Original ventilator days (%) * LOCF ventilator days (%) ** Optimal model HR 95% CI p‐value AIC Relevant exposure window & number of knots for best‐fitting model HR & 95%CI Adjusted p‐ value (multiple testing)***** AIC ivSDD 2‐day delay 986 28 days Fig 3B 0.003 973 WCE Yes 1160 (14.7) 1160 (14.7) 0.43 (0.15, 1.26) 0.123 1 knot No 6710 (85.2) 6712 (85.3) Ref missing 2 (0.03) 0 (0.0) Intestinal prophylaxis Current 970 12 days Fig 3E 0 965 WCE Yes 2395 (30.4) 2395 (30.4) 0.12 (0.04, 0.41) 0.0007 2 knots No 5475 (69.6) 5477 (69.6) Ref missing 2 (0.03) 0 (0.0) Oropharyngeal prophylaxis Current 967 21 days Fig 3D 0 971 Current Yes 4783 (60.8) 4784 (60.8) 0.14 (0.05, 0.38) 0.0001 1 knot No 3087 (39.2) 3088 (39.2) Ref (unconstrained) missing 2 (0.03) 0 (0.0) * The total number of ventilation days was 7872. ** The number of ventilation days after reducing the numbers of missings for time‐dependent covariates with the ‘last observation carried forward (LOCF) approach *** Selected as the model with the lowest AIC among the current exposure‐risk, 1‐day delay exposure‐risk, and 2‐day delay exposure‐risk models in Table C in S1 File. **** Preference given to a non‐cumulative model. WCE model selected if the WCE model had an AIC that was 4 units lower than that of the best‐fitting non‐cumulative model. ***** The p‐value was estimated using 1000 bootstrapped data sets to account for multiple testing when selecting the best‐fitting WCE model. 74 Chapter 4
RkJQdWJsaXNoZXIy MTk4NDMw