|Home | About | Journals | Submit | Contact Us | Français|
Ths report describes the change in atherosclerotic renovascular disease (AS-RVD) among hypertensive adults referred for renal duplex sonography (RDS) scan.
From Oct 1993 through July 2008, 20,994 patients had RDS at our center. A total of 434 hypertensive patients with two or more RDS exams without intervention comprised the study cohort. Patient demographics (blood pressures, medications, serum creatinine levels, and data from RDS) were collected. Analyses of longitudinal changes in Doppler scan parameters, blood pressures, and renal function were performed by fitting linear growth-curve models. After confirming the linearity of change in Doppler scan parameters among patients with variable number of studies, estimates of mean slopes were calculated using maximum likelihood techniques. For changes in renal function, quadratic growth curves were required to describe longitudinal change.
A total of 434 subjects (212 men [49%] and 222 women [51%]; mean age, 64.6 ± 12.2 years) provided 1351 studies (mean, 3.2 ± 2.4; range, 2 to 18) for 863 kidneys over a mean follow-up of 34.4 ± 25.1 months. At baseline, 20.6% of kidneys demonstrated hemodynamically significant stenosis. On follow-up, 72 kidneys (9.1%) demonstrated anatomic progression of disease. A total of 54 kidneys (6.9%) progressed to significant stenosis and 18 (2.3%) progressed to occlusion. Controlling for progression of disease, baseline renal artery status demonstrated a strong association with baseline kidney length (P = .0006). Significant annualized change in renal length was observed (cm change/year ± standard error of the mean [SEM]: 0.042 ± 0.011; P = .0002) among both kidneys with and without critical disease at baseline, however, decline in length was significantly greater among kidneys exhibiting progression of renovascular disease (−0.152 ± 0.028 cm/year; comparison of slopes between groups P = .0005). In the absence of progression, the presence or absence of critical renal artery stenosis at baseline did not affect the rate of decline in renal length. Fitted models for the natural log transform of serum creatinine demonstrated a significant increase during follow-up (P < .0001). No association was observed between change in serum creatinine and baseline renovascular disease status, or its progression.
A total of 32% of hypertensive adults referred for RDS demonstrated hemodynamically significant renal artery stenosis. Regardless of the presence or absence of baseline disease, a small percentage of patients demonstrated anatomic progression of AS-RVD. A total of 9.1% demonstrated anatomic progression and 2.3% progressed to occlusion. Although anatomic progression of AS-RVD was associated with an increased rate of decline in renal length, progression did not predict a decline in excretory renal function. Intervention for AS-RVD should be selective and reserved for strict indications.
In July of 2007, the Centers for Medicare and Medicaid Services (CMS) held a public hearing to discuss percutaneous intervention for atherosclerotic renovascular disease (AS-RVD).1 This was prompted by the exponential growth in percutaneous transluminal renal angioplasty (PTRA) during the past 5 years. After extensive public debate, the CMS and invited experts concluded that there were insufficient data to reject or accept PTRA as appropriate treatment for RVD. Rather, a CMS recommendation regarding PTRA for RVD was postponed until results from five large prospective randomized studies were available for analysis.2-6
In part, questions regarding appropriate management of RVD are unanswerable because natural history data for the hypertensive population are incomplete. Based on data derived from angiographic case series and prospective studies of select hypertensive patients, intervention is often undertaken for anatomic RVD whenever discovered.7,8 Inevitable progression of RVD with associated decline in kidney size and function among hypertensive patients is cited as rationale for this treatment. However, in a recent prospective study of RVD among Cardiovascular Health Study (CHS) participants, providing a mean follow-up of 8 years, the annualized rate of anatomic progression of prevalent RVD was only 1.3% per year.9 Incident RVD on follow-up, but not prevalent disease at first examination, demonstrated significant and independent associations with increased diastolic blood pressure and decreased renal length. Importantly, no CHS participant demonstrated severe hypertension or renal insufficiency consistent with a renovascular etiology.
This retrospective study was undertaken to examine incident RVD and progression of RVD among hypertensive patients referred for renal duplex sonography scan. Specific aims included determination of: (1) the prevalence of RVD among a referred hypertensive adult population; (2) the rate of incident disease and progression of prevalent disease on follow-up; and (3) the change in kidney size and kidney function in the presence or absence of RVD and its anatomic progression.
From Oct 1993 through Jul 2008, 20,994 patients had renal duplex sonography (RDS) scans at our institution. With Institutional Review Board approval, potential subjects were identified by electronic medical record review of patients referred to the Clinical Vascular Laboratory at Wake Forest University Baptist Medical Center. All patients with two or more RDS exams without intervention for renal artery (RA) revascularization were included in the initial sample. The most common reason for referral was hypertension (97%). Four hundred thirty-four hypertensive patients with two or more RDS exams without intervention for AS-RVD comprised the study cohort. Patient demographics, blood pressures, presence of atherosclerotic risk factors, medications, serum creatinine levels, and data from RDS were collected. Estimated glomerular filtration rate (eGFR) was calculated using the Modification of Diet in Renal Disease equation.10
The technique of RDS has been described in detail.9,11-13 After an overnight fast, the patient was placed in the supine position and a 2.25 or 3.0 MHz ultrasound scan probe was coupled to the abdominal skin with acoustic gel 3 or 4 cm inferior to the xiphoid process. Sagittal B-mode scan images were obtained of the upper abdominal aorta, celiac axis, and superior mesenteric arteries.
