Similarity as a risk factor in drug-name confusion errors: The look-alike (orthographic) and sound-alike (phonetic) model

BACKGROUND. One of every four medication errors reported in the United States is a nameconfusion error. The rate of name-confusion errors might be reduced if new and confusing names were not allowed on the market and if safeguards could be put in place to avoid confusion between existing names. OBJECTIVES. To evaluate several prognostic tests of drug-name confusion, alone and in combination, with respect to their sensitivity, specificity, and overall accuracy. RESEARCH DESIGN. Case-control study. Twenty-two different computerized measures of orthographic similarity, orthographic distance, and phonetic similarity were used to compute similarity/distance scores for n = 1,127 cases (ie, pairs of names that appeared in published error reports or national error databases) and n = 1,127 controls. MAIN OUTCOME MEASURES. Mean similarity/ distance scores were compared across cases and controls. The performance of each measure at distinguishing between cases and controis was evaluated by tenfold crossvalidation. Dose-response relationships were examined. Univariate and multivariate logistic regression models were formed and evaluated by 10 fold crossvalidation. RESULTS. Cases had significantly higher similarity scores than controls. Every measure of similarity proved to be a significant risk factor for error. There was a significant increasing trend in the odds-ratio as a function of similarity. A three-predictor logistic regression model had crossvalidated sensitivity of 93.7%, specificity of 95.9% and accuracy of 94.8%. CONCLUSIONS. A sensitive and specific test of drug-name confusion potential can be formed using objective measures of orthographic similarity, orthographic distance, and phonetic distance.