The Data-Driven Definition of "Complication" in Medicine

The concept of "complication" is widely used in the medical domain to designate unfavorable events in the course of medical care. However, the medical community has not yet established a strict and generally accepted definition of "complication". This makes it much harder to systematically track and ensure the safety of medical care, whether in an individual clinic or across the entire healthcare system. This study aimed to define the concept of "complication" by identifying its generic concept and key distinguishing features using natural language processing.

We conducted linguistic and statistical analysis of the term "complication" using a large corpus of medical texts from 90,688 completed neurosurgical cases in the digital archive of the N.N. Burdenko National Medical Research Center for Neurosurgery, Ministry of Health of the Russian Federation, spanning 2000 to 2017. The corpus was tokenized and normalized to obtain a vocabulary of 40,121 lexemes. A total of 5853 lexemes were selected as the lexicon of adverse medical events (LAME), supposed to be found in the context of complications. Using n-gram vector representations trained on our corpus, we obtained vector representations of LAME words and selected 4416 words as the sub-LAME core based on their positive cosine similarity with the vector for "complication". From the nouns, adjectives, and verbs in the sub-LAME, we extracted features that generalize, characterize, and classify complications. "Pathology" was identified as the generic concept for complication. The distinguishing features of complications were determined to be their novelty and emergence during observation of a primary phenomenon.Thus, we propose the following definition of "complication" for medical care safety monitoring:A complication (in medicine) is an intercurrent pathology detected during observation of an underlying disease, physiological process, or the result of intervention.Our patented method presented in this paper enables the development of scientifically grounded definitions for unclear or poorly defined concepts.