Leveraging Electronic Dental Record Data to Classify Patients Based on Their Smoking Intensity

Abstract

Background Smoking is an established risk factor for oral diseases and, therefore, dental clinicians routinely assess and record their patients' detailed smoking status. Researchers have successfully extracted smoking history from electronic health records (EHRs) using text mining methods. However, they could not retrieve patients' smoking intensity due to its limited availability in the EHR. The presence of detailed smoking information in the electronic dental record (EDR) often under a separate section allows retrieving this information with less preprocessing. Objective To determine patients' detailed smoking status based on smoking intensity from the EDR. Methods First, the authors created a reference standard of 3,296 unique patients’ smoking histories from the EDR that classified patients based on their smoking intensity. Next, they trained three machine learning classifiers (support vector machine, random forest, and naïve Bayes) using the training set (2,176) and evaluated performances on test set (1,120) using precision (P), recall (R), and F-measure (F). Finally, they applied the best classifier to classify smoking status from an additional 3,114 patients’ smoking histories. Results Support vector machine performed best to classify patients into smokers, nonsmokers, and unknowns (P, R, F: 98%); intermittent smoker (P: 95%, R: 98%, F: 96%); past smoker (P, R, F: 89%); light smoker (P, R, F: 87%); smokers with unknown intensity (P: 76%, R: 86%, F: 81%), and intermediate smoker (P: 90%, R: 88%, F: 89%). It performed moderately to differentiate heavy smokers (P: 90%, R: 44%, F: 60%). EDR could be a valuable source for obtaining patients’ detailed smoking information. Conclusion EDR data could serve as a valuable source for obtaining patients' detailed smoking information based on their smoking intensity that may not be readily available in the EHR.

Natural language processing and machine learning to enable automatic extraction and classification of patients’ smoking status from electronic medical records

“…It allows the algorithm to always converge without regard of the dimensionality (17). A recent study on dental health records from the US found that SVM performed best to classify patients according to smokers, non-smokers, and unknowns, with a PPV and sensitivity of 98% and F-score of 0.98 (8). In addition, that study included an assessment of the patient's tobacco consumption, but it was limited to three classifiers instead of four as in our study and there was no consideration of the cost matrix.…”

“…Most of the previous work addressing the methods for classifying patients' smoking status has been conducted as a result of the 'Smoking challenge' (7,21,22,(28)(29)(30)(31)(32), and by others who continued building on that work (5,6,33). A recent US study on dental health records developed a similar model based on machine learning (8). However, to our knowledge, there is no such prior work performed on Swedish data.…”

“…Most of the research specifically addressing the problems of classifying smoking status based on secondary data sources was conducted in conjunction with the 2006 'Smoking challenge' announced by 'Informatics for Integrating Biology and the Bedside' (i2b2), a Centre for Biomedical Computing funded by the National Institute of Health in USA, and by those who continued building on that work (5)(6)(7). In addition, a recent US study on dental health records developed a similar model as presented here, but focussing on tobacco consumption (8). However, despite the recent advancements in text mining and machine learning, making use of unstructured information from EMRs such as smoking status still presents a challenge for researchers.…”

Developing Automated Computer Algorithms to Track Periodontal Disease Change from Longitudinal Electronic Dental Records

“…Another study reported a deep-learning model that automated the staging and grading of periodontitis [ 26 ]. Finally, a few studies have reported methods to extract PD risk factors information, such as smoking, diabetes, and cardiovascular diseases from the EDR [ 27 , 28 , 29 , 30 ]. Yet, to the best of our knowledge, no study has utilized longitudinal EDR data to study PD change and its clinical course over time.…”

Developing and testing a prediction model for periodontal disease using machine learning and big electronic dental record data

“…The increased availability of longitudinal patient care data electronically through the electronic dental record (EDR) offers an opportunity to characterize the present patient population's demographics, disease profiles to develop prediction models with up-to-date information (Song et al, 2013 ; Patel et al, 2018 ; Thyvalikakath et al, 2020 ). Moreover, advanced machine learning (ML) methods provide us with an opportunity to develop sophisticated data-driven models for PD.…”