Multiple Pre-operative and Intra-operative Factors Predict Early Fistula Thrombosis in the Hemodialysis Fistula Maturation Study

Abstract

OBJECTIVES

Early thrombosis (ET) contributes to autogenous arteriovenous fistula (AVF) failure. We studied patients undergoing AVF placement in the Hemodialysis Fistula Maturation (HFM) Study, a prospective, observational cohort study, using a nested case-control analysis to identify pre-operative and intra-operative predictors of ET.

METHODS

ET cases were compared to controls who were matched on gender, age, diabetes, dialysis status, and surgeon fistula volume. ET was defined as thrombosis diagnosed by physical exam or ultrasound within 18 days of AVF creation. Conditional logistic regression models were fit to identify risk factors for ET.

RESULTS

Thirty-two ET cases (5.3%) occurred among 602 study participants; 198 controls were matched. ET was associated with female gender (OR=2.75, CI 1.19–6.38, P=0.018), fistula location (forearm vs. upper arm) (OR=2.76, CI 1.05–7.23, P=0.039), feeding artery (radial vs. brachial) (OR=2.64, CI 1.03–6.77, P=0.043) and arterial diameter (OR=1.52, CI 1.02–2.26, P=0.039, per mm smaller). Draining vein diameter was nonlinearly associated with ET, with highest risk in 2–3 mm veins. Surprisingly, ET risk was lower in diabetics (OR=0.19, CI 0.07–0.47, P=0.0004), lower with less nitroglycerin-mediated brachial artery dilatation (NMD%) (OR=0.42, CI 0.20–1.92, P=0.029 for each 10% lower) and higher with lower carotid-femoral pulse wave velocity (OR=1.49, CI 1.02–2.20, P=0.041, for each m/sec lower). Intraoperative protamine use was associated with a higher ET risk (OR 3.26, CI 1.28-∞, P=0.038). Surgeon’s intraoperative perceptions were associated with ET: surgeons’ greater concern about maturation success (likely, marginal, unlikely) was associated with higher thrombosis risk (OR 8.09, CI 4.03-∞, p<0.0001, per category change), as were absence vs. presence of intraoperative thrill (OR 21.0, CI 5.07-∞, P=0.0002) and surgeons’ reported frustration during surgery (OR 6.85, CI 2.70-∞, P=0.0004). Reduced extent of intraoperative thrill (proximal, mid or distal third of the forearm or upper arm, based on AVF placement) was also associated with ET (OR 2.91, CI 1.31-∞, P=0.014, per diminished level). Oral antithrombotic medication use was not significantly associated with ET.

CONCLUSIONS

ET was found to be associated with female gender, forearm AVF, smaller arterial size, draining vein diameter of 2–3 mm, and protamine use. Paradoxically, diabetes and stiff, noncompliant feeding arteries were associated with lower frequency of ET. Absent or attenuated intraoperative thrill, and both surgeon frustration and concern about successful maturation during surgery, were strongly correlated with ET.

INTRODUCTION

Catheters and both autogenous and prosthetic arteriovenous access are used in maintenance hemodialysis, with arteriovenous fistulas (AVF) preferred because of longer patency, less frequent intervention to maintain patency and lower health care costs^1–3 However, AVF maturation failure rates are high varying from 18%–53%^{3, 4}. Early thrombosis (ET),^4–14 which may represent the most aggressive form of primary access failure, has been reported to occur in 6.3–19.5% of fistulas, and is associated with preoperative patient factors including thrombophilic factor gene polymorphisms¹⁵ and radial artery diameter¹⁶, intraoperative factors such as surgeon experience¹⁷, end-diastolic velocity in the proximal feeding artery after AVF construction, and absence of bruits and postoperative factors including arterial resistive index¹⁸ and use of anticoagulants^{4, 12}, as well as intraoperative⁶ and postoperative¹⁰ fistula blood flow.

Evaluation of risk factors for ET has been hampered, in part, by lack of a standard definition and the reporting of ET as a component of maturation failure rather than as a separate outcome¹⁹. Identifying patients at high risk for ET could lead to better selection of candidates for AVF creation and/or suggest more appropriate remedial post-operative procedures. We studied participants in the Hemodialysis Fistula Maturation (HFM) Study, a 7-center prospective, observational cohort study of patients with newly created AVFs, to identify pre- and intra-operative risk factors for ET.

