ASBecker/exam_forecast

Companion analysis to a paper published in the Journal of Digital Imaging.

Paper Repository: Automatic Forecasting of Radiology Examination Volume Trends for Optimal Resource Planning and Allocation

by Anton S. Becker · Joseph P. Erinjeri · Joshua Chaim · Nicholas Kastango · Pierre Elnajjar · Hedvig Hricak · H. Alberto Vargas

This is the companion analysis to a paper published in the Journal of Digital Imaging.

Preparation

To run the analysis in ExamForecast.Rmd, the following packages need to be installed:

install.packages(
  c(
    "here",
    "kableExtra",
    "knitr",
    "magrittr",
    "prophet",
    "readr",
    "rmarkdown",
    "stringr",
    "tidyverse",
    "timeDate"
    # Recommended:
    "rticles"
    "skimr"
  )
)

Data Format

The synthetic toy data is provided in the "Data" folder (gzipped csv). It can be read natively by {{readr::read_csv}} or alternatively unzipped by R's native read.csv function. To run a prophet forecast with our own data, replace the csv files with your own data with at least two columns: Date and number of examinations. In order to recycle the code from this repository a column modality_code should be added containing either "CT" or "MRI".

library(dplyr)
here::here("Data", "per_diem_msk.csv.gz") %>%  
  readr::read_csv(show_col_types = FALSE) %>% 
  skimr::skim()

Simple forecast example

For a more comprehensive documentation please refer to the accompanying ExamForecast.Rmd and the official Prophet documentation. Below is a minimal example of a forecast for the next month:

library(dplyr)
library(prophet)

exams <- here::here("Data", "per_diem_msk.csv.gz") %>% 
  readr::read_csv() %>%
  filter(modality_code == "CT") %>% 
  transmute(
    ds = as_date(exam_date),
    y = n_exams,
  )

m <- prophet(exams)

pred <- make_future_dataframe(m, 31, "days", include_history = TRUE)

forecast_exams <- predict(m, pred)

License

All source code is made available under a MIT or file-specific license. You can freely use and modify the code, without warranty, so long as you provide attribution to the authors/cite the article. See LICENSE.md for the full license text.