Comprehensive Summary
Machine learning and computerized natural language processing (NLP) algorithms show potential to streamline patient identification, overcoming challenges associated with the manual review of Electronic Healthcare Records (EHRs). In this work, Agaronnik et al. use NLP to identify patients who have used wheelchairs for one or more years, manually curating the chronicity of such use. EHRs from 14,877 patients between 21 and 75 years old diagnosed with colorectal cancer from 2005 to 2007 were extracted from the Research Patient Data Repository in Massachusetts. A curated keyword library associated with wheelchair use was submitted to Clinical Regex NLP software, which analyzed patients’ EHRs (starting 5 years before their diagnosis) identifying clinical notes with information related to wheelchair use, which were manually reviewed. NLP screening revealed that 0.5% (1482) of clinical notes, representing 2.8% (420) patients, contained wheelchair-associated keywords. However, only 19.3 % (286) of the identified clinical notes (referring to 105 patients) detailed the reasons and duration. Manual revision of clinical notes brought to light issues regarding inclusion of context and inadequate documentation of reasons or duration of wheelchair use. Overall, this work reveals that the main limitation to NLP’s applicability is the quality and completeness of EHRs.
Outcomes and Implications
Impaired mobility is the leading cause of disability in the U.S.A., associated with higher comorbidity rates, and affects 12.9% of American adults. Patients with chronic mobility disability have higher comorbidity rates and confront healthcare disparities. Improving the quality of care for this population starts with patient identification. Agaronnik et al. explore the applicability of NLP to identify patients who have used wheelchairs for one or more years, as a direct indicator of mobility disability. Despite the NLP algorithm improving efficiency in the identification of clinically relevant notes, these still required manual revision to confirm if patients used wheelchairs, the reasons, and the duration of such use. Still, only 19.3% of such notes contained sufficient detail to determine chronic wheelchair use, highlighting the need for complete documentation, and the development of documentation and ontology standards.