After a sagittal aortic and superior mesenteric artery signal was obtained, the probe was rotated 90° to obtain a B-mode scan image of the aorta and proximal superior mesenteric artery in cross-section. The left renal vein was identified in longitudinal section. Using the left renal vein as a reference, the aortic origins of the main renal arteries were identified. While maintaining an angle of insonation of <60°, Doppler scan samples were taken from each RA from aortic origin to the renal hilum, for a total of approximately 10 Doppler scan sample sites per RA.
Renal artery peak systolic velocity (RA-PSV) and enddiastolic velocity were estimated from the spectral analysis of the Doppler scan-shifted signals. After Doppler scan interrogation in the supine position, RA-PSV was estimated from a flank approach with the participant in the right or left lateral decubitus positions. B-mode scan imaging of each kidney determined the greatest pole-to-pole renal length. The RDS study was considered negative or positive for significant RVD or inadequate for interpretation according to the following criteria: (1) RDS was negative for significant RVD when RA-PSV from aortic origin to renal hilum was <1.8 m/second; (2) RDS was positive for hemodynamically significant RVD when there was a focal increase in RA-PSV ≥1.8 m/second (≥60% RA diameter-reducing stenosis) in the proximal RA or no Doppler scan signal was obtained from an imaged artery (RA occlusion); and (3) RDS was technically inadequate for interpretation when RA-PSV could not be determined from the aortic origin to the renal hilum.
Prevalent RVD was present at the first RDS. Incident disease was defined as new, hemodynamically significant stenosis or occlusion on follow-up RDS. RAs without prevalent RVD were considered to have categorical progression of disease if RA-PSV demonstrated an increase to critical stenosis or RA occlusion. Significant increase in PSV was examined by comparing annualized change in PSV to two times the standard deviation (SD) of the linear slope from a regression of PSV vs follow-up time in years, as described in the statistical analysis.
A total of four ultrasonographers, all registered vascular technologists (RVTs), participated in the ultrasonographic scan evaluation of these patients over the time period defined. It is estimated that two of these four RVTs performed greater than 65% of all examinations. All examinations were conducted with one of two ultrasound scan machines (Philips IU22 and Philips HDI5000; Philips Healthcare, Andover, Mass).
Descriptive statistics were computed for the entire cohort as a group and stratified by presence or absence of AS-RVD (in either kidney) at the first RDS exam. Continuous characteristics were compared between AS-RVD groups using t tests, and χ2 tests (or Fisher's exact tests) were used for categorical factors. Continuous variables were expressed as mean ± SD; categorical variables were expressed as frequency and percentage of total (n[%]).
Analyses of longitudinal changes in Doppler scan parameters and renal function were performed by fitting random effects linear growth-curve models. After confirming the linearity of change in Doppler scan parameters among patients with variable numbers of studies, estimates of mean slopes were calculated using restricted maximum-likelihood.14 For changes in renal function, quadratic growth curves were required to describe longitudinal change.
Details of statistical methods for calculating the variability of PSV slopes and statistical models employed are included in the Statistical Appendix (online only). Briefly, significant increase of PSV was determined independently by estimating the variability of linear change in PSV in the subset of patients with three or more RDS exams. Progression was defined as an increase in RA-PSV above what might be expected due to variability of the measure. Variance components were estimated using side-specific models of linear change in RA-PSV fitted against time (in years) from initial scan.15
At each follow-up RDS exam, individual renal arteries were classified as having significant change in RA disease from the most immediate prior exam if the time-adjusted change in velocity was greater than or equal to the signifi-cant change in PSV (ie, two times the inherent variability). For patients with only two exams, these values were 7 cm/seconds/year in the right RA and 13 cm/seconds/ year in the left. For patients with three or more RDS exams, a jackknife technique was employed where observations from individual patients were excluded from the variance estimation used to classify them. Classification of each RDS result as “progression” or “nonprogression” yielded individual patient response profiles. Evaluation of risk factors for PSV progression was performed using logistic regression models with generalized estimating equations and robust standard errors to account for within-subject correlation.16
All analyses were performed using SAS statistical software (SAS/STAT v 9.2, SAS Institute Inc, Cary, NC).