METHODS

The design of the HFM Study has been described in detail previously²⁰. Eligibility criteria included age < 80 if not yet on maintenance dialysis, life expectancy of > 9 months, current or anticipated need for maintenance hemodialysis within 3 months of AVF construction and placement of an autogenous, single-stage AVF. Information obtained pre-operatively included demographic characteristics, comorbid illnesses and self-reported use of medications. Brachial artery flow-mediated (FMD) and nitroglycerin-mediated (NMD) dilatation, arterial pulse wave velocity (PWV) measurement and venous occlusion plethysmography were performed at baseline to assess vascular function. Duplex ultrasound measures including vessel diameters, flow rates, and arterial calcification were used to assess vascular anatomy. Processes of care including training and experience of the surgical team, technique for AVF creation, intraoperative management, dialysis practices, perioperative medications, vessel size, anesthesia type, procedural details and surgical duration were noted. The attending surgeon’s intraoperative assessment of AVF thrill (absent or extending to proximal, mid, or distal third of upper arm or forearm), expressed frustration (yes/no), and prediction of success (likely, marginal, unlikely) were recorded. Intraoperative fistula vein samples were obtained for histopathological analysis of hyperplasia and calcification. Postoperative serial duplex ultrasounds were performed within 3 days and at 2 and 6 weeks after surgery in 94%, 92% and 88% of participants, respectively, and as clinically requested by the participant’s physician. (See Supplemental Materials for methods of ultrasound, vascular function, and histological assessments.) The HFM study was approved by the Institutional Review Boards of all participating institutions and all patients were consented to participate.

Thrombosis was diagnosed clinically during scheduled HFM post-operative visits and also at duplex ultrasound examinations. We postulated that early thrombotic cases represented a distinct set of failure mechanisms that would manifest themselves rapidly. We defined ET as those cases that occurred within 18 days after fistula creation, allowing incorporation of the 2-week ultrasound exam and after which a gap and substantial decline in frequency of ET diagnoses was observed. (Supplemental Materials figure 1.)

Due to the statistical limitations of multivariable modeling with few outcome cases per predictor, we performed a nested case-control analysis. Control patients were those without ET who could be simultaneously matched to any ET patient (case) based on sex, age (± 8 years), diabetes, whether or not on maintenance dialysis at time of enrollment and surgeon past experience in creating AVFs based on the attending surgeon’s reported number of AVF operations performed in 2007 through 2009 (just prior to initiation of the HFM Study consortium, ± 70). These controls were optimally partitioned into disjoint sets, matched to each individual case²¹ (Appendix A). To study gender in relation to ET, adjusted for other matching variables, we created new matched sets after removing gender from the matching criteria, and proceeded analogously for other matching variables.

The method of case-control matching, variables for case-control comparisons, choices of one- or two-sided testing and criterion for statistical significance were predetermined for each variable. We also designated outcome hierarchies (primary, secondary, other) or pre-specified Bonferroni multiple comparison adjustments, to limit the potential occurrence and emphasis placed on false positive results from multiple testing of correlated variables. (Appendix B)

Conditional logistic regression within the resulting matched sets was used to estimate and test associations of ET with candidate variables, with splines used to screen for general nonlinear effects (Appendix A). Statistically significant associations were reexamined after covariate-adjustment for oral antithrombotic medication defined as the use of aspirin, clopidogrel, warfarin, or aspirin/dipyridamole reported at the screening visit (yes/no).

Since FMD% and NMD% may not adequately account for variation in vessel size between individuals,²² we included two allometrically-adjusted alternatives to these percentage measures, based on linear regression of log (post-stimulus vessel diameter) on log (pre-stimulus vessel diameter), as exploratory measures. These replace each percentage measure as predictor respectively by i) the patient’s residual from this regression, or ii) the ratio of post-stimulus diameter to (pre-stimulus diameter)^b, where b is the estimated regression coefficient.²²

Joint distributions of several intraoperative variables found to be strongly associated with ET were examined for redundancy. To assess and illustrate the potential predictive capacity of intraoperative surgeon perceptions, guided by bivariate associations we devised a simple ad hoc algorithm combining two predictors, and calculated its sensitivity and specificity (the respective fractions of ET cases and non-cases correctly anticipated by the algorithm), and its positive (PV+) and negative (PV−) predictive values (the respective fractions of the algorithm’s ET and no-ET predictions that proved to be correct) within the full HFM cohort.²³

Computations were performed using SAS 9.3 PROC LOGISTIC and other components, and the LGTPHCURV9 macro²⁴ for spline fitting. Unless otherwise noted, statistical significance was assessed using two-sided 5% level tests.

RESULTS

Cases and matched sets

The 602 HFM Study participants had a mean age of 55.1±13.4 years with 37% being over 60 years old. Thirty percent of participants were female, 44% were African-American, 59% had diabetes and 64% were on maintenance dialysis (see the Supplemental Materials and Supplemental Table I for additional summary information regarding this cohort and the specific timing of all ET diagnoses through 60 days, post-operatively). Thirty-five patients developed ET within the first 30 days after AVF creation and 32 patients (5.5% of HFM enrollees) were diagnosed within the first 18 days of surgery. Of these 32 patients, 9 were diagnosed by physical exam, 2 by duplex ultrasound and 21 using both modalities.

