Binary logistic regression model of factors predictive of death at 120 days from the point of diagnosis of pneumomediastinum (all patients eligible for mechanical ventilation, n=315)
| B±se | OR or % increase per unit (95% CI) | p-value | |
| Mechanically ventilated (at diagnosis) | 1.40±0.26 | 4.0 (2.4–6.7) | <0.001 |
| Age | −0.38±0.12 | 3.7% (1.4–5.9%) per year | <0.01 |
| Diabetes mellitus | 0.59±0.34 | 1.8 (0.9–3.5) | 0.08 |
All variables significantly associated with mortality in univariate analyses were entered into the model backwards stepwise. The model produces a prediction accuracy for outcome of 68.4% versus a 51.1% default accuracy. Model R2=0.251, Nagelkerke, χ2(6)=65.2, p<0.001. Constant B±se=2.38±0.79. Variables in the regression but not listed: subcutaneous emphysema (p=0.12), hypertension (p=0.35) and ischaemic heart disease/left ventricular systolic dysfunction (p=0.50).