added readme.md

DanielaGawehns · Jan 25, 2024 · 893c83d · 893c83d
1 parent 2af63ac
commit 893c83d
Show file tree

Hide file tree

Showing 40 changed files with 165 additions and 19 deletions.
diff --git a/.DS_Store b/.DS_Store
diff --git a/AnalysisPipelineSTART_HERE.R b/AnalysisPipelineSTART_HERE.R
@@ -103,7 +103,7 @@ Data24hrs21<- normalize24hrs30min(perPerson21)
 ### For 2021: Medlo, 24hrs, estimators (MAD and ENMO)
 MedloData <- NormalizedTo30min #
 Normalized_24hrs_Data <- Normalized24hrsTo30min21 #
-keydata<- Link19_21
+keydata<- Link19_21 [,1:2]
 
 ### create linkedData for different ENMO/MAD epochs to compare results
 
@@ -128,6 +128,7 @@ Frag=sapply(MAD_AllEpochs5sec,fragment),
 #time spend above 100 (relative)
 Rel= sapply(MAD_AllEpochs5sec, function (x) length(which(x >100))) )
 
+#create dataset for correlation matrix (this means some data will be lost bc only complete sets are kept!)
 LinkedData_21_5sec<- link21data (MedloData, Normalized_24hrs_Data, ENMO_MAD_Data, keydata)
 
 

diff --git a/BarchartPercentagesMedlo.pdf b/BarchartPercentagesMedlo.pdf
diff --git a/BoxplotAllCatsMedlo.pdf b/BoxplotAllCatsMedlo.pdf
diff --git a/BoxplotAllCatsSamsung.pdf b/BoxplotAllCatsSamsung.pdf
diff --git a/BoxplotAllCatsSamsungPercent.pdf b/BoxplotAllCatsSamsungPercent.pdf
diff --git a/BoxplotsAllCatsSamsung.pdf b/BoxplotsAllCatsSamsung.pdf
diff --git a/CorrelationMatrix_5sec.pdf b/CorrelationMatrix_5sec.pdf
diff --git a/CorrelationMatrix_5sec_hc.pdf b/CorrelationMatrix_5sec_hc.pdf
diff --git a/LinkData_Correlations.R b/LinkData_Correlations.R
@@ -22,22 +22,28 @@ link21data<- function (MedloData, Normalized_24hrs_Data, ENMO_MAD_Data, keydata)
   WatchData_21 <- merge(SensorData_21,Dat21_24hrs, by = "ResidentID" )
 
   #Sensors, 24hrs, Medlo in 21
-  MovementData_21 <- merge(WatchData_21, MedloData, by.x = "NIVEL.ID..Compas.2019.", by.y = "bwcode")
-  MovementData_21 <- MovementData_21[,-c(13,14,15,16,17)] #remove empty cols from keyfile and an empty column for "cycling" 
+  MovementData_21 <- merge(WatchData_21, MedloData, by.x = "NIVEL.ID..Compas.2019.", by.y = "bwcode") #this is where "cycling" disappears 
+  #(only full observations are kept - for correlations later)
+
+  #MovementData_21 <- MovementData_21[,-c(21,22,23,24)] #remove empty cols from keyfile - not needed anymore, cleaned keyfile when importing data
+
   colnames(MovementData_21) <- c( "NIVEL_ID", "LIACS_ID", 
-                                  "MeanMAD", "MeanEN", "MeanENMO",
+                                  "MADmean", "MeanEN", "MeanENMO",
                                   "MinMAD","MinEN", "MinENMO",
                                   "MaxMAD", "MaxEN", "MaxENMO",
+                                  "VarMAD", "VarEN", "VarENMO",
                                   "NoEpochs",
-                                  "Inactive", "Light", 
+                                  "MADsd", "MADmedian", "MADqant", "MADfrag","MADrel",
+                                  "Cycling","Inactive", "Light", 
                                   "NotOnWrist", "Sleeping", 
                                   "MLO1", "MLO2", "MLO3", "MLO4", "MLO5",
                                   "MLO6", "MLO7")
 
