Derivation and validation of a decision rule for predicting seat belt utilization

Information from 3,108 health risk appraisals completed by Tennessee residents in 1986 was used to develop a decision rule for predicting seat belt utilization. The data set was randomly divided into derivation and validation sets. The dependent variable was self-reporte seat belt use (percentage). Using multiple linear regression, the following rule was derived: score = [age (years) x 0.24] + [mood-affecting drug use x 4.09] + [miles driven per year x 5.08] + [education level x 11.18] - [race x 18.31] - [cigarette use x 2.73] - [satisfaction with life x 3.50] - [body mass (kg/m²) x 0.83] - [urban/rural residence x 4.08]. Likelihood ratios for persons stating 0 to 25 percent seat belt use were compared with those for persons stating 76 to 100 percent use. The prevalence of 0 to 25 percent seat belt use was 31 percent in the derivation set and 33 percent in the validation set. At the lowest quintile of score (-1 or less), the likelihood ratios were 4.18 and 3.31 in the derivation and validation sets, respectively. At the highest quintile of score (26 or more) the likelihood ratios were 0.29 and 0.38, respectively. At score levels than 10 the decision rule had a sensitivity of 59 percent and 55 percent and a specificity of 80 percent and 81 percent in the derivation and validation sets, respecitively. This decision rule may be used by primary care physicians to identify persons likely not to use seat belts and target them for health promotion efforts.