Consistent with prior studies, ET was more frequent among women than men (7.6 vs. 5.0%), and in AVFs constructed from vessels 2.0–3.0 mm than from those > 3.0 mm in diameter (arteries 8.5 vs. 4.8%, veins 8.8 vs. 4.5%). ET occurred in 3 of 12 (25%) AVFs constructed from < 2.0 mm arteries but, interestingly, in only 3 of 93 constructed from < 2.0 mm veins, including 0 of 30 such AVFs in women. See Supplemental Table 2 for details.

A total of 198 HFM Study participants without ET were matched to ET cases, with the number of controls varying from 1 to 38 per case. Table I summarizes the five matching variables, as well as antithrombotic usage, for cases, all non-cases and the averages of controls matched to each case. The proportions of cases and weighted proportions of controls with each dichotomous matching factor were identical, by construction. Mean age was matched within 1.4 years, mean surgeon fistula creation experience within 1.1 surgeries and baseline reported oral antithrombotic usage was noted to be within 2.2%.

Table I.

Matching Variables and Maintenance Antithrombotic Therapy Among HFM Study Participants with Early Fistula Thrombosis, those without ET and Controls.

	Subgroup
	Participants without ET (N= 570)	Participants with ET (N=32)	Weighted1, Matched Controls (N=198)
Variable	%, or mean ± SD
Female	29.3%	40.6%	40.6%
Age (years)	55.5 ± 13.4	49.2 ± 11.8	50.5 ± 11.2
Diabetes	60.1%	21.9%	21.9%
On maintenance hemodialysis	64.2%	65.6%	65.6%
Attending surgeon fistula surgery experience2	169.7 ± 169.7	211.3 ± 145.8	210.3 ± 145.5
Maintenance antithrombotic therapy use at screening	53.3%	37.5%	39.6%

ET: early fistula thrombosis. N: sample size. SD: standard deviation.

¹Weighted statistics are simple averages of summaries (means or %s) within the sets of controls matched to each individual ET case.

²Fistulas (total number) created by attending surgeon in the 3 years prior to initiation of the HFM Study.

Baseline clinical and demographic characteristics (Table II)

Table II.

Association of baseline clinical and demographic characteristics with early fistula thrombosis.¹

Risk Factor	N2	Odds Ratio	95% CI	P-value
Demographic Factors:
Female vs. Male3	327	2.75	1.19–6.38	0.018
Age (per decade4)	445	0.86	0.65–1.14	0.28
African-American vs. Other Races	222	1.90	0.80–4.50	0.15
Other Non-clinical Factors:
BMI>30 (kg/m2)	230	1.19	0.53–2.70	0.67
Ever Smoker	228	0.70	0.31–1.57	0.39
Pack-Years, Among Ever-Smokers	71	1.01	0.99–1.03	0.41
Years Since Last Smoked Among Former Smokers (per decreasing decade)	49	1.08	0.96–1.22	0.20
Preoperative Diagnoses:
Diabetes5	360	0.19	0.07–0.47	0.0004
Known hypercoagulable state	230	0.714	0.14–3.58	0.68
Kidney Disease Diagnosis
Glomerulonephritis vs. Pooled Diabetic Nephropathy/Ischemic Renal Disease	102	1.45	0.14–15.3	0.82
Hypertensive Nephrosclerosis vs. Pooled Diabetic Nephropathy/Ischemic Renal Disease		0.98	0.10–9.43
Glomerulonephritis vs. Hypertensive Nephrosclerosis		1.49	0.42–5.26
Current therapies:
Chronic (Maintenance) Dialysis6	331	1.15	0.46–2.89	0.77
Ipsilateral Catheter Use	155	0.99	0.11–8.55	0.99
No Antithrombotic Medication7	230	1.23	0.52–2.93	0.63

N: sample size. CI: confidence interval.

¹Unless otherwise noted, case-control comparisons are matched on baseline sex, age, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium), and hypothesis tests are two-tailed with p ≤ 0.05 required for statistical significance.

²Number of patients in matched sets that are statistically informative, in the sense that data on the risk factor was present for both at least one case and one matched control.

³Matched on baseline age, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium).

⁴Matched on baseline sex, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium).

⁵Matched on baseline age, sex, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium).

⁶Matched on baseline age, sex, diabetes, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium)

⁷No current use of aspirin, clopidogrel, warfarin, or aspirin/dipyridamole reported at screening visit.

Early thromboses were significantly more common in females (OR=2.75, CI 1.19–6.38, P=0.018) and seen more frequently among African-Americans than other races, though the latter effect was not statistically significant (OR=1.90, CI 0.80–4.50, P=0.15) (Table II). Estimated ORs for age, Body Mass Index (BMI), smoking, hypercoagulability disorder, renal diagnosis, current maintenance hemodialysis, ipsilateral catheter use and use of antithrombotic medications were not significant.

