download.file("https://github.com/g-walley/cegpy/raw/refs/heads/main/data/Falls_Data.xlsx", "Falls_Data.xlsx")
data <- readxl::read_xlsx("Falls_Data.xlsx")
data$Treatment[is.na(data$Treatment)] <- "Not Referred & Not Treated"
res1 <- reshape2::colsplit(data$HousingAssessment, " ", names = c("Housing", "Assessment"))
res2 <- reshape2::colsplit(data$Treatment, " & ", names = c("Referred", "Treatment"))
falls_long <- cbind(res1, data[, "Risk"], res2, data[, "Fall"])
file.remove("Falls_Data.xlsx")
usethis::use_data(falls_long, overwrite = TRUE)Staged Trees and CEG discovery with Falls Data
Loading and Cleaning
To download the falls data from github:
Intro to the Falls Data
Housing: Community or communal housing.
The study it was adapted from categorized housing as older people living at home, older people living in residential/nursing homes, older people in hospital, and older people in A&E (accident and emergency departments) (Eldridge). In the falls data, the categorization of communal living was categorized as older people living in nursing homes, care homes, and those in hospital environments (either the hospital or A&E). Older people living on their own were categorized as living in a community (Shenvi). Living situations can greatly impact the likelihood of a fall from conditions as seemingly minor as the length of staircases or the number of stories in the house. According to a geospatial study done in Olmsted County by the Mayo Clinic, it was determined that certain fall “hotspots” existed especially in people that often live in older homes, homes with 2 stories and fewer stairs, as well as being of lower SES (Takahashi et. al).
Assessment Risk: Whether the patient’s risk was assessed.
The assessment risk was determined using a Falls Risk Assessment Tool (FRAT) to see if the individual’s fall risk was assessed (Shenvi). The FRAT tool was administered by a wide variety of people, mainly depending on the housing status. If there was an older person living at home, the fall risk assessment (FRA) was conducted by social services or the primary health care team. If the older person was living in a residential/nursing home, the FRA was conducted by residential home staff or support services for residential homes. If the older person was living in a hospital, the FRA was conducted by a hospital consultant and physiotherapist. If the older people were in A&E, then the FRA was conducted by nursing staff and OT (Eldridge). It is important to note that this variable is significant in the sense that some people administering the surveys could be overly optimistic regarding the outcomes or have underlying cognitive biases. For example, in a study conducted in Japan, it was found that Nurse’s fall risk assessments were influenced by whether patients called for assistance or had other medical interventions such as have a tube or drain in place (Takase et al.). An additional point of variation to consider is the path to referral with the amount of providers the patient needs to see before finally reaching the falls clinic. In practice, if patients must go through many referrals, they are often “Lost to Care”.
Referred: Referral to various services.
This variable has to do with the referral status of a patient post receiving a FRA. In general, the patient could either first be referred to a general practitioner and/or primary health care team and then referred to a falls clinic or be directly referred to a falls clinic (Eldridge). A falls clinic serves to have the needs of a patient that has had unexplained falls be examined by a multidisciplinary team including a consultant, nurse, OT, and physiotherapist (Eldridge). A patient could also be referred to an OT and physiotherapy services for balance and gait problems after first being referred to a general practitioner and/or primary health care team (Eldridge).
Treatment: Received treatment to prevent or due to fall.
This variable assesses whether the patient received treatment as prescribed by the clinician for their increased fall risk (Shenvi). This would be received after the patient had a referral to a falls clinic or had seen their primary care physician, so it is reasonable to assume that every patient who received a referral also received some form of treatment. The objective of receiving treatment is to prevent future falls or help mitigate underlying causes that could lead to a fall.
Fall: Whether the patient fell or not.
The last variable simply assesses if the patient had a fall or not.
head(falls_long) Housing Assessment Risk Referred Treatment Fall
1 Community Not Assessed Low Risk Not Referred Not Treated Fall
2 Community Not Assessed High Risk Not Referred Not Treated Fall
3 Community Not Assessed Low Risk Not Referred Not Treated Don't Fall
4 Community Not Assessed Low Risk Not Referred Not Treated Don't Fall
5 Community Not Assessed Low Risk Not Referred Not Treated Fall
6 Community Not Assessed Low Risk Not Referred Not Treated Don't FallCreate a Staged Tree
Install the staged trees package in order to model the variables.
install.packages("stagedtrees")library(stagedtrees)
tree <- falls_long |> full() |> stages_bhc() |> stndnaming(uniq = TRUE)
palette("Tableau")
par(mfrow = c(1,1))
par(cex = 0.3)
plot(tree, col = "stages", cex_label_edges = 2)
References
- Shenvi, A., Smith, J. Q., Walton, R., & Eldridge, S. (2018, July). Modelling with non-stratified chain event graphs. In International Conference on Bayesian Statistics in Action (pp. 155-163). Cham: Springer International Publishing.