Health status as a potential mediator of the association between hemodialysis vascular access and mortality.

BACKGROUND It is unknown whether the selection of healthier patients for arteriovenous fistula (AVF) placement explains higher observed catheter-associated mortality among elderly hemodialysis patients. METHODS From the United States Renal Data System 2005-2007, we used proportional hazard models to examine 117 277 incident hemodialysis patients aged 67-90 years for the association of initial vascular access type and 5-year mortality after accounting for health status. Health status was defined as functional status at dialysis initiation and number of hospital days within 2 years prior to dialysis initiation. RESULTS Patients with catheter alone had more limited functional status (25.5 versus 10.8% of those with AVF) and 3-fold more prior hospital days than those with AVF (mean 18.0 versus 5.4). In the unadjusted model, the likelihood of death was higher for arteriovenous grafts (AVG) {Hazard ratio (HR) 1.20 [95% CI (1.16-1.25)], catheter plus AVF [HR 1.34 (1.31-1.38)], catheter plus AVG [HR 1.46 (1.40-1.52)] and catheter only [HR 1.95 (1.90-1.99)]}, compared with AVF (P < 0.001). The association attenuated -23.7% (95% CI -22.0, -25.5) overall (AVF versus all other access types) after adjusting for the usual covariates (including sociodemographics, comorbidities and pre-dialysis nephrology care) {AVG [HR 1.21 (1.17-1.26)], catheter plus AVF [HR 1.27 (1.24-1.30)], catheter plus AVG [HR 1.38 (1.32-1.43)] and catheter only [HR 1.69 (1.66-1.73)], P < 0.001}. Additional adjustment for health status further attenuated the association by another -19.7% (-18.2, -21.3) overall but remained statistically significant . CONCLUSIONS The observed attenuation in mortality differences previously attributed to access type alone suggests the existence of selection bias. Nevertheless, the persistence of an apparent survival advantage after adjustment for health status suggests that AVF should still be the access of choice for elderly individuals beginning hemodialysis until more definitive data eliminating selection bias become available.