Of note was the low frequency of ET among participants with diabetes. Only 2% of the study participants with diabetes experienced ET compared to 10% of those without diabetes. The inverse association of diabetes with ET was highly statistically significant after matching on age, gender, dialysis status, and surgeon fistula creation experience (OR=0.19, CI 0.07–0.47, P=0.0004) (Table II). The proportions of upper arm (vs. forearm) AVFs were similar for cohort members with and without diabetes (79.6% and 73.4%, respectively). Participants with a self-reported history of diabetes more frequently reported antithrombotic medication use at baseline than those without diabetes (61.5% and 39.8%, respectively) and had stiffer arteries as measured by carotid-femoral PWV (means of 11.74 vs. 9.40 m/sec). Adjustments for these and other potential confounders, including fistula location, feeding artery and its diameter, only modestly attenuated the inverse association between diabetes and ET (data not shown).

Baseline vascular anatomic characteristics (Table III)

Table III.

Association of baseline vascular anatomical characteristics with early fistula thrombosis.¹

Risk Factor	N2	Odds Ratio	95% CI	P-value
Fistula Configuration Variables:
Forearm vs. Upper Arm	225	2.76	1.05–7.23	0.039
Radial vs. Brachial	225	2.64	1.03–6.77	0.043
Basilic vs. Cephalic	225	1.27	0.47–3.44	0.64
Forearm Cephalic vs. Upper Arm Cephalic	225	3.38	1.17–9.77
Upper arm Basilic vs. Upper Arm Cephalic	225	1.89	0.61–5.82	0.077
Upper arm Basilic vs. Forearm Cephalic	225	0.59	0.17–1.85
Feeding Artery Diameter (per 1 mm thinner)	222	1.52	1.02–2.26	0.039
Feeding Artery Cross-Sectional Area (per 1 mm2 reduction)	222	1.98	0.97–4.04	0.060
Minimum Draining Vein Diameter (per 1 mm thinner)	225	Varies with diameter: model P=0.036, nonlinearity P=0.0357.
Minimum Draining Vein Cross-Sectional Area (per 1 mm2 reduction)	225	2.03	0.86–4.77	0.10
Pre-operative Vascular Pathology Measures:
Feeding Artery Ultrasound Calcification Index (per severity level)	224	1.34	0.63–2.86	0.44
Draining Vein Histological Calcification (intimal and/or medial vs. none)	203	1.17	0.24–5.78	0.85
Draining Vein Histological Neointimal Hyperplasia Index (per 10% increase)	101	1.19	0.94–1.50	0.15

N: sample size. CI: confidence interval.

¹Unless otherwise noted, case-control comparisons are matched on baseline sex, age, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium), and hypothesis tests are two-tailed with p ≤ 0.05 required for statisticalsignificance.

²Number of patients in matched sets that are statistically informative, in the sense that data on the risk factor was present for both at least one case and one matched control.

The artery on which the anastomosis was based (radial vs. brachial), and feeding artery and draining vein diameters, pre-specified as primary anatomical variables, were each significantly associated with ET (radial vs. brachial OR=2.6, CI 1.0–6.8, P=0.043; feeding artery diameter OR=1.5, CI 1.0–2.3, P=0.039 per mm narrower; draining vein diameter nonlinearly related, P=0.036, with highest risk over a plateau for 2–3 mm diameter veins). Fistulas located in the forearm had higher odds of ET than those placed in the upper arm (OR 2.8, CI 1.05–7.2, P=0.039). An overall test of differences among the three most common configurations (forearm cephalic, upper arm basilic, upper arm cephalic) approached but did not achieve statistical significance (P=0.08), although forearm cephalic AVF had higher ET odds than upper arm cephalic AVF in direct pairwise comparison (OR 3.38, CI 1.2–9.8). Feeding artery and draining vein calcification, vein neointimal hyperplasia, and, cross-sectional areas were not significantly associated with ET.

Baseline vascular function (Table IV)

Table IV.

Associations of preoperative measures of vascular function with early fistula thrombosis.¹

Risk Factor	N2	Odds Ratio	95% CI	P-value
Brachial Artery Dilation Measures:
FMD% (per 10% decline)	198	0.51	0.21–1.22	0.13
NMD% (per 10% decline)	140	0.42	0.20–0.92	0.029
Arterial Stiffness Measures:
Carotid-Radial Pulse Wave Velocity (per m/sec decline)	101	1.06	0.79–1.43	0.68
Carotid-Femoral Pulse Wave Velocity (per m/sec decline)	101	1.49	1.02–2.20	0.041
Fistula Forearm Vein Function Measures:
Capacitance Slope (per 0.1%/10mmHg pressure decline)	217	1.14	0.94–1.38	0.19
Maximum Output Slope (per (ml/100ml/min)/10mmHg pressure decline)	217	1.21	0.96–1.51	0.10

N: sample size. CI: confidence interval. FMD: flow-mediated dilation. NMD: nitrogen-mediated dilation.

¹Unless otherwise noted, case-control comparisons are matched on baseline sex, age, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium), and hypothesis tests are two-tailed with p ≤ 0.05 required for statistical significance.

