Risk prediction is central in disease prevention and clinical management. Accordingly, several major clinical guidelines incorporate risk predic- tion models based on traditional risk factors (e.g.,
blood pressure, cholesterol, and smoking status) for clinical
decision making. One commonly used tool, the AHA/ACC
Atherosclerotic Cardiovascular Disease (ASCVD) risk calculator, is recommended when choosing treatment for patients
The data used to derive these risk equations often come
from epidemiologic cohort studies and randomized control
trials, with meticulous protocols for when and how to take
clinical measurements, which lead to essentially complete
ascertainment of the information needed and with consistent
methods of measurement. In real-life clinical practice, however, since predictors for risk equations are only “routinely”
assessed, many patients are missing measurements for one
or more risk factors.
Figure 1 shows rates of measurement for risk factors
required to estimate ASCVD risk with the AHA/ACC calculator.
Using electronic health record data from 22 different AMGA
members, we identified 2. 6 million patients aged 40–79, with
a diagnosis of hypertension and > 1 ambulatory office visit
(07/01/2017–06/30/2018). Among this population of patients
with hypertension, 4% had no recorded value for systolic
blood pressure in the 24 months prior to their most recent
office visit, 15% had no information for smoking status, and
30% had no information for HDL or total cholesterol. Only 64%
of patients had information for all four risk factors; 37% of
patients were missing one or more risk factor and would not
have had the appropriate information to predict ASCVD risk
using the AHA/ACC calculator.
Across 22 healthcare organizations, the proportion of
patients with complete information ranged from 47% to 82%.
Furthermore, clinical or lab measurements are generally not
random, and can be compounded by the presence of comor-bid conditions or other factors, presenting methodological
challenges for imputing values from the unmeasured to the
Further research is needed, evaluating the validity of
these global tools in local settings and exploring strategies
to optimize or tailor tools to local health system data (e.g.,
recalibrating and refitting models), as well as strategies to
handle missing information for predictors (e.g., adding dummy
variables, single or multiple imputation, stratified models
based on available information).
it’s all in the data
Data source: This analysis was conducted using a subset of the Optum® Analytics database, comprised of longitudinal ambulatory electronic health record (EHR) data from 22 healthcare
organizations that pool their EHR data as part of a national learning collaborative. Optum extracts data from multiple sources, cleans, normalizes, and validates it, making it possible to
conduct accurate lateral analysis and comparisons.
Rates of Measurement for ASCVD Risk Factors,
by Percent of Patients
SBP Smoke HDL Total cholesterol
All four risk factors 30 30%
Available in EHR ( 24 months)
Not available or missing