Skip to main content

Table 2 Three sequential random effects and multilevel Cox regression models were used to evaluate the impact of frailty on both 30-days survival

From: The impact of frailty on survival in elderly intensive care patients with COVID-19: the COVIP study

  Model 1 unadjusted HR (95% CI) p value Model 2 adjusted HR (95% CI) p value Model 3 adjusted HR (95% CI) p value
Survival at 30 days
 Vulnerable versus fit 1.75 (1.35–2.25)  < 0.001 1.55 (1.14–2.10) 0.011 1.14 (0.79–1.65) 0.4811
 Frail versus fit 3.20 (2.56–4.13)  < 0.001 2.41 (1.77–3.27)  < 0.001 1.86 (1.36–2.52)  < 0.001
Treatment limitation
 Vulnerable versus fit 2.26 (1.73–2.96)  < 0.001 1.7 (1.21–2.38) 0.0021   
 Frail versus fit 3.98 (3.08–5.21)  < 0.001 2.79 (1.96–3.91)  < 0.001   
Mechanical ventilation
 Vulnerable versus fit 0.83 (0.67–1.01) 0.055 0.92 (0.73–1.16) 0.5   
 Frail versus fit 0.75 (0.62–0.92) 0.005 0.69 (0.54–0.87) 0.0043   
Non-invasive ventilation
 Vulnerable versus fit 1.58 (1.11–2.25) 0.011 1.22 (0.79–1.88) 0.37   
 Frail versus fit 1.58 (1.12–2.24) 0.009 1.26 (0.8–1.95) 0.32   
Non-invasive ventilation/mechanical ventilation
 Vulnerable versus fit 0.94 (0.78–1.13) 0.51 1 (0.81–1.24) 0.99   
 Frail versus fit 0.84 (0.7–1.02) 0.069 0.74 (0.58–0.91) 0.0096   
Vasoactive drugs
 Vulnerable versus fit 0.99 (0.81–1.22) 0.93 1.09 (0.86–1.39) 0.47   
 Frail versus fit 1.01 (0.84–1.25) 0.88 0.9 (0.7–1.15) 0.44   
Renal replacement therapy
 Vulnerable versus fit 1.23 (0.81–1.86) 0.33 1.14 (0.7–1.87) 0.61   
 Frail versus fit 1.62 (1.1–2.45) 0.014 1.01 (0.59–1.65) 0.98   
  1. First, we estimated the impact of frailty on outcome without adjustment on confounding using a baseline model including only frailty (model 1). Second, to estimate the impact of frailty when adjusting on patients’ baseline characteristics the following covariates were added to model 1: age, sex, comorbidities, SOFA score, BMI, PaO2/FiO2. Third, to evaluate whether the effect of frailty was independent of ICU management strategies, both organ support and treatment limitation (model 3) were added to model 2 as time-dependent covariates. For all outcomes, significance of the random centre effect was tested comparing the likelihood of two models including frailty with and without random effect. Random effect was significant for all outcomes. No violation of the proportional hazard assumption was detected in the models