In ZachLevine-11/rMHF: Software package for daily-level, individual level, and weekly level aggregation, statistical modelling, hypothesis testing, and analysis for the medical records of a cardiac rehabilitation program.

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Individual level variables of interest

1. Met minutes
2. Age

Group level variables of interest.

1. Volume of visits
2. Attendance
3. %Females

We'll be using the automatic ARIMA model fitting machinery from the forecast package to simplify this a bit.

library(lubridate)
library(forecast)
library(rstatix)

First let's generate the three data frames of interest; mets, prescheduled appointments (volume), and noshows. Split each into only appointments in 2019 and only appointments in 2020.

##Change the year to select the year
splitbyyear <- function(df, year = "2019"){
  df[lubridate::year(df$Date) == year,]
}
mets2019 <- splitbyyear(RMHF::read_individual_data(fn = "mets.csv"), "2019")
mets2020 <- splitbyyear(RMHF::read_individual_data(fn = "mets.csv"), "2020")
##Order by date.
mets2019 <- mets2019[order(mets2019$Date),]
mets2020 <- mets2020[order(mets2020$Date),]

volume2019 <- splitbyyear(RMHF::read_individual_data(fn = "volume.csv"), "2019")
volume2020 <- splitbyyear(RMHF::read_individual_data(fn = "volume.csv"), "2020")
##Order by date.
volume2019 <- volume2019[order(volume2019$Date),]
volume2020 <- volume2020[order(volume2020$Date),]

Now, let's split off into seperate analyses for each variable. For each, we ask two questions: 1. Does 2019 differ from 2019? 1. If that difference exists, is it before and after the implementation of COVID-19 social distancing in Ontario?

Individual level data - Met minutes, sex and age.

First, aggregate data on the daily level.

##Should do the heavy lifting for us.
aggregate_date <- function(date, df, colselect = "Mets"){
  dfdate <- df[df$Date == date, colselect]
  return(mean(dfdate))
}
##Vectorized, gives data at the daily level.
meanmets2020 <- sapply(unique(mets2020$Date), aggregate_date, df = mets2020)
meanmets2019 <- sapply(unique(mets2019$Date), aggregate_date, df = mets2019)
meanage2020 <- sapply(unique(volume2020$Date), aggregate_date, df = volume2020, colselect = "Age")
meanage2019 <- sapply(unique(volume2019$Date), aggregate_date, df = volume2019, colselect = "Age")

We'd expect some autoregressive (ARIMA) structure in the data, so let's check for that.

library(forecast)
auto.arima(meanmets2019)
auto.arima(meanmets2020)
auto.arima(meanage2019)
auto.arima(meanage2020)

ARIMA models mostly come up up as white noise processes. So we can't predict anything. Defaulting to anovas and t - tests.

Mets

##Normal stuff, t - test between data in different years.
library(rstatix)
t.test(meanmets2020, meanmets2019[1:length(meanmets2020)], paired = TRUE)
covidIndex <- as.numeric(min(which(unique(mets2020)$Date == lubridate::ymd("2020/03/16"))))
df2020before <- mets2020[1:covidIndex]
df2020after <- mets2020[covidIndex + 1:length(df2020)]

Age

##Normal stuff, t - test between data in different years.
t.test(meanage2020, meanage2019[1:length(meanage2020)], paired = TRUE)
covidIndex <- as.numeric(min(which(unique(mets2020)$Date == lubridate::ymd("2020/03/16"))))
df2020before <- meanage2020[1:covidIndex]
df2020after <- meanage2020[covidIndex + 1:length(df2020)]

So there were no individual level differences between years. Or due to COVID. On to group - level stuff.

Group level variables of interest.

1. Volume of visits
2. Attendance
3. %Females

Volume of visits.

df <- RMHF::read_group_data()
df2019 <- as.numeric(df[df$Year == "2019",]$"Prescheduled appointments")
df2020 <- as.numeric(df[df$Year == "2020",]$"Prescheduled appointments")

##Between 2019 and 2020.
t.test(df2019, df2020, paired = TRUE)
##Withint 2020 before and after COVID-19 closure.
t.test(df2020before, df2020after, paired = TRUE)

Attendance

df <- RMHF::read_group_data()
df2019 <- as.numeric(df[df$Year == "2019",]$"% of patients who were no-shows")
df2020 <- as.numeric(df[df$Year == "2020",]$"% of patients who were no-shows")

##Between 2019 and 2020.
t.test(df2019, df2020, paired = TRUE)
##Withint 2020 before and after COVID-19 closure.
t.test(df2020[1:5], df2020[6:10], paired = TRUE)

%Females

df <- RMHF::read_group_data()
df2019 <- as.numeric(df[df$Year == "2019",]$"% Females")
df2020 <- as.numeric(df[df$Year == "2020",]$"% Females")


##Between 2019 and 2020.
t.test(df2019, df2020, paired = TRUE)
##Withint 2020 before and after COVID-19 closure.
t.test(df2020[1:5], df2020[6:10], paired = TRUE)

Correlations between attendance and met-minutes.

mets2020 <- RMHF::read_group_data()[RMHF::read_group_data()$Year == "2020",]$"Met-minutes"
mets2019 <- RMHF::read_group_data()[RMHF::read_group_data()$Year == "2019",]$"Met-minutes"
att2020 <- RMHF::read_group_data()[RMHF::read_group_data()$Year == "2020",]$"% of patients who were no-shows"
att2019 <- RMHF::read_group_data()[RMHF::read_group_data()$Year == "2019",]$"% of patients who were no-shows"

#The ith element in meanmets is the mean of the mets collected on ith day in that year.
cor(as.numeric(mets2019), as.numeric(att2019))
cor(as.numeric(mets2020), as.numeric(att2020))

ZachLevine-11/rMHF documentation built on Oct. 26, 2020, 11:16 a.m.