   #select a subset of entries for correlation matrix:
-  data<- MovementData_21[, c(3,5,7,
-                             13,14,15,16,
-                             17,18,19,20,21,22) ]
+  data<- MovementData_21[, c(3, #MADmean
+                             16,17,18,19,20, #sd,median, qant,frag,rel
+                             22,23,24,25, #inactive, light, sleeping, not on wrist
+                             26,27,29,30) ]#ml1,ml2,ml4, ml5
   data[is.na(data)] <- 0
 
   return(data)

diff --git a/MADSpreadHighVariance.pdf b/MADSpreadHighVariance.pdf
diff --git a/MADSpreadHighVariance0to420.pdf b/MADSpreadHighVariance0to420.pdf
diff --git a/MADSpreadHighestVariance.pdf b/MADSpreadHighestVariance.pdf
diff --git a/MADSpreadLowVariance0to420.pdf b/MADSpreadLowVariance0to420.pdf
diff --git a/MADSpreadLowestvariance.pdf b/MADSpreadLowestvariance.pdf
diff --git a/MADspreadHighVariance.png b/MADspreadHighVariance.png
diff --git a/MADspreadLowVariance.pdf b/MADspreadLowVariance.pdf
diff --git a/PerResidentMAD.png b/PerResidentMAD.png
diff --git a/PerResidentMAD5sec.png b/PerResidentMAD5sec.png
diff --git a/PerResidentMADResidentID1.png b/PerResidentMADResidentID1.png
diff --git a/PerResidentMADResidentID10.png b/PerResidentMADResidentID10.png
diff --git a/PerResidentMADResidentID11.png b/PerResidentMADResidentID11.png
diff --git a/PerResidentMADResidentID12.png b/PerResidentMADResidentID12.png
diff --git a/PerResidentMADResidentID13.png b/PerResidentMADResidentID13.png
diff --git a/PerResidentMADResidentID14.png b/PerResidentMADResidentID14.png
diff --git a/PerResidentMADResidentID15.png b/PerResidentMADResidentID15.png
diff --git a/PerResidentMADResidentID16.png b/PerResidentMADResidentID16.png
diff --git a/PerResidentMADResidentID2.png b/PerResidentMADResidentID2.png
diff --git a/PerResidentMADResidentID3.png b/PerResidentMADResidentID3.png
diff --git a/PerResidentMADResidentID4.png b/PerResidentMADResidentID4.png
diff --git a/PerResidentMADResidentID5.png b/PerResidentMADResidentID5.png
diff --git a/PerResidentMADResidentID6.png b/PerResidentMADResidentID6.png
diff --git a/PerResidentMADResidentID7.png b/PerResidentMADResidentID7.png
diff --git a/PerResidentMADResidentID8.png b/PerResidentMADResidentID8.png
diff --git a/PerResidentMADResidentID9.png b/PerResidentMADResidentID9.png
diff --git a/PercentageTimeMedloActive.pdf b/PercentageTimeMedloActive.pdf
diff --git a/PercentageTimeSamsungActive.pdf b/PercentageTimeSamsungActive.pdf
diff --git a/PlottingScripts.R b/PlottingScripts.R
@@ -1,4 +1,4 @@
-#Plotting Scripts
+Plotting Scripts
 
 ##############################
 #Barcharts - MEDLO and 24hrs data
@@ -11,7 +11,7 @@ datMedlo <- data.frame(
                                             "Standing", "Standing activity/walking around", "Walking Activity/Cycling", 
                                             "Sports/Whole Body Movement")),
 
-  TimeSumsMedlo = c(colSums(LinkedData_21_30min [,8:13], na.rm = TRUE),0)
+  TimeSumsMedlo = c(colSums(LinkedData_21_5sec [,8:13], na.rm = TRUE),0)
 )
 
 p <- ggplot(data=datMedlo, aes(x=MovementCategoryMedlo, y=TimeSumsMedlo, fill=MovementCategoryMedlo)) +
@@ -71,9 +71,10 @@ corrplot(cor(LinkedData_21_5sec),
                #lower = "number", 
                #upper = "circle",
                tl.col = "black",
-               title = "5secEpochs",
+               title = "",
                mar=c(0,0,2,0),
-               order = "hc", 
+               order = "hclust", 
+              hclust.method = "complete",
                addrect = 2)
 
