Multivariable logistic regression model used to predict postoperative atrial fibrillation (n = 109) in a subgroup of 9789 patients who had NT-proBNP measured before surgery*
| Variable | OR (95% CI) |
|---|---|
| NT-proBNP, ng/L | |
| 100 | 1 (Ref.) |
| 200 | 1.31 (1.15–1.49) |
| 1500 | 2.07 (1.27–3.36) |
| 3000 | 2.39 (1.26–4.51) |
| Age, yr | |
| 50 | 1 (Ref.) |
| 65 | 2.06 (0.92–4.61) |
| 80 | 4.37 (1.91–9.98) |
| Type of surgery | |
| Low-risk | 1 (Ref.) |
| Major nonthoracic | 3.23 (1.67–6.25) |
| Major thoracic | 4.71 (1.57–14.14) |
| Demographic and clinical characteristics | |
| Male sex | 1.20 (0.81–1.78) |
| Hypertension | 0.93 (0.61–1.43) |
| Diabetes mellitus | 1.21 (0.76–1.92) |
| Coronary artery disease | 1.12 (0.69–1.81) |
| Chronic obstructive pulmonary disease | 2.04 (1.24–3.33) |
Note: CI = confidence interval, NT-proBNP = N-terminal pro–brain-type natriuretic peptide, OR = odds ratio, Ref. = reference category.
↵* We modelled NT-proBNP and age as continuous variables using restricted cubic splines. Splines are best interpreted using a figure, but they also allow the comparison of any 2 desired values of the predictor that can be easily interpreted as an OR without the need to refit the model. We used previously established thresholds for NT-proBNP and clinically meaningful increments for age. Odds ratios presented in the table can be interpreted as relative risks owing to the low baseline risk of the outcome.