Modeling the information-value decay of medical problems for problem list maintenance

The medical problem list is a key attribute of electronic health records, according to the Institute of Medicine, serving as a dynamic 'table-of-contents' to the patient's record. However, the problem list is infrequently maintained and often contains irrelevant and resolved problems. Other work has focused on adding missing information but not on removing resolved problems. Therefore we present: first, a model of the information-value decay of diagnosis information over time; second, a method to estimate this decay function using survival analysis applied to proxies of diagnosis lifetime in medical order entry data. We evaluate our method using data on 38,824 patients seen at a local hospital in a three-year period, comparing the derived survival functions to clinical expectation. We find the method useful in comparing decay and rank-ordering the relative durations of medical problems. Discriminating on the forty problems with highest and lowest area under the survival curve, our method was able to select chronic problems (as defined in HCUP's list of chronic conditions) with 100% sensitivity and 80% specificity. This straightforward statistical method applied to routinely-collected empirical data can help clean up the problem list by removing 'decayed' problems, which will lead to improvements both in patient care and medical research.