Tjallie van der Kooi

Proc phreg data=total covs(aggregate); class hospital Patientno sex; model CVCtime*infection2(0)= sex age1 age2 age3 ICUstay1 ICUstay2 ICUstay3 bacteraemia insertiondepartment1 insertiondepartment2 vein1 vein2 vein3 vein4 admissiontype1 admissiontype2 ICUtype1 ICUtype2 ICUtype3 ICUtype4 ICUtype5 ICUtype6 ICUtype7 ICUtype8 UseforAntibiotics UseforTPV UseforBloodProducts intervention1 intervention2 intervention3 / rl; id Patientno; strata hospital; run; Table M3 shows both the hazard ratios (HR) resulting from the sub‐distribution hazard analysis as well as from the event‐specific hazard analysis, i.e. without taking competing events into account. As can be seen, the hazard ratios are quite similar, suggesting that the competing events were only slightly affected by the intervention. This can be checked by performing a Cox regression on these outcomes as well. Table M3. Hazard ratios (HR) with and without taking competing events into account, with 95% confidence intervals Sub‐distribution HR (95% CI) Event‐specific HR (95% CI) HHint (intervention1) 0.46 (0.28 – 0.74) 0.50 (0.31 – 0.81) CVCint (intervention2) 0.59 (0.43 – 0.81) 0.64 (0.46 – 0.89) BOTHint (intervention3) 0.33 (0.24 – 0.47) 0.42 (0.30 – 0.58) Trend analysis There was a significant decreasing the baseline quarters, with a sub‐distribution hazard ratio (HRsub) of 0∙93 per quarter (0∙84 – 1.02). Therefore we also analysed the results accounting for a baseline that is an autonomous hospital‐specific trend during the entire study period (variable period 2) with a post‐intervention trend on top of that (variable post period x), allowing random trends for intervention arms. The intended random multilevel Poisson model did not converge, however. Therefore a sub‐distribution hazard regression, with hospital and the interaction of hospital with baseline trend as covariates was performed (again by extending the CVC duration for catheters without infection); Proc phreg data=total covs(aggregate); class hospital Patientno sex; model CVCtime*infection2(0)= hospital – covariates as above – intervention1 intervention2 intervention3 periode2 hospital*periode2 postperiodx interventiontype*postperiodx /rl; id Patientno; run; The formal start of the CVC intervention was not associated with a reduction in the CRBSI risk anymore (HRsub CVCi 1∙16 (0∙63 –2∙16) but the formal start of the interventions in both the HHi and COMBi groups successfully further reduced the CRBSI risk (HRsub HHi and COMBi 0∙37 (0∙16 – 0∙87) and 0∙47 (0∙27 – 0∙83) respectively). Multivariable analysis with Apache II score Apache II scores were recorded – for limited periods only – in 11 hospitals. SAPS II scores were recorded in five hospitals. We performed the multivariable analysis allowing for trends in the subset with known Apache II scores and found comparable results (Table M4). An Apache II score of 10‐19 was associated with a marginally significant hazard ratio of 1.45 (p‐value 0.07) in this analysis.

RkJQdWJsaXNoZXIy MTk4NDMw