New PAP.adh data, renamed old one to PAP.adh1y (#130)

R/data.R

@@ -50,10 +50,44 @@ NULL
 
 
 #' @name PAP.adh
-#' @title Biweekly Mean Therapy Adherence of OSA Patients over 1 Year
+#' @title Weekly Mean PAP Therapy Usage of OSA Patients in the First 3 Months
 #' @description
-#' A simulated longitudinal dataset comprising 500 patients with obstructive sleep apnea (OSA) during their
-#' first year on CPAP therapy.
+#' A simulated longitudinal dataset comprising 301 patients with obstructive sleep apnea (OSA) during their first 91 days (13 weeks) of PAP therapy.
+#' The longitudinal patterns were inspired by the adherence patterns reported by Yi et al. (2022), interpolated to weekly hours of usage.
+#' @format A `data.frame` comprising longitudinal data of 500 patients, each having 26 observations over a period of 1 year.
+#' Each row represents a patient observation interval (two weeks), with columns:
+#' \describe{
+#'   \item{Patient}{`integer`: The patient identifier, where each level represents a simulated patient.}
+#'   \item{Week}{`integer`: The week number, starting from 1.}
+#'   \item{UsageHours}{`numeric`: The mean hours of usage in the respective week.
+#'   Greater than or equal to zero, and typically around 4-6 hours.}
+#'   \item{Group}{`factor`: The reference group (i.e., adherence pattern) from which this patient was generated.}
+#' }
+#'
+#' \insertRef{yi2022identifying}{latrend}
+#' @seealso [latrend-data] [PAP.adh1y]
+#' @examples
+#' data(PAP.adh)
+#'
+#' if (require("ggplot2")) {
+#'   plotTrajectories(PAP.adh, id = "Patient", time = "Week", response = "UsageHours")
+#'
+#'   # plot according to cluster ground truth
+#'   plotTrajectories(
+#'     PAP.adh,
+#'     id = "Patient",
+#'     time = "Week",
+#'     response = "UsageHours",
+#'     cluster = "Group"
+#'   )
+#' }
+"PAP.adh"
+
+
+#' @name PAP.adh1y
+#' @title Biweekly Mean PAP Therapy Adherence of OSA Patients over 1 Year
+#' @description
+#' A simulated longitudinal dataset comprising 500 patients with obstructive sleep apnea (OSA) during their first year on CPAP therapy.
 #' The dataset contains the patient usage hours, averaged over 2-week periods.
 #'
 #' The daily usage data underlying the downsampled dataset was simulated based on 7 different adherence patterns.
@@ -77,25 +111,20 @@ NULL
 #' \insertRef{aloia2008time}{latrend}
 #' @seealso [latrend-data]
 #' @examples
-#' data(PAP.adh)
+#' data(PAP.adh1y)
 #'
 #' if (require("ggplot2")) {
-#'   plotTrajectories(PAP.adh, id = "Patient", time = "Biweek", response = "UsageHours")
+#'   plotTrajectories(PAP.adh1y, id = "Patient", time = "Biweek", response = "UsageHours")
 #'
 #'   # plot according to cluster ground truth
 #'   plotTrajectories(
-#'     PAP.adh,
+#'     PAP.adh1y,
 #'     id = "Patient",
 #'     time = "Biweek",
 #'     response = "UsageHours",
 #'     cluster = "Group"
 #'   )
 #' }
-"PAP.adh"
-
-#' @rdname PAP.adh
-#' @description The `PAP.adh1y` dataset is a subset of `PAP.adh`, comprising only patients who used therapy for at least 1 year.
-#' The subset does not contain the Non-users and Early drop-out groups.
 "PAP.adh1y"
 