²Number of patients in matched sets that are statistically informative, in the sense that data on the risk factor was present for both at least one case and one matched control.

Of note, ET was associated with greater brachial artery nitrogen-mediated dilation (OR per 10% increase in NMD%, a primary variable, = 2.36, CI 1.09–5.10, P=0.029), and with less trunk artery stiffness as assessed by PWV (OR per 1 m/sec decline in carotid-femoral PWV, a secondary variable, = 1.49, CI 1.20–2.20, P=0.041) but not with carotid-radial PWV, a primary variable (OR=1.06, CI 0.79–1.43). The OR for FMD% was similar in magnitude to that for NMD%, but did not achieve statistical significance (Table IV). Allometric alternatives to FMD% and NMD% yielded similar results with slightly higher variability and p-values. The association of ET with NMD was attenuated after allometric adjustment (P=0.062, data not shown). The remaining measures of vascular function were not statistically significantly associated with ET.

Surgeon and surgical factors (Table V)

Table V.

Associations of surgeon and intraoperative variables with early fistula thrombosis.¹

Risk Factor	N2	Odds Ratio	95% CI	P-value
Surgeon Factors:
Recommends post-operative ball squeezing less often (per category: always or almost always, sometimes, never)	230	0.41	0.16–1.01	0.053
Cardiothoracic (CT) Subspecialty Certification3	No CT certified HFM surgeons.
Transplant Subspecialty Certification3	230	2.55	0.68–9.50	0.16
Vascular Subspecialty Certification3	230	0.35	0.12–1.04	0.06
Surgical Subspecialty Certification3	230	0.41	0.94–1.78	0.16
Vascular Access Surgeries in 3 Pre-HFM Years (per 10 fewer surgeries4)	417	0.98	0.95–1.01	0.15
Intra-operative Factors:
Local or Regional Anesthesia	230	2.09	0.82–5.38	0.13
Arteriotomy Length (per mm shorter)	112	0.72	0.51–1.03	0.071
Topical Thrombin5	230	1.06	0.38-∞	0.46
Heparin5,6	230	0.88	0–1.07	0.15
Fixed (vs. weight-based) Heparin Dose6	98	1.68	0.72–3.93	0.23
Protamine5,6	98	3.26	1.28-∞	0.038
Desmopressin5	Virtually unused, hence not tested.
Anastomosis by Fellow or Resident	230	0.42	0.11–1.58	0.20
Topical Vasodilator	230	1.96	0.59–6.49	0.27
Vessel Loops	230	0.99	0.35–2.81	0.99
Surgeon Assessments of Fistula:
No Thrill5	228	21.0	5.07-∞	0.0002
If Thrill, Reduced Extent (per 1/3 forearm/upper arm5)	146	2.91	1.31-∞	0.014
Predicted Success (per <category: likely, marginal, unlikely5)	228	8.09	4.03-∞	<0.0001
Surgeon Frustrated4	217	6.85	2.70-∞	0.0004

N: sample size. CI: confidence interval.

¹Unless otherwise noted, case-control comparisons are matched on baseline sex, age, diabetes, chronic dialysis, and surgeon’s fistula creation surgeries (during 2007–9, preceding HFM Consortium), and hypothesis tests are two-tailed with p ≤ 0.05 required for statistical significance.

²Number of patients in matched sets that are statistically informative, in the sense that data on the risk factor was present for both at least one case and one matched control.

³Prespecified test criterion p ≤ 0.0125 for multiple comparison control.

⁴Matched on baseline sex, age, diabetes, and chronic dialysis.

⁵One-tailed testing with significance criterion p ≤ 0.05.

⁶Protamine use and type of heparin dosing (fixed or weight-based) did not show statistically significant interaction (P=0.80).

No statistically significant associations of ET with subspecialty training of the attending surgeon, or whether the attending or a trainee performed the anastomosis, were found. The following factors were significantly associated with increased risk for ET: Protamine use to reverse heparin-induced anticoagulation (OR 3.26, CI 1.28-∞, P=0.038); surgeon’s perception of thrill (OR=21.0, CI 5.07-∞, P=0.0002 for absence vs. presence of thrill and OR 2.91, CI 1.31-∞, P=0.014 for decreased extent of thrill, when present (proximal vs. mid or mid vs. distal third of either the forearm or upper arm, depending on AVF placement)); and, surgeon’s prediction of fistula outcome (OR 8.09, CI 4.03-∞, p≪0.0001 for each successive worse level (likely, marginal, unlikely) of prognostic concern). Surgeon-reported frustration at the end of surgery was also strongly associated with ET (OR 6.85, CI 2.70-∞, P=0.0004). A common correlate of diminished thrill, prognostic concern and surgeon frustration was surgical duration, which in the full cohort was longer by ~ 0.75 hours when the surgeon was frustrated or when success was viewed as unlikely vs. likely, and by 0.46 hours for surgeries without perceptible thrill at closing. A composite of the surgeon’s belief that success was unlikely or marginal combined with the surgeon’s expressed frustration predicted ET with 44% sensitivity, 97% specificity, 54% positive and 97% negative predictive values. Thus, roughly half (54%) of such fistulas thrombosed early. Absent or attenuated thrill on surgical closing was a related and less subjective negative prognostic sign. Estimated ORs for regional or local vs. general anesthesia, post-operative ball-squeezing recommendation, or use of topical vasodilator, thrombin, or vessel loops, were not statistically significant. Desmopressin was not tested due to very infrequent use.