 # Closing the graphical device
@@ -112,24 +113,115 @@ pairs(LinkedData_21_60sec[,c(4,5,6,7,8,9,10,11,12,13)], pch = 19)
 
 MAD_AllEpochs5sec<-readRDS("epochsPerResident5sec.rds")
 allData<-numeric(length=3)
+allMeans<-numeric()
 
 for (i in 1:length(MAD_AllEpochs5sec)){
 
-dat<-data.frame(MAD=MAD_AllEpochs5sec[[i]],
-                 time= seq (1: length( MAD_AllEpochs5sec[[i]])),
-                 resident=rep(i, length(MAD_AllEpochs5sec[[i]])))
+dat<-data.frame(MAD=MAD_AllEpochs5sec[[i]][,1],
+                 measurement= seq (1:  length( MAD_AllEpochs5sec[[i]][,1])),
+                 resident=rep(i, length(MAD_AllEpochs5sec[[i]][,1])))
 
 allData<-rbind(allData,dat)
+allMeans<-c(allMeans, mean(dat$MAD,na.rm=TRUE))
 }
+
+allData<- allData [-1,] #remove empty first line
+var(allMeans)
+min(allMeans)
+max(allMeans)
+#########
+
+
+
+
 library(ggplot2)
-allData<- allData [-1,]
-ggplot(allData, aes(time, MAD)) +
+###! This takes a long time to render!
+p<- ggplot(allData, aes(x=measurement, y=MAD)) +
+    geom_point()+
+    facet_wrap(~allData$resident, nrow=3)
+p
+ggsave(filename="PerResidentMAD.png")
+
+#####This is more doable,saves plots individually
+for (j in 1:length(unique(allData$resident))) {
+  filename <- paste0("PerResidentMADResidentID",j,".png")
+  sampleResident <- allData[which(allData$resident == j),]
+  q<-ggplot(sampleResident, aes(x=measurement, y =MAD)) + geom_point()
+  q
+  ggsave(filename)
+}
+
+lowvarianceResIDs<- c(6,2,15)
+highvarianceResIDs<- c(8,14,13)
+
+p<- ggplot(allData [which(allData$resident %in%lowvarianceResIDs),], 
+           aes(x=measurement, y=MAD)) +
   geom_point()+
-  facet_wrap(~allData$resident, nrow=3)
+  ylim(0,420)+
+  facet_wrap(~allData [which(allData$resident %in%lowvarianceResIDs),]$resident, 
+             nrow=1)
+p
+ggsave("MADSpreadLowVariance0to420.pdf")
 
-p <- ggplot(allData, aes(x=as.factor(allData$resident), y=MAD)) + 
+q<- ggplot(allData [which(allData$resident %in%highvarianceResIDs),], 
+           aes(x=measurement, y=MAD)) +
+  geom_point()+
+  ylim(0,420)+
+  facet_wrap(~allData [which(allData$resident %in%highvarianceResIDs),]$resident, 
+             nrow=1)
+q
+ggsave("MADSpreadHighVariance0to420.pdf")
+
+#### violin plots of MAD data per resident
+p <- ggplot(allData [which(allData$resident %in%lowvarianceResIDs),], 
+            aes(x=as.factor(allData [which(allData$resident %in%lowvarianceResIDs),]$resident), 
+                y=MAD)) + 
   geom_violin()
 p
 
