README.md

Source code for Learning and Predicting from Dynamic Models for COVID-19 Patient Monitoring"

Repository for paper "Learning and Predicting from Dynamic Models for COVID-19 Patient Monitoring"

• 00_loadlibs.R: setup file directories and load required pacakges

• Code folder

  • generatePrediction

    1. multinom_results_alldata.R: run multinomial regression models for prospective and retrospective appraoch using all available data, and generate coefficient tables.

    2. jtgaussian_results_alldata.R: run multinomial regression models for prospective and retrospective appraoch, and generate Figure 1.

    3. model_perform_forward.R: run overall and cross-validated prospective linear mixed effects model; run cross-validated prospective multinomial model; generate prospective method predictions for each patient provided patients are hospitalized for 0, 2, 4, 8 days (baseline days).

    4. model_perform_backward.R: run overall and cross-validated retrospective linear mixed effects model; run cross-validated retrospective multinomial model; generate retrospective method predictions for each patient provided patients are hospitalized for 0, 2, 4, 8 days.

  • modelChecking

    1. mixed_model_checking.R: produce residuals versus fitted deciles plot and residuals Q-Q plot against standard Gaussian for prospective and retrospective linear mixed models.

    2. discrim_auc_script.R: read predictions from model_perform_forward.R and model_perform_backward.R and calculate AUC from predictions for each baseline day.

    3. calib_chisqstats_script.R: read prediction outputs and calculate chisquare statistics for calibration power.

    4. produce_calib_discrim_figures.R: read outputs from discrim_auc_script.R and calib_chisqstats_script.R and generate Figure 4.