BACKGROUND: Targeted screening for cardiovascular disease (CVD) can be carried out using existing data from patient medical records. However, electronic medical records in UK general practice contain missing risk factor data for which values must be estimated to produce risk scores. OBJECTIVE: To compare two methods of substituting missing risk factor data; multiple imputation and the use of default National Health Survey values. METHODS: We took patient-level data from patients in 70 general practices in Ealing, North West London. We substituted missing risk factor data using the two methods, applied two risk scores (QRISK2 and JBS2) to the data and assessed differences between methods. RESULTS: Using multiple imputation, mean CVD risk scores were similar to those using default national survey values, a simple method of imputation. There were fewer patients designated as high risk (>20%) using multiple imputation, although differences were again small (10.3% compared with 11.7%; 3.0% compared with 3.4% in women). Agreement in high-risk classification between methods was high (Kappa = 0.91 in men; 0.90 in women). CONCLUSIONS: A simple method of substituting missing risk factor data can produce reliable estimates of CVD risk scores. Targeted screening for high CVD risk, using pre-existing electronic medical record data, does not require multiple imputation methods in risk estimation.
BACKGROUND: Targeted screening for cardiovascular disease (CVD) can be carried out using existing data from patient medical records. However, electronic medical records in UK general practice contain missing risk factor data for which values must be estimated to produce risk scores. OBJECTIVE: To compare two methods of substituting missing risk factor data; multiple imputation and the use of default National Health Survey values. METHODS: We took patient-level data from patients in 70 general practices in Ealing, North West London. We substituted missing risk factor data using the two methods, applied two risk scores (QRISK2 and JBS2) to the data and assessed differences between methods. RESULTS: Using multiple imputation, mean CVD risk scores were similar to those using default national survey values, a simple method of imputation. There were fewer patients designated as high risk (>20%) using multiple imputation, although differences were again small (10.3% compared with 11.7%; 3.0% compared with 3.4% in women). Agreement in high-risk classification between methods was high (Kappa = 0.91 in men; 0.90 in women). CONCLUSIONS: A simple method of substituting missing risk factor data can produce reliable estimates of CVD risk scores. Targeted screening for high CVD risk, using pre-existing electronic medical record data, does not require multiple imputation methods in risk estimation.
Authors: Douglas G Manuel; Meltem Tuna; Richard Perez; Peter Tanuseputro; Deirdre Hennessy; Carol Bennett; Laura Rosella; Claudia Sanmartin; Carl van Walraven; Jack V Tu Journal: PLoS One Date: 2015-12-04 Impact factor: 3.240