Patients who consume more than two drinks a day prior to elective surgery are at increased risk of experiencing a myriad of surgical complications, readmissions, and prolonged hospital stays. Fortunately, short-term pre-operative abstinence from alcohol mitigates many surgical risks, and carefully timed interventions can prevent complications and alcohol withdrawal syndrome. However, implementation of pre- operative alcohol interventions requires accurate identification of patients with risky alcohol use at least four weeks prior to surgery. Pre-operative clinics frequently fail to screen for alcohol use or do so too close to the surgery date to allow time for intervention. Electronic health records (EHRs) offer an unprecedented amount of accessible clinical data that can be leveraged to identify risky alcohol use early in the surgical episode of care. Innovative methods are needed to identify data elements and create algorithms to capture risky alcohol use from structured and unstructured EHR data. Natural language processing (NLP) and other machine learning (ML)-based approaches are best suited to extract and analyze alcohol-related clinical narratives, and to synthesize heterogeneous alcohol-related data through computer-assisted methods. The proposed study will leverage EHR data to identify and characterize risky alcohol use among surgical patients to identify cohorts who could benefit from pre-operative alcohol intervention. The study aims are to: 1) develop an electronic, automated computable phenotype to classify risky alcohol use prior to surgery using NLP and ML; 2) validate the algorithm through prospective data collection; and 3) longitudinally evaluate the association between risky alcohol use phenotypes and adverse surgical outcomes including complications and hospital readmissions. Innovative applications of NLP and ML will support evaluation of unstructured EHR data (e.g. clinical notes) and will enable integration of heterogeneous alcohol use data to create the computable phenotype.
The aims will be achieved through collaboration of experts in key clinical domains and advanced methodologies. This study will create and validate the first alcohol-specific phenotype-based algorithm for surgical patients, which will support future clinical applications and research into alcohol-related surgical interventions and health outcomes. Study outcomes are expected to have immediate value for identifying cohorts for future implementation research and lead to a new clinical tool for surgical clinics.
Risky alcohol use is one of the most common risk factors for surgical complications among elective surgical patients. This study will develop an automated computable phenotype to classify risky alcohol use prior to surgery using electronic health record data, and longitudinally evaluate the association between risky alcohol use and adverse surgical outcomes. Accurate and early identification of patients is needed to implement critical pre-operative alcohol interventions.
