement per participant-day

Distribution of clicks and engagement per participant and day

Distribution of clicks and engagement per participant (across the entire study period)

Number of clicks and engagement per participant and day over time

Mean daily clicks and engagement over time

Clicks and engagement by weekday in the pre, intervention and post intervention period

Mean clicks and engagement, by weekday and period

Total time use

MVPA

Distribution/frequency of MVPA - with and without including 0s

Plot of minutes of MVPA per day over time -- overall and comparing engagers and non-engagers

Boxplot of MVPA per day and participant, by week

LPA

Distribution/frequency of LPA - with and without including 0s

Plot of minutes of LPA per day over time -- overall and comparing engagers and non-engagers

Boxplot of LPA per day and participant, by week

Sedentary Behaviour

Distribution/frequency of SB - with and without including 0s

Plot of minutes of SB per day over time -- overall and comparing engagers and non-engagers

Boxplot of SB per day and participant, by week

Sleep

Distribution/frequency of Sleep - with and without including 0s

Plot of minutes of Sleep per day over time -- overall and comparing engagers and non-engagers

Boxplot of SB per day and participant, by week

High-level: All movement behaviours over time

Plots

Line Graph - with engagement and clicks overlayed

Any clicks across all three periods

Intervention clicks during intervention period

Box plot - per period - one panel per behaviour

Box plot - per week - one panel per behaviour

Models

Exploratory: Linear regression

(a) only independent variable is intervention period

(b) only independent variable is a binary indicator for Mon/Wed

(c) only independent variable is whether the participant ever engaged (clicked or engaged) with app content

(d) only independent variable is dayCountClickAnys

(e) only independent variable is countEngage

(f) intervention, countClicks, engagers, and univeristy