Surgeon Factors Have a Larger Effect on Vascular Access Type and Outcomes than Patient Factors
Timothy Copeland, Peter Lawrence, Karen Woo.
UCLA, Los Angeles, CA, USA.
Objective:To determine influence of unmeasured individual surgeon factors on initial vascular access (VascAcc) selection and outcomes in patients initiating hemodialysis with a tunneled catheter(TC).
Methods:The 2011 to 2016 Optum Clinformatics claims database was used. Primary outcomes were VascAcc type(arteriovenous graft versus fistula), likelihood of repeat VascAcc, time to TC removal and time to repeat VascAcc. Multivariable models were fitted with (WithSurg) and without (NoSurg) accounting for individual surgeons.
Results:5299 VascAcc met inclusion criteria:4119(78%) fistula and 1180(22%) graft performed by 1215 surgeons with mean 4.3 cases/surgeon(range 2-44). Median follow-up was 593 days(range 1-2543). Mean patient age was 66.7, with 3035(57.3%) male and 2380(44.9%) White. The NoSurg model predicting VascAcc type had AUC 0.64; the WithSurg model had AUC 0.86(Table), indicating that accounting for individual surgeons significantly improved model discrimination. The patient demographic/comorbidity covariate with the largest effect in the NoSurg and WithSurg models predicting VascAcc type was male sex(OR 0.57 and 0.52, respectively)(Table). Accounting for individual surgeons did not significantly change the effect size of male sex or other covariates, indicating that accounting for individual surgeons offers additional predictive value independent of covariates. The surgeon median odds ratio(MOR) in the VascAcc type model was 2.68(Table), indicating that between two patients with identical characteristics but two different surgeons, the odds of undergoing graft varied by a factor of 2.68, making surgeon effect 5.6 times larger than male sex effect. A similar pattern was found in the odds of repeat VascAcc model(Table). In time to TC removal and time to repeat VascAcc models, WithSurg models had significantly lower Akaike Information Criterion(AIC) and
Bayesian Information Criterion(BIC), without changing covariate effect sizes, indicating improved prediction precision and independent additional predictive value of accounting for individual surgeons(Table). In the time to TC removal model, only VascAcc type had a larger effect than surgeon. In the time to repeat VascAcc model, surgeon effect was larger than the effect of all other covariates(Table).
Conclusion:Unmeasured factors associated with surgeon account for more variation in VascAcc type and outcomes of VascAcc than measurable patient demographics/co-morbidities. Future research should focus on identifying which surgeon factors are associated with improved outcomes.
|Model Resultsa||Without Surgeon Random Effect||With Surgeon Random Effect|
|Outcome: Odds of Vascular Access Type=Graft|
|Outcome: Time to Tunneled Hemodialysis Catheter Removal|
|Graft (vs Fistula) HRb||1.83**||1.78**|
|Male Hazard Ratio||1.14***||1.18***|
|Surgeon Median HR||-||1.57|
|Outcome: Odds of Repeat Vascular Access|
|Graft (vs Fistula) ORc||0.37*||0.35*|
|Outcome: Time to Repeat Vascular Access|
|Graft (vs Fistula) HR||0.41*||0.39*|
|AUC Area Under the Curve; OR Odds Ratio; MOR Median Odds Ratio; AIC Akaike Information Criterion; BIC Bayesian Information Criterion; MHR Median Hazard Ratio; *significant at P<0.05; **significant at P<0.01; ***significant at P<0.001; a, all models include age, race, sex, vascular access type, age and vascular access type interaction, diabetes, cardiac arrhythmia, congestive heart failure, peripheral vascular disease and obesity; b, Hazard Ratio for a 68 year old patient; c, Odds Ratio for a 68 year old patient|
Back to 2019 Abstracts