Intelligent data entry for physicians by machine learning of an anticipative task model

We report empirical work toward development of an adaptive interface for physician´s data entry of electronic medical records (EMRs) in general practice. The goal is to improve useability of EMR systems by having the computer anticipate physicians´ data entry actions. We investigate generation of short menus (hot lists) that offer likely selections to the user. A task model from which we derive hot lists is formed by machine learning from a database of 3085 records of past encounters. The hot lists anticipate a patient´s drug treatment (from among 332 generic names) using already entered problem codes. Based on simulated data entry using records held back from training, hot lists of length 12 contain just under 70% of drug selections. 86% hit rates are found with anticipation of drug categories. The results show promise for development of useful task models via machine learning for complex domains such as medicine