Antithrombotic usage

Self-reported antithrombotic use ascertained at screening exhibited no clear association with ET. Results for other variables were not materially changed by further adjustment for antithrombotic usage.

DISCUSSION

Early Thrombosis (within 18 days of surgery) occurred in 5.5% of patients in the HFM Study, an unusually large and comprehensive prospective, multi-center, observational cohort study of men and women with newly created AVFs. We examined the Dialysis Access Consortium (DAC) Study database, and found that a virtually identical 6.0% of patients experienced early thrombosis within 18 days (retrieved from DAC Study database by M. Radeva, DAC Study DCC, personal communication). Early Thrombosis may represent the worst functional outcome on the AVF maturation spectrum. Identifying ET predictors could facilitate recognition of patients for whom placing an alternative dialysis access may be preferable to creating an AVF.

Early Thrombosis risk was higher in women than in men, for forearm than upper arm fistulas, for radial than brachial artery fistulas, and for fistulas constructed from smaller caliber arteries or veins, although the trend for veins surprisingly did not extend below 3 mm diameter vessels. Early Thrombosis risk was unexpectedly much lower in diabetic than in non-diabetic patients, and unexpectedly higher in patients with better brachial artery nitrogen-mediated dilation or more compliant central arteries. Absent or attenuated intra-operative thrill and surgeon frustration and concern about the fistula’s prospects at the close of surgery were strongly associated with ET.

Studies support the predictive value of surgeon immediate post-operative risk assessments and frustration during surgery in predicting complications of thyroid, abdominal, and hernia surgeries^24,25,26, although intraoperative surgeon opinion on the success of endoscopic repair of vesicoureteral reflux only poorly predicted cure²⁷, but associations of surgeon factors with AVF thrombosis have not been well studied. In the United States, dialysis access surgery is performed by general, vascular, transplant, and cardiothoracic surgeons, who may or may not be board certified in their respective fields. Our findings support our hypothesis that these surgeons can intraoperatively anticipate the prospects for ET of a newly constructed AVF. However, because our predictive algorithm is ad hoc and estimates of its sensitivity and PV+ are imprecise and vulnerable to over-optimism bias, its use in practice would be inappropriate without first observing and confirming its value in another large series of new AVFs.

Early Thrombosis odds almost tripled in women. Others have also found female gender to be associated with maturation failure²⁸. Diabetes was strongly inversely associated with ET, even after adjustment for potential confounders. This is at odds with prior studies which have shown that diabetes has been directly associated with primary fistula failure¹¹ and patency loss within 6 months of use.²⁹ However, it is roughly consistent with the less pronounced DAC Study results where 4.6% (19/416) of diabetics and 7.3% (33/450) of non-diabetics developed ET (retrieved from DAC Study database by M. Radeva, DAC Study DCC, personal communication). Only a small minority of primary maturation failures are due to ET. In studies of maturation failure by any cause, an inverse association of diabetes with ET may be concealed by direct associations of diabetes with later failure mechanisms, including perhaps later thromboses by other mechanisms than give rise to ET.

Vascular function tests may contribute to cardiovascular risk prediction in uremic and related populations.^30–33 Early Thrombosis was associated with higher brachial artery NMD and lower carotid-femoral PWV, both counterintuitive findings because ET was hypothesized to be associated with impaired arterial dilatation (lower NMD) and increased arterial stiffness (higher PWV). Arterial function is diminished in diabetes, suggesting the unanticipated findings for diabetes and arterial function could be due to confounding among these variables, but confounding by diabetes was reduced by matching on diabetes status, and the diabetes finding was only modestly attenuated by adjustments for other variables.

Forearm AVF experienced ET more often than upper arm fistulas, as did radial artery-based compared to brachial artery-based fistulas. Others have reported higher risks of forearm than upper arm AVFs for primary failure^{11, 29,} and failure to mature.²⁸ Feeding artery diameter was inversely correlated with ET. Pre-operative arterial diameters of radiocephalic AVF has been found to be associated inversely with immediate post-operative thrombosis¹⁶ and thrombosis up to 1 week after surgery⁶. Finally, draining vein diameter was nonlinearly associated with ET. Low fistula vein diameter has been associated with ET⁶ and non-maturation³⁴. Although we observed a decreased incidence of ET incidence with increasing vein diameter >3 mm, ET were also relatively less common in veins < 2 mm wide.

