WHO Cardiovascular Disease Risk Calculator

Implements the 2019 World Health Organization (WHO) cardiovascular disease (CVD) risk prediction models, as described in Kaptoge et al. (2019) . Provides two validated models for estimating 10-year risk of fatal and non-fatal cardiovascular events (myocardial infarction and stroke): a laboratory-based model using age, sex, systolic blood pressure, total cholesterol, smoking status, and diabetes history; and a non-laboratory-based model substituting body mass index (BMI) for cholesterol and diabetes, suitable for resource-limited settings. Risk estimates are recalibrated to 21 Global Burden of Disease regions (or to individual countries) using the authors' recalibration parameters, fitted to Global Burden of Disease 2017 incidence rates and Non-Communicable Disease Risk Factor Collaboration risk factor levels. Functions are fully vectorized for efficient batch calculations and support automatic country-to-region mapping via ISO 3166-1 alpha-3 country codes.

WHORiskCalculator: WHO Cardiovascular Disease Risk Calculator

An R package implementing the 2019 WHO cardiovascular disease (CVD) risk prediction models for 21 global regions.

Overview

This package calculates 10-year risk of cardiovascular disease (fatal and non-fatal myocardial infarction and stroke) using the WHO risk charts published in The Lancet Global Health (2019).

Reference: > Kaptoge S, Pennells L, De Bacquer D, et al. World Health Organization cardiovascular disease risk charts: revised models to estimate risk in 21 global regions. Lancet Glob Health. 2019;7(10):e1332-e1345. doi:10.1016/S2214-109X(19)30318-3{.uri}

Features

• Two models available:
  • Laboratory-based model (requires cholesterol measurement)
  • Non-laboratory-based model (uses BMI instead, suitable for resource-limited settings)
• Fully vectorized for efficient batch calculations
• Zero dependencies - uses only base R
• Automatic country-to-region mapping using ISO 3166-1 alpha-3 country codes
• 21 global regions with region-level recalibration by default (as in the published WHO risk charts), plus optional country-level recalibration
• Matches the authors' implementation: recalibration parameters taken from the authors' Stata program whocvdrisk

Installation

# From CRAN
install.packages("WHORiskCalculator")

# Development version from GitHub
# install.packages("remotes")
remotes::install_github("andredot/who_cvd_calculator")

Usage

Laboratory-Based Model

library(WHORiskCalculator)

# Single patient calculation using region
risk <- calculate_who_risk(
  age = 55,
  sex = "male",
  sbp = 140,
  cholesterol = 5.5,  # mmol/L
  smoking = TRUE,
  diabetes = FALSE,
  region = "western_europe"
)
print(paste0("10-year CVD risk: ", round(risk * 100, 1), "%"))

# Using country code instead of region
risk <- calculate_who_risk(
  age = 60,
  sex = "female",
  sbp = 130,
  cholesterol = 6.0,
  smoking = FALSE,
  diabetes = TRUE,
  country = "ITA"  # Italy -> western_europe
)

# Vectorized calculation for multiple patients
risks <- calculate_who_risk(
  age = c(45, 55, 65, 75),
  sex = c("male", "female", "male", "female"),
  sbp = c(120, 140, 160, 150),
  cholesterol = c(5.0, 6.0, 7.0, 5.5),
  smoking = c(FALSE, TRUE, FALSE, FALSE),
  diabetes = c(FALSE, FALSE, TRUE, TRUE),
  country = c("USA", "GBR", "IND", "JPN")
)

Non-Laboratory-Based Model

For settings where cholesterol measurement is not available:

risk <- calculate_who_risk_nonlab(
  age = 55,
  sex = "male",
  sbp = 140,
  bmi = 28,
  smoking = TRUE,
  region = "south_asia"
)

Recalibration: Region (Default) or Country

Both functions take a recalibration argument:

Region-level (default). Use this to get chart-equivalent risks. If you pass country, it is only used to find the region, so all countries in a region get the same result.

