Selection bias in clinical research when subjects are excluded because of failure to estimate left ventricular mass by echocardiography
Whalley GA., Gamble GD., Doughty RN., MacMahon S., Sharpe N.
Quantitative echocardiographic imaging often is used in clinical research, but the method is not successful in all patients. The goals of this study were as follows: (1) to determine the frequency with which left ventricular mass (LVM) could be estimated by echocardiography, (2) to characterize persons in whom LVM could not be estimated from echocardiography, and (3) to determine whether failure to estimate LVM was independently related to any common cardiovascular risk factors, thereby suggesting a potential for selection bias. This research was part of a community-based case-control study of risk factors in hypertension. Five hundred hypertensive subjects and 506 normotensive age- and sex-matched control subjects were studied prospectively. The mean age was 61 years (range: 39 to 76 years), the mean weight was 74 kg (range: 37 to 127 kg), 46% of the subjects were men, and 56% had a history of smoking. Two- dimensionally guided M-mode echocardiography was attempted in all subjects for determination of LVM (Penn convention). LVM was estimated when the endocardial borders could be accurately identified from a true radial cavity diameter. LVM was estimated in 803 subjects (79.8%). Subjects in whom estimates of LVM could not be obtained were older and had the following characteristics: higher systolic blood pressure; larger chest, waist, and abdominal circumferences; higher waist/hip ratio; and larger body mass index (all P < .02). In multivariate analysis the only independent predictors of failure to estimate LVM by echocardiography were advanced age (P = .0001), larger chest size (P = .0001), and history of smoking (P = .004). The inability to estimate LVM in 20% of subjects has important implications to our understanding the strength of association between potential risk factors and LVM, an independent measure of cardiovascular risk. LVM is more likely to be missing in older subjects who have a large chest and a history of smoking. The effects of such a selection bias should be considered in any analysis, and a characterization of the excluded subjects should be developed.