 #############################
+#instead of barcharts of the summed data, make boxplots of the percentage of time
+#a resident spend in one movement category
+
+Medlo_2019$codebew
+
+percentageActiveMedlo21<- MedloData [1:13 ,2:8] * 100/30 #!!! Use on first 13 entries for 2021 entries
+percentageActiveMedlo21[is.na(percentageActiveMedlo21)] <- 0 #clear NAs
+
+percentageActiveSamsungCat<- t(Normalized_24hrs_Data) *100/30
+percentageActiveSamsungCat[is.na(percentageActiveSamsungCat)] <- 0 #clear NAs
+
+percentageActiveMedloLong21<- data.frame(
+  percentage=unlist(c(percentageActiveMedlo21)),
+  activity=c(rep("ML01", 13), rep("ML02", 13),
+             rep("ML03", 13),
+             rep("ML04", 13), rep("ML05", 13),
+             rep("ML06", 13), rep("ML07", 13))
+)
+
+percentageActiveSamsungLong<- data.frame(
+  percentage=unlist(c(percentageActiveSamsungCat)),
+  activity=c(rep("Cycling", 16), rep("Inactive", 16), 
+             rep("Light", 16), 
+             rep("NotOnWrist", 16), rep("Sleeping", 16) )
+)
+
+boxplot(percentageActiveMedloLong21$percentage ~ percentageActiveMedloLong21$activity,
+        col='steelblue',
+        main='Percentage of Time spent per activity',
+        xlab='Activity',
+        ylab='Percentage of Time') 
+
+# Opening the graphical device
+pdf("BoxplotAllCatsSamsungPercent.pdf")
+boxplot(percentageActiveSamsungLong$percentage ~ percentageActiveSamsungLong$activity,
+        col='steelblue',
+        main='Percentage of Time spent per activity',
+        xlab='Activity',
+        ylab='Percentage of Time') 
+
+# Closing the graphical device
+dev.off() 
+
+
 
diff --git a/epochsPerResident5sec.rds b/epochsPerResident5sec.rds
diff --git a/readme.md b/readme.md
@@ -0,0 +1,47 @@
+# Using Consumer Wearables to Measure Physical Activity of Nursing Home Residents with Dementia
+
+This is the repository with the code needed to rerun our analysis of the manuscript: "Using Consumer Wearables to Measure Physical Activity of Nursing Home Residents with Dementia", available as preprint here: https://osf.io/preprints/psyarxiv/mqg86
+
+
+## Data
+
+The data necessary to run this script and to recreate the plots is currently not available. 
+There are three types needed
+a) aggregated behavioral observations (Medlo Data)
+b) raw accelerometer data (Acc Data)
+c) aggregated 24hrs log data from the Samsung activity recognition application (24hrs Data)
+
+From the raw accelerometer data one can calculate the MAD derivatives.
+With the two aggregated datasets the correlation matrix can be recreated
+
+
+__Data Contact__
+
+You can contact Daniela Gawehns, the owner of the repro and first author of the manuscript regarding the data: <gawehnsd@liacs.leidenuniv.nl> or <d.gawehns@umcg.nl> 
+
+
+## Getting Started
+
+The .r file named AnalysisPipelineSTART_HERE.R is the easiest entry point to rerun the analysis. This is where the data gets imported, cleaned and normalized into the format used for the plots. 
+The file PlottingScripts.R contains all code used to plot the figures in the manuscript. 
+
+
+### Requirements
+
+The scrips have the following dependencies: ggplot2, lubridate, haven and eeptools
+(eeptools is only used in a script to extract demographic data from participant files, which will not be shared out of privacy reasons)
+
+## Publications
+
+Gawehns, D., Portegijs, S., van Beek, A. P. A., & van Leeuwen, M. (2024, January 26). Using Consumer Wearables to Measure Physical Activity of Nursing Home Residents with Dementia. Retrieved from osf.io/preprints/psyarxiv/mqg86
+
+
+van Dijk, R.M.K., Gawehns, D. and van Leeuwen, M., 2023. WEARDA: Recording Wearable Sensor Data for Human Activity Monitoring.  Journal of Open Research Software,  11(1), p.13.DOI: https://doi.org/10.5334/jors.454
+
+
+
+## License
+
+This project is licensed under CC-BY-SA-4.0. 
+
+