Introduction: We have recently shown (Alavi et al. Circulation, (2021)144: A13745-A13745) that adverse effects of nicotine delivered chronically by electronic cigarette (EC) vapor or standard cigarettes on left ventricular systolic function can be captured using intrinsic frequency (IF) method applied on carotid waveforms. Here, we propose a hybrid IF-machine learning (ML) method to detect nicotine effect on cardiovascular system using a carotid waveform. Methods: Total number of n=117 young healthy adult male and female Sprague Dawley rats (49% (n=57) female, weight ~200-250 g) were randomized and exposed to: 1) purified air, n=32; 2) EC vapor without nicotine (EC NIC-), n=26; 3) EC vapor plus nicotine (EC NIC+), n=27; and 4) standard cigarette smoke from reference combustion cigarettes (3R4F), n=32. All the exposures (nose-only) took place for duration of 5 hours/day, 4 days/week for total of 8 weeks. Third-generation type EC from VaporFi, tank model: Volt 2 was used. E-liquid was tobacco flavored with a 50/50 propylene glycol/glycerin ratio. Similar nicotine amount as the standard cigarette was delivered to the rats of EC NIC+ group. After 8 weeks of exposure, IFs were computed from invasively measured carotid waveforms. A support vector machine (SVM) classifier with radial basis function kernel was trained using IF data from 83 rats to detect (non)-nicotine groups. The k -fold cross-validation ( k =10) was used to avoid overfitting. The remaining rats were used for generalization test (n=14) and stratified blind test (n=20). Results: Our SVM model showed positive and negative predictive values of 66.7% and 76.9%, respectively. Sensitivity and specificity were 82.4% and 58.8% for test data (Fig1). Conclusions: Our results suggest that nicotine delivered by ECs or cigarettes can be detected by a physics-based ML model from a single carotid waveform. This method can potentially be used for detecting adverse effect of nicotine on cardiovascular system noninvasively.
Read full abstract