Table 3.
Logistic regression models with and without random effects for the risk of ICU admission and/or in-hospital death
| Fixed-effects model | Mixed-effects model | |||||
|---|---|---|---|---|---|---|
| Predictors | Odds ratios | CI | p | Odds ratios | CI | p |
| LUSs | 1.28 | 1.08 – 1.65 | 0.016 | 5.64 | 1.19 – 26.68 | 0.029 |
| Age | 0.91 | 0.80 – 1.00 | 0.101 | 0.75 | 0.51 – 1.10 | 0.138 |
| Random effects | ||||||
| σ2 | 3.29 | |||||
| τ00 | 4151.89 Subject | |||||
| ICC | 1.00 | |||||
| R2 Tjur | 0.350 | 0.037/0.999 | ||||
| AIC | 28.026 | 20,534 | ||||
The most informative models according to AIC were those with LUSs and age as fixed effects terms; in the mixed effects model, addition of random between-subjects intercepts did improve the AIC but did not lead to improved model predictivity. The addition of other terms as specified in the Methods section did not significantly improve the AIC in either the fixed effects or mixed model. AIC Akaike’s information criterion, ICU intensive care unit