Data quality assessment of general practitioner collected data on smoking status
Machine Learning: Health · 31.12.2026 · 2 (2) · 025006
DOI: 10.1088/3049-477x/ae866d ↗Kurz gefasst
Fachartikel in der Zeitschrift Machine Learning: Health, online erschienen am 16.07.2026. Verfasst von Jiayue Wang und 4 weiteren Autor:innen. Die Arbeit nennt 64 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research Technology Services at UNSW Sydney, Medical Research Future Fund, Australian National Health and Medical Research Council.
Abstract (Originalsprache)
Smoking status is a routinely documented behavioural risk factor within electronic health record (EHR) systems, in general practice (GP). However, variability in its data quality may constrain its fitness for providing care and secondary use. Robust and generalizable machine learning (ML) models in healthcare depend on high-quality EHR data, and incomplete or inconsistently recorded exposure variables such as smoking status, may introduce bias, degrade predictive performances, and limit external validity. This study aims to systematically evaluate the data quality of GP-recorded smoking-related information and examine variations in documentation across smoking-related and smoking-unrelated conditions. In this cross-sectional study, we investigated a GP dataset of 133,123 patients from South-Western Sydney using an evidence-based structured data quality framework. Two-proportion Z-tests and chi-square tests were used to measure variations in the collection of data on smoking status amongst different conditions. Although 92% of the patient records conformed to predefined assessment values of data quality checks, smoking status was recorded for only 14% of patients overall. After age adjustment, smoking status was not statistically associated with smoking-related conditions. Smoking status documentation completeness varied substantially across clinical subgroups, with disproportionately higher recording in pregnancy-related encounters (65%) compared with other diagnostic categories (23% among patients with myocardial infarction and 20% among those with smoking-unrelated diagnoses), indicating heterogeneous data quality within GP. The low rates of collection of data on smoking status in both smoking-related and smoking-unrelated conditions highlights the need for potential improvements in data quality. These findings highlight key data quality limitations that may impact the development of robust and generalizable ML models using GP EHR data.
Förderung
- Research Technology Services at UNSW Sydney
- Medical Research Future Fund
- Australian National Health and Medical Research Council
Zitieren
APA
Wang, J., Rakhshanda, S., Lukitasari, M., Sitas, F., & Jonnagaddala, J. (2026). Data quality assessment of general practitioner collected data on smoking status. Machine Learning: Health, 2(2), 025006. https://doi.org/10.1088/3049-477x/ae866d
Vancouver
Wang J, Rakhshanda S, Lukitasari M, Sitas F, Jonnagaddala J. Data quality assessment of general practitioner collected data on smoking status. Machine Learning: Health. 2026;2(2):025006. doi:10.1088/3049-477x/ae866d
BibTeX
@article{Wang2026,
author = {Wang, Jiayue and Rakhshanda, Shagoofa and Lukitasari, Mifetika and Sitas, Freddy and Jonnagaddala, Jitendra},
title = {Data quality assessment of general practitioner collected data on smoking status},
journal = {Machine Learning: Health},
year = {2026},
volume = {2},
number = {2},
pages = {025006},
doi = {10.1088/3049-477x/ae866d}
}
EUVIMED