Predicting Flu Hospitalizations Paper

Installation

Download R-3.6.3 from CRAN.
Download a .zip file of this repository from the GitHub Clone tab and unzip the archive into a location of your choice.
1. Another option would be to clone this repository to your computer. However, cloning takes a long time due to large file sizes and large version control history.
Navigate to the root of the unzipped repository on your computer and start an R session using R-3.6.3. You should see messages related to the renv package.
Once renv has installed itself, run the code below. Ideally, the user could run renv::restore() to install all dependencies, but the sl3 and randomForest versions required can’t be installed automatically.

renv::install("./renv/cellar/randomForest_4.6-14.tar.gz")
renv::install("devtools")
devtools::install_local("./renv/cellar/sl3-1.3.7.tar.gz")
renv::restore()

## from the root directory of the project 
devtools::install(quick = TRUE, build = FALSE, dependencies = FALSE)

Please post an issue if you have trouble installing.

Abstract

Accurate forecasts can inform response to outbreaks. Most efforts in influenza forecasting have focused on predicting influenza-like activity, but fewer on influenza-related hospitalizations. We conducted a simulation study to evaluate a super learner’s predictions of three seasonal measures of influenza hospitalizations in the United States: peak hospitalization rate, peak hospitalization week, and cumulative hospitalization rate. We trained an ensemble machine learning algorithm on 15,000 simulated hospitalization curves and generated weekly predictions. We compared the performance of the ensemble (weighted combination of predictions from multiple prediction algorithms), the best-performing individual prediction algorithm, and a naive prediction (median of a simulated outcome distribution). Ensemble predictions performed similarly to the naive predictions early in the season but consistently improved as the season progressed for all prediction targets. The best-performing prediction algorithm in each week typically had similar predictive accuracy compared to the ensemble, but the specific prediction algorithm selected varied by week. An ensemble super learner improved predictions of influenza-related hospitalizations, relative to a naive prediction. Future work should examine the super learner’s performance using additional empirical data on influenza-related predictors (e.g., influenza-like illness). The algorithm should also be tailored to produce prospective probabilistic forecasts of selected prediction targets.

Package directory (one level deep):

## .
## ├── DESCRIPTION
## ├── FluHospPrediction.Rproj
## ├── LICENSE.md
## ├── Makefile
## ├── NAMESPACE
## ├── R
## │   ├── calendar_mgmt.R
## │   ├── loss_absolute_error.R
## │   ├── simcrv_funs.R
## │   ├── sl_procedure.R
## │   └── summaries.R
## ├── README.Rmd
## ├── README.md
## ├── data
## │   ├── cleaned
## │   └── raw
## ├── inst
## │   ├── 01_data_cleaning_empdat.R
## │   ├── 02_simulate_hospcurves.R
## │   ├── 03_create_analysis_dataset.R
## │   ├── 04_run_superlearner.R
## │   ├── 05_run_sqerrloss_sensitivity.R
## │   ├── 06.1_tables-figures-setup.R
## │   ├── 06.2_tables-figures-simul.R
## │   ├── 06.3_tables-figures-main.R
## │   ├── 06.4_tables-figures-sens-1se.R
## │   ├── 06.5_tables-figures-sens-elastrf.R
## │   ├── 06.6_tables-figures-sens-sqerrloss.R
## │   ├── 06.7_tables-figures-sens-comb.R
## │   ├── 07_run_sl_prospective.R
## │   ├── 08_run_sl_observed.R
## │   ├── 09_tables-figures-sub-prosp-obs.R
## │   ├── 10_ensemble-cv.R
## │   ├── 11_ensemble-cv-sqerrloss.R
## │   ├── 12_ensemble-cv-1se.R
## │   ├── 13_ensemble_optimism.R
## │   ├── 14_run_sl_observed_allseasons.R
## │   ├── 15_extract-obstrain-preds.R
## │   ├── 16_tables-figures-obstrain.R
## │   ├── 17_figure-peakrate-week16.R
## │   ├── 99_test-script.R
## │   └── fmt_emp_season.R
## ├── man
## │   ├── calendar_mgmt.Rd
## │   ├── loss_absolute_error.Rd
## │   ├── predcurves.Rd
## │   ├── simcrv.Rd
## │   ├── simdist.Rd
## │   ├── summary_functions.Rd
## │   └── super_learner_proc.Rd
## ├── renv
## │   ├── activate.R
## │   ├── cellar
## │   ├── library
## │   ├── settings.dcf
## │   ├── settings.json
## │   └── staging
## └── renv.lock

Code files

All analytic code files are stored in /inst directory.
Functions to handle calendar date management, curve simulation, and risk table formatting located in /R

Data Sources

FluSurv-NET

Chaves SS, Lynfield R, Lindegren ML, Bresee J, Finelli L. The US Influenza Hospitalization Surveillance Network. Emerg Infect Dis. 2015 Sep;21(9):1543–50. Available from: http://dx.doi.org/10.3201/eid2109.141912

FluView: Influenza Hospitalization Surveillance Network, Centers for Disease Control and Prevention. WEBSITE. (Emerging Infections Program data)