Between Oct 1993 and Jul 2008, 20,994 patients had RDS at our center. A total of 434 hypertensive subjects provided 1351 RDS studies for 863 kidneys over a mean follow-up of 34.4 ± 25.1 months. Table I summarizes the demographic, renal function and atherosclerotic risk factor characteristics of all patients at the time of their baseline examination, and for patients with and without significant RA disease at baseline. Patients had a mean age of 64.6 ± 12.2 years, 51.2% were female, 48.8% male, and 26.3% were African American. The mean systolic blood pressure at the time of the baseline scan was 158.0 ± 29.6 mm Hg, mean diastolic pressure was 87.4 ± 16.6 mm Hg. All but 4 of the 434 patients included in the cohort had hypertension noted in their medical history, were on anti-hypertensive medication, or were hypertensive (ie, had blood pressure >140/ 90) at the time of one or more duplex scan exams. Of the remaining 4 patients, 2 had elevated serum creatinine levels. The remaining 2 patients were referred for reasons that were uncertain. Patients were taking a mean of 2.6 ± 1.3 blood pressure medications at the time of the first scan. The mean baseline serum creatinine was 1.4 ± 0.8 mg/dL, and 90 patients (24.5%) had a serum creatinine ≥1.8. A history of diabetes was noted in 176 patients (43.1%), hyperlipidemia in 49 (11.3%), and current or former tobacco use in 270 (69.6%). Two hundred nine patients (51.5%) were taking lipid-lowering agents at the time of their first scan.
Significant RVD was present at the first examination in 20.6% of kidneys and 32% of patients. Demographic, renal function, and atherosclerotic risk factor characteristics of these patients, segregated by presence or absence of baseline RVD (in one or both kidneys) are presented in Table I. Patients with anatomic RVD at baseline were older (68.9 ± 8.9 vs 62.6 ± 12.9 years; P < .001), less frequently black (14% vs 32.2%; P < .001), had lower diastolic blood pressure (82.9 ± 13.0 vs 89.9 ± 17.8 mm Hg; P < .001), had smaller kidney lengths (10.5 ± 1.3 vs 10.9 ± 1.3 cm; P = .006), more prevalent renal insufficiency (31.6% vs 20.3%; P = .019), more prevalent peripheral vascular disease (PVD) (34.1% vs 15.2%; P < .001), and more prevalent history of smoking (81% vs 65%; P = .002). Patients with baseline disease were also taking fewer blood pressure medications at baseline (2.4 ± 1.3 vs 2.7 ± 1.4; P = .069). At the time of their baseline exam, 113 patients (30%) had a resistive index (RI) ≥0.8. Resistive indices were not significantly different between those with and without baseline renovascular disease (0.8 ± 0.1 vs 0.7 ± 0.1; P = .847).
On follow-up, 72 kidneys (9.1%) demonstrated categorical progression of renovascular disease. Fifty-four kidneys (6.9%) progressed to significant stenosis and 18 (2.3%) progressed to RA occlusion.
Significant progression based on increased PSV was more common than categorical disease progression. Results of a random-effects logistic regression model to estimate follow-up PSV progression are presented in Table II and Fig 1. Table II shows model-based estimates of 5-year RVD progression for varying values of covariates found to be significantly associated with progression. Factors associated with an increase in PSV progression included: presence of RVD in the contralateral RA at baseline exam (P = .010), patient history of hyperlipidemia (P = .010), and non-black race (P = .064). There was an accelerated rate of increase in PSV over time in kidneys with the presence of RVD at baseline. This effect is illustrated in Fig 1.
Controlling for categorical progression of disease, baseline RA status demonstrated a strong association with baseline kidney length (P = .0006). Significant annualized change in renal length was observed (cm change/year ± standard error of the mean [SEM]: 0.042 ± 0.011; P < .001) among both kidneys with and without critical renovascular disease at baseline, however, decline in length was significantly greater among kidneys exhibiting categorical progression of RVD (−0.152 ± 0.028 cm/year; comparison of slopes between groups P < .001). In the absence of progression, the presence or absence of critical RA stenosis at baseline did not affect the rate of decline in renal length. These relationships are depicted graphically in Fig 2. Progression of PSV was not associated with baseline renal lengths or change in renal length over follow-up.
Relationships between predictive factors and decline in renal function as measured by eGFR are presented in Figs 3 and and4.4. Predicted mean eGFR for patients with and without baseline RVD, and with and without PSV progression, are shown in Fig 3. Patients with unilateral (but not bilateral) RVD at baseline exam had significantly lower eGFR, however, patients with bilateral PSV progression had higher eGFR at initial renal function assessment than did those patients without PSV progression. Neither baseline RVD nor PSV progression was significantly associated with a change in the rate of renal function decline.
Diabetes and hyperlipidemia demonstrated significant and independent association with poorer renal function. A history of hyperlipidemia accelerated the projected decline in eGFR, however, the effect of hyperlipidemia on decline in eGFR moderated over the observation period. Diabetes was also associated with greater renal function decline, and the effect of diabetes on eGFR decline appeared to be sustained throughout the period of renal function observation. Effects of hyperlipidemia and diabetes are depicted graphically in Fig 4.
On mean follow-up of 3 years of AS-RVD in hypertensive patients, we observed that 2.3% of significant RVD progressed to RA occlusion and 9.1% of kidneys demonstrated new, incident RVD. In the absence of categorical progression, the presence or absence of critical RA stenosis at baseline did not affect the