Country-level (optional). Use this when you want estimates recalibrated to a specific country's GBD 2017 incidence rates and NCD-RisC risk factor levels. Results can differ substantially from region-level ones:

# 45-year-old male smoker, UK, SBP 140 mmHg, total cholesterol 5.5 mmol/L
calculate_who_risk(45, "male", 140, 5.5, smoking = TRUE, diabetes = FALSE,
                   country = "GBR")                            # region:  ~7.0%
calculate_who_risk(45, "male", 140, 5.5, smoking = TRUE, diabetes = FALSE,
                   country = "GBR", recalibration = "country") # country: ~9.4%

Country-specific estimates are not available for South Sudan (SSD), the US Virgin Islands (VIR), American Samoa (ASM), Guam (GUM) and the Northern Mariana Islands (MNP), nor for Greenland (GRL), Bermuda (BMU) and Puerto Rico (PRI) before 2015. For these, the region-level parameters are used and a warning is issued.

Calendar year. year (default 2017, as in the charts) selects the year of the recalibration target population. Parameters exist for 1990, 1995, 2000, 2005, 2010, 2015, 2016 and 2017; other years up to 2015 use the 5-year period they fall in (e.g. 2003 uses 2000).

calculate_who_risk_nonlab(
  age = 55, sex = "female", sbp = 135, bmi = 26, smoking = FALSE,
  country = "KEN", recalibration = "country", year = 2010
)

Utility Functions

# Get list of valid regions
get_regions()

# Get country code mappings
get_country_codes()

# Look up specific countries
get_country_codes(c("USA", "GBR", "IND"))

# Map country to region
country_to_region(c("FRA", "DEU", "ITA"))  # All return "western_europe"

Model Details

Variables and Centering

Valid Ranges

• Age: 40-80 years (model derivation range)
• Systolic BP: Typically 70-250 mmHg
• Cholesterol: 2-12 mmol/L (divide mg/dL by 38.67)
• BMI: 12-60 kg/m²

21 Global Regions

Algorithm

Risk is calculated separately for (a) myocardial infarction / CHD death and (b) stroke, in three steps, following Kaptoge et al. (2019) and the authors' Stata implementation:

1. Uncalibrated risk from the derivation (ERFC) Cox model:

   P = 1 - S₀^exp(LP)
   

   where S₀ is the 10-year baseline survival of the derivation cohort (specific to sex, endpoint and model) and LP is the linear predictor, including main effects and age interactions.

2. Recalibration on the complementary log-log scale:

   P_cal = 1 - exp(-exp(cons + slope × ln(-ln(1 - P))))
   

   The intercept cons and slope slope are specific to region (or country, with recalibration = "country"), sex, endpoint, model and year. The authors fitted them to GBD 2017 incidence rates and NCD-RisC risk factor levels; this package takes them from their Stata program (net install whocvdrisk, from("https://raw.githubusercontent.com/skkceu/stata/main")). See data-raw/calibration_parameters.R for how they are extracted.

3. Combination, assuming independence of the two endpoints:

   P(CVD) = 1 - (1 - P_cal(MI)) × (1 - P_cal(Stroke))
   

Cholesterol Unit Conversion

The model uses mmol/L. To convert from mg/dL:

cholesterol_mmol <- cholesterol_mg_dl / 38.67

Limitations

1. Age range: Model was derived for ages 40-80. Extrapolation outside this range is less reliable.

2. Non-laboratory model: Does not include diabetes, so may underestimate risk in diabetic patients.

3. Calibration: Based on GBD 2017 estimates; actual risk may vary within a country or population. Because the models are recalibrated to contemporary populations, they may not be well suited to historically collected data (use year with care).

4. Risk factors not included: Family history, ethnicity, HDL cholesterol, triglycerides, etc.

License

MIT License. See LICENSE file for details.

The underlying risk models are from the WHO CVD Risk Chart Working Group. The recalibration parameters are from the whocvdrisk Stata program by Stephen Kaptoge and Lisa Pennells (Cardiovascular Epidemiology Unit, University of Cambridge).