ISARIC WHO CCP-UK study: 4C Mortality Score public repo

Public repo containing functions and code used in the development of the 4C Mortality Score

Purpose

The functions in this repo are for use with ISARIC WHO CCP-UK study.

Data should be prepared, cleaned, and carefully checked as per standard procedures.

Collaboration with consortium members is advised to ensure limitations and vagaries of data are understood.

We are grateful if any bugs or errors can be reported in issues. Many thanks!

Data sharing

We welcome applications for data and material access via our Independent Data And Material Access Committee (https://isaric4c.net).

Data preparation

Clone and apply

https://github.com/SurgicalInformatics/cocin_ccp.

01_data_prep.R

1. Packages.
2. Variable definitions.
3. Extreme values in continuous variables.
4. Training and test definitions.
5. Geographical split definitions

Functions

02_functions.R

1. ggplot templates / extraction functions.
2. Table defaults.
3. Prognostic scoring.

Multiple imputation by chained equations

03_mice.R

1. Missing data inspection, description and characterisation .
2. Basic mice() function and approach.

Generalised additive models

04_gam.R

1. GAM with complete data.
2. GAM with mice data.
3. purrr methods for metrics, combining by Rubin's rules.

glmnet regression

05_lasso.R

1. Apply variables changes across imputed datasets.
2. glmnet() using MICE datasets.
3. Extract glmnet coefficients, combine and scale.

4C mortality score

06_4c_mortality_score.R

1. 4C mortality score function.
2. Prognostic index discrimination using mice data.
3. Score distribution.
4. Calibration.

Risk tables

07_risk_tables.R

1. prognos function for easy cut-off table generation.
2. Risk tables generation at specified cut-offs.

Comparisons with pre-existing scores

08_score_comparisons.R

1. Generate risk scores.
2. ff_aucroc() function for applying AUCROC and counts across multiple scores.
3. Apply ff_auroc() to risk scores.

XGBoost

09_xgboost.R

1. Variable definition.
2. Derivation and validation matrices created.
3. XGBoost training.
4. Discrimination (AUROC).

Decision curve analysis

10_decision_curve_analysis.R

1. Apply comparison scores in derivation and validation data.
2. DCA: fit in derivation and predict in validation (recalibration of comparators).
3. DCA: fit and predict in validation data (recalibration of all).

Mortality table for each score

4C mortality score	Mortality (%)
1	0.3
2	0.8
3	2.3
4	4.8
5	7.5
6	7.8
7	11.7
8	14.4
9	19.2
10	22.9
11	26.9
12	32.9
13	40.1
14	44.6
15	51.6
16	59.1
17	66.1
18	75.8
19	77.4
20	82.9
21	87.5