Modeling of Clinical Judgment Using Lasso Regression and Reduction of Assessment Items

Abstract

In the medical field, accidents involving the fall of inpatients around beds have become an increasing problem. To prevent such accidents, nurses record the condition of inpatients using a check sheet referred to as the Fall Assessment Score Sheet (Fall AS). These sheets are utilized for each inpatient and provide useful information for daily clinical judgment. However, compiling these sheets adds to the excessive workload of nurses, as the Fall AS contains numerous items to be checked and assessments frequently necessitate revision. We sought to model the clinical judgment of nurses using Lasso regression, with the aim of reducing the number items of the Fall AS. Three nurses were assigned a questionnaire that included 200 patterns of Fall AS and seven patterns of patient posture, and were requested to fill in the fall risk (0% to 100%) for all 1400 pattern combinations. The clinical judgment model was constructed based Lasso regression, adopting each fall risk and Fall AS as objective and explanatory variables, respectively. The proposed model succeeded in reducing the number of Fall AS items from 50 to 25. Moreover, the mean error of estimated values for the clinical judgment increased by only 3.44% compared with the general least-squares method.