Systemic heparin for anticoagulation was associated with a modestly but not statistically significantly lower ET risk, but risk was higher when protamine was used to reverse heparin’s effect than when heparin was used alone. Protamine is commonly used in vascular surgery³⁵, including dialysis access³⁶, to reverse the effect of heparin. Despite its use, clinical thrombosis is relatively rare having been described after deployment of drug eluting stents³⁷. Notwithstanding, protamine has been described to significantly diminish the risk of bleeding complications without increasing the risk of clinical thrombosis in carotid artery procedures^{38, 39}.

The many potential risk factors and relatively few ETs precluded development of a statistically reliable comprehensive prediction model using all risk factors as covariates. For simplicity, parsimony, and face validity in confounder control, we took a nested case-control perspective, matching controls to cases on age, gender, diabetes, prior dialysis, and surgeon experience. Maintenance antithrombotic therapy was not explicitly matched, but was well balanced between cases and controls. This is important as antiplatelet agents have been demonstrated to decrease the incidence of AVF thrombosis,⁴⁰ although this increase was associated with a randomized prescribed clopidogrel intervention and our HFM results pertain to self-reported usage at screening.

This study has several important positive features. Cases and controls were nested in a large vascular access cohort from 7 geographically dispersed centers. Cases were specifically restricted to the early maturation period. We studied a large number of potential risk factors, obtained under standardized protocols. Details of statistical analysis followed a formal analysis plan, precluded ex post facto searches for statistical significance.

However, we focused on the specific, uncommon problem of ET because we hypothesized that ET might occur through mechanisms distinct from those of later thrombosis and other causes of maturation failure. Thus, results here do not apply to later thromboses or maturation failure generally. Due to this specific focus ET frequency was low despite a relatively large AVF sample size, permitting consideration of only a modest number of potential predictors simultaneously. Measurement of NMD and PWV could not be obtained for substantial fractions of participants. Most importantly, associations from such observational studies are vulnerable to unanticipated confounders, and hence can suggest but not imply causality.

In conclusion, we found ET to be associated with female gender and, unexpectedly, relatively less common among persons with diabetes and, additionally, those whose arteries were stiffer and less able to dilate. Risk was higher when protamine was used to reverse heparin-induced anticoagulation. The surgeon’s intraoperative assessment of thrill, expressed frustration, and fistula prognosis were each strongly associated with ET.

Appendix A: Pre-specified Candidate Predictors¹

Baseline variables

Demographic

Secondary

Age, sex, self-identified African-American vs. other race

Comorbidities

Secondary

Diabetes, obesity (BMI > 30), hypercoagulable state,² renal diagnosis (comparisons of glomerulonephritis to pooled nephropathy/ischemic renal disease; hypertensive nephrosclerosis to pooled diabetic nephropathy/ischemic renal disease; glomerulonephritis to hypertensive nephrosclerosis), chronic dialysis.

Vascular anatomy

Primary

Radial vs. brachial feeding artery, feeding artery diameter, draining vein diameter.

Secondary

Fistula location (all forearm vs. all upper arm), basilic vs. cephalic draining vein, fistula configuration (3 categories: upper arm basilic, upper arm cephalic, forearm cephalic).

Exploratory

Draining vein cross-sectional area, forearm vs. upper arm among cephalic fistulas, basilic vs cephalic among upper arm fistulas, upper arm basilic vs. forearm cephalic fistulas.

Vascular function and pathology

Primary

Fistula arm brachial artery flow-mediated dilation (FMD), expressed as FMD%.

Fistula arm brachial artery nitrogen-mediated dilation (NMD), expressed as NMD%.

Feeding artery ultrasound calcification index (none, mild, moderate to severe).

Draining vein neointimal hyperplasia index.

Draining vein intimal and/or medial calcification. Carotid-radial pulse-wave velocity.

Fistula forearm vein capacitance slope.

Secondary

Fistula arm carotid-femoral pulse wave velocity.

Fistula forearm maximum vein output slope.

Fistula draining vein neointimal hyperplasia areal index.

Exploratory

Rescaled alternatives to FMD%: log (Post to pre inflation vein diameter ratio); log scale post on pre inflation vein diameter regression residual; allometrically adjusted post to pre inflation vein diameter ratio.

Rescaled alternatives to NMD%, as above.

Therapy

Primary

Current ipsilateral catheter use

Secondary

Antithrombotic medication

Cigarette smoking

Secondary

Ever smoker, pack-years (among ever smokers), years since last smoked (among former smokers).

Surgeon factors

No surgical subspecialty certification.³

No cardiothoracic subspecialty certification.3

No transplant surgery subspecialty certification.³

No vascular surgery subspecialty certification.³

Fistulas created within three years prior to HFM Study.

Frequency of recommending post-operative ball squeezing: never, sometimes, always or almost always.

