Background: There has been ample speculation about potential future applications of natural language processing and related technologies for clinical documentation, and yet the mechanics of clinical documentation have changed very little since the passage of the HITECH Act in 2009.

Purpose: While awaiting the integration of more advanced technologies into electronic health records (EHRs), we sought to leverage an existing “low-tech” feature – spell check – to improve clinical documentation integrity (CDI) in a way that also makes the process of documentation less burdensome.

Description: Our group created a custom, shareable spell check dictionary (Figure 1) within our EHR (Epic Systems; Verona, WI) that recognizes words which are commonly used in diagnostic terminology but are not congruent with Centers for Medicare and Medicaid (CMS) terminology. These words are each paired with a suggestion for CMS-congruent ways to document the same diagnosis.For example, “delirium” is a commonly used term, but it does not contain the diagnostic specificity needed to be recognized by CMS. With our dictionary, when a user types “delirium” into their note, a red wavy line will appear beneath “delirium,” indicating a “misspelled” word. When the user clicks on the word, they are given the option to replace that word with “[menu] encephalopathy,” providing a drop-down menu containing different CMS-recognized forms of encephalopathy such as “toxic-metabolic.” (Figure 2)Currently, the main mechanism to ensure CMS-congruent documentation consists of queries sent by clinical documentation specialists when they manually identify noncompliant documentation. This process is burdensome both to these specialists and to clinicians, and the time it consumes contributes to the administrative costs of healthcare. We hypothesize that our custom dictionary may be able to partially obviate the need for this process by intercepting and correcting noncompliant documentation as it is being written.We have implemented a 4-week pilot of the dictionary with a group of 13 hospitalists to assess feasibility and acceptability. After pilot testing, we plan to conduct a prospective interrupted time series study to investigate whether CDI queries decrease with full implementation of the custom dictionary across the entire division and to assess the custom dictionary’s subjective impact on documentation burden.

Conclusions: Our custom dictionary repurposes a ubiquitous technology in a novel way which may improve and reduce the burden of documentation. In behavioral economics terms, this dictionary functions as a “nudge” for users to improve their documentation. It also creates “choice architecture” by suggesting different options for alternative ways to document. This type of design is what makes spell check an effective, unobtrusive, and widely accepted tool. We believe that our custom dictionary will be similarly effective and acceptable to users for these reasons.Furthermore, this is a base feature of EHRs, and custom dictionaries can be easily exported and shared within or across institutions via a small ASCII text file. Thus, this intervention can be easily disseminated at no cost, and its other potential applications are vast. For example, our group has begun developing a similar dictionary for stigmatizing language (e.g., it suggests “declined” if users type “refused”), and the autocorrect feature can save time and make notes more readable by replacing acronyms or jargon with language that patients can understand.

IMAGE 1: Figure 1

IMAGE 2: Figure 2