Background

Machine learning algorithms are promising tools for smoking status classification in big patient data sets. Smoking is a risk factor for postoperative complications in major surgery. Whether this applies to all surgery is unknown. The aims of this retrospective cohort study were to develop a machine learning algorithm for clinical record-based smoking status classification and to determine whether smoking and former smoking predict complications in all surgery types.

Methods

All surgeries performed in a Finnish hospital district from 1 January 2015 to 31 December 2019 were analysed. Exclusion criteria were age below 16 years, unknown smoking status, and unknown ASA class. A machine learning algorithm was developed for smoking status classification. The primary outcome was 90-day overall postoperative complications in all surgeries. Secondary outcomes were 90-day overall complications in specialties with over 10 000 surgeries and critical complications in all surgeries.

Results

The machine learning algorithm had precisions of 0.958 for current smokers, 0.974 for ex-smokers, and 0.95 for never-smokers. The sample included 158 638 surgeries. In adjusted logistic regression analyses, smokers had increased odds of overall complications (odds ratio 1.17; 95 per cent c.i. 1.14 to 1.20) and critical complications (odds ratio 1.21; 95 per cent c.i. 1.14 to 1.29). Corresponding odds ratios of ex-smokers were 1.09 (95 per cent c.i. 1.06 to 1.13) and 1.09 (95 per cent c.i. 1.02 to 1.17). Smokers had increased odds of overall complications in all specialties with over 10 000 surgeries. ASA class was the most important complication predictor.

Conclusion

Machine learning algorithms are feasible for smoking status classification in big surgical data sets. Current and former smoking predict complications in all surgery types.Machine learning algorithms are feasible for smoking status classification in large data sets. Current and former smoking associate with overall and critical complications in all types of inpatient and outpatient surgery of various invasiveness.