Intraoperative variables

Primary

Arteriotomy length, heparin use,⁴ heparin dosing (fixed or weight-based) when heparin used; protamine,⁴ when heparin used; topical thrombin,⁴ arteriotomy length, absence of thrill,⁴ surgeon prognosis (success likely, marginal, unlikely).⁴

Secondary

General vs. other anesthesia,² desmopressin,⁴ topical vasodilator, vessel loops, thrill extent, if present (in forearm or upper arm thirds),⁴ surgeon frustration (dichotomy),⁴ anastomosis surgeon (Fellow or Resident vs. Attending).

Appendix B. Details of matching and statistical analysis

Details of statistical analysis followed a formal analysis plan. The set of all non-ET cases matched to any case within the prescribed tolerances was optimally partitioned into disjoint subsets matched to each individual case by optimal full matching²¹ which i) enforced matching of sex, diabetes, and dialysis status, and ii) under this constraint, chose the partition minimizing the sum of the following weighted index of case-control age and surgeon experience disparities:

$$\begin{array}{c}2\times \mid \text{case}\text{age}-\text{control}\text{age}\mid /(\text{age}\text{tolerance})\\ +\\ 1\times (\mid \text{case}\text{surgeon}’\mathrm{s}\text{prior}\text{AVFs}-\text{control}’\mathrm{s}\text{surgeon}’\mathrm{s}\text{prior}\text{AVFs}\mid )/(\text{prior}\text{AVF}\text{tolerance})=\\ \mid \text{case}\text{age}-\text{control}\text{age}\mid /4+\mid \text{case}\text{surgeon}’\mathrm{s}\text{prior}\text{AVFs}-\text{control}’\mathrm{s}\text{surgeon}’\mathrm{s}\text{prior}\text{AVFs}\mid /70,\end{array}$$

Optimizations were performed by the R package optmatch⁴¹.

Relationships of ET to continuous predictors were initially examined for nonlinearity using natural cubic splines with internal knots at the first and third quartiles, and boundary knots at the observed extrema of the predictor, by comparing the fitted spline and linear logistic regression models based on the same predictor. If significant nonlinearity was found by this 2 degree of freedom (df) comparison, then significance of the predictor’s association with ET was based on the 3 df test of the full spline model; otherwise, the single df linear regression Wald test was used. For ordinal predictors, we compared the model using category indicator variables to the pseudo-linear model based on equally-spaced category scores, and based a final test on the latter or a categorical model, depending on whether significant nonlinearity in the equally-spaced scores was found.

Hemodialysis Fistula Maturation (HFM) Study Group

Members of the HFM Study Group are as follows:

Chair, Steering Committee, University of Pennsylvania: H. Feldman; Clinical Centers, Boston University: L. Dember (PI), A. Farber, J. Kaufman, L. Stern, P. LeSage, C. Kivork, D. Soares, M. Malikova; University of Alabama: M. Allon (PI), C. Young, M. Taylor, L. Woodard, K. Mangadi; University of Cincinnati: P. Roy-Chaudhury (PI), R. Munda, T. Lee, R. Alloway, M. El-Khatib, T. Canaan, A. Pflum, L. Thieken, B. Campos-Naciff; University of Florida: T. Huber (PI), S. Berceli, M. Jansen, G. McCaslin, Y. Trahan; University of Texas Southwestern: M. Vazquez (PI), W. Vongpatanasin, I. Davidson, C. Hwang, T. Lightfoot, C. Livingston, A. Valencia, B. Dolmatch, A. Fenves, N. Hawkins; University of Utah: A. Cheung (PI), L. Kraiss, D. Kinikini, G. Treiman, D. Ihnat, M. Sarfati, I. Lavasani, M. Maloney, L. Schlotfeldt; University of Washington: J. Himmelfarb (PI), C. Buchanan, C. Clark, C. Crawford, J. Hamlett, J. Kundzins, L. Manahan, J. Wise; Data Coordinating Center, Cleveland Clinic: G. Beck (PI), J. Gassman, T. Greene, P. Imrey, L. Li, J. Alster, M. Li, J. MacKrell, M. Radeva, B. Weiss, K. Wiggins; Cores: Histology Core, University of Washington: C. Alpers (PI), K. Hudkins, T. Wietecha; Ultrasound Core, University of Alabama at Birmingham: M. Robbin (PI), H. Umphrey, L. Alexander, C. Abts, L. Belt; Vascular Function Core, Boston University: J. Vita (PI), M. Duess, A. Levit; Repositories: NIDDK Biosample Repository, Fisher BioServices: H. Higgins, S. Ke, O. Mandaci, C. Snell; NIDDK DNA Repository, Fred Hutchinson Cancer Research Center: J. Gravley, S. Behnken, R. Mortensen; External Expert Panel: G. Chertow (Chair), A. Besarab, K. Brayman, M. Diener-West, D. Harrison, L. Inker, T. Louis, W. McClellan, J. Rubin; NIDDK: J. Kusek, R. Star.