Skip to main content
. 2019 Nov 1;14(11):e0224582. doi: 10.1371/journal.pone.0224582

Table 2. Summary of key AF risk factors inferred from the final risk model.

Risk factor
Patient demographics (age, sex, race, smoking status) at baseline
History of antihypertensive medication use at baseline
History of type 1 or type 2 diabetes at baseline
History of cardiovascular comorbidities at baseline
Presence of a cardiovascular event in the past year
BMI in each of the latest four quarters
Change in BMI between the latest two quarters
High pulse pressure in the latest quarter
Negative absolute change in DBP or positive absolute change in SBP between the latest two quarters
Increasing frequency of DBP, SBP and BMI recording in the latest quarter

BMI: body mass index; DBP: diastolic blood pressure; SBP: systolic blood pressure. Risk factors were identified using model inferences from variable importance plots[23] and partial dependence plots[24] for the fitted baseline and time-varying neural networks. Smoking status was defined as: current smoker, former smoker, non-smoker, passive smoker and unknown. Cardiovascular comorbidities/events considered were: hypertension (diagnosed), hypertension (receiving antihypertensive medication), heart failure, coronary heart disease, congenital heart disease, myocardial infarction left ventricular hypertrophy, type 1 diabetes, type 2 diabetes. Pulse pressure is the difference between systolic and diastolic blood pressure, corresponding to the force generated by cardiac contraction. “High” pulse pressure refers to elevated (>120 mmHg) SBP combined with a low to normal (≤80 mmHg) DBP. For details, see S3 Fig. Frequency with which clinical characteristics were recorded was a continuous variable representing the number (count) of DBP, SBP, etc. recordings, rather than actual values recorded over a given quarter.