 #' @export

data-raw/PAP.adh.R

@@ -0,0 +1,116 @@
+library(magrittr)
+library(assertthat)
+library(data.table)
+library(latrend)
+
+# Cluster 1: "adherers"
+# y = 385, 326, 281
+# Cluster 2: "improvers"
+# y = 256, 297, 365
+# Cluster 3: "non-adherers"
+# y = 175, 137, 107
+
+# convert daily usage-hour variance to weekly usage minutes SD
+matrix(rnorm(1e6 * 7L, sd = 1.5 * 60), ncol = 1e6L) %>% colMeans() %>% sd()
+
+# define group trajectories
+groupTrajMat = cbind(
+  Adherers = approx(x = c(1, 4, 13), y = c(385, 328, 281), xout = 1:13, method = 'linear')$y,
+  Improvers = approx(x = c(1, 4, 13), y = c(256, 293, 365), xout = 1:13, method = 'linear')$y,
+  `Non-adherers` = approx(x = c(1, 4, 13), y = c(176, 138, 108), xout = 1:13, method = 'linear')$y
+)
+
+cluster1Data = generateLongData(
+  sizes = 162L,
+  id = 'Patient',
+  cluster = ~ Week + Adherers,
+  random = ~ 1,
+  data = data.frame(Week = 1:13, Adherers = groupTrajMat[, 'Adherers']),
+  clusterCoefs = cbind(c(0, 0, 1)),
+  clusterNames = 'Adherer',
+  randomScales = cbind(62.7),
+  noiseScales = 23,
+  seed = 1L
+)
+
+plotTrajectories(cluster1Data, time = 'Week', id = 'Patient', cluster = 'Class', response = 'Value') +
+  expand_limits(y = 0) +
+  scale_x_continuous(breaks = seq(1, 13, by = 2))
+
+cluster2Data = generateLongData(
+  sizes = 56L,
+  id = 'Patient',
+  cluster = ~ Week + Adherers,
+  random = ~ 1,
+  data = data.frame(Week = 1:13, Adherers = groupTrajMat[, 'Improvers']),
+  clusterCoefs = cbind(c(0, 0, 1)),
+  clusterNames = 'Improver',
+  randomScales = cbind(68.9),
+  noiseScales = 43,
+  seed = 2L
+) %>%
+  .[, Patient := Patient + max(cluster1Data$Patient)]
+
+cluster3Data = generateLongData(
+  sizes = 83L,
+  id = 'Patient',
+  cluster = ~ Week + Adherers,
+  random = ~ 1,
+  data = data.frame(Week = 1:13, Adherers = groupTrajMat[, 'Non-adherers']),
+  clusterCoefs = cbind(c(0, 0, 1)),
+  clusterNames = 'Non-adherer',
+  randomScales = cbind(53.7),
+  noiseScales = 34,
+  seed = 3L
+) %>%
+  .[, Patient := Patient + max(cluster2Data$Patient)]
+
+dataset = rbind(cluster1Data, cluster2Data, cluster3Data) %>%
+  .[, UsageHours := fifelse(Value > 0, Value / 60, 0)] %>%
+  setnames('Class', 'Group')
+
+plotTrajectories(dataset, time = 'Week', id = 'Patient', cluster = 'Group', response = 'UsageHours') +
+  expand_limits(y = 0) +
+  scale_x_continuous(breaks = seq(1, 13, by = 2))
+
+PAP.adh = subset(dataset, select = c('Patient', 'Week', 'UsageHours', 'Group')) %>%
+  as.data.frame()
+head(PAP.adh)
+stopifnot(uniqueN(PAP.adh$Patient) == 301)
+
+usethis::use_data(PAP.adh, overwrite = TRUE)
+
+
+# parallel
+nCores <- parallel::detectCores(logical = FALSE)
+if (.Platform$OS.type == "windows") {
+  cl <- parallel::makeCluster(nCores)
+  doParallel::registerDoParallel(cl)
+} else {
+  doMC::registerDoMC(nCores)
+}
+
+# test gbtm
+options(latrend.id = "Patient", latrend.time = "Week")
+gbtmMethod <- lcMethodLcmmGBTM(
+  fixed = UsageHours ~ Week,
+  mixture = ~ Week,
+  maxiter = 20, #not recommended; used for speed
+  idiag = TRUE
+)
+gbtmMethods <- lcMethods(gbtmMethod, nClusters = 1:5)
+gbtmList <- latrendBatch(gbtmMethods, data = PAP.adh, parallel = TRUE)
+
+plotMetric(gbtmList, c("Dunn", "WMAE", "BIC", "estimationTime"))
+
+# test gmm
+gmmMethod <- lcMethodLcmmGMM(
+  fixed = UsageHours ~ Week,
+  mixture = ~ Week,
+  idiag = TRUE
+)
+gmmMethods <- lcMethods(gmmMethod, nClusters = 1:5)
+system.time({
+  gmmList <- latrendBatch(gmmMethods, data = PAP.adh, parallel = TRUE)
+})
+plotMetric(gmmList, c("Dunn", "WMAE", "BIC", "estimationTime"))

data-raw/osa-adherence-data.R → data-raw/PAP.adh1y.R

@@ -245,13 +245,5 @@ plotTrajectories(dataset, id = 'Patient', time = 'Biweek', response = 'UsageHour
 dataset[, Biweek := as.integer(Biweek)]
 dataset[, MaxDay := as.integer(MaxDay)]
 
-PAP.adh = as.data.frame(dataset)
-usethis::use_data(PAP.adh, overwrite = TRUE)
-
-# PAP.adh1y
-PAP.adh1y = as.data.frame(
-  dataset[not(Group %in% c('Early drop-outs', 'Non-users'))]
-)
-PAP.adh1y$Patient = factor(PAP.adh1y$Patient)
-PAP.adh1y$Group = factor(PAP.adh1y$Group)
+PAP.adh1y = as.data.frame(dataset)
 usethis::use_data(PAP.adh1y, overwrite = TRUE)

inst/REFERENCES.bib

@@ -434,4 +434,16 @@ @Article{denteuling2021comparison
   timestamp = {2021-03-25},
 }
 
+@Article{yi2022identifying,
+  author           = {Yi, Huijie and Dong, Xiaosong and Shang, Shaomei and Zhang, Chi and Xu, Liyue and Han, Fang},
+  journal          = {Frontiers in Neurology},
+  title            = {Identifying longitudinal patterns of {CPAP} treatment in {OSA} using growth mixture modeling: Disease characteristics and psychological determinants},
+  year             = {2022},
+  month            = nov,
+  pages            = {1063461},
+  volume           = {13},
+  doi              = {10.3389/fneur.2022.1063461},
+  modificationdate = {2023-03-10T11:31:38},
+}
+
 @Comment{jabref-meta: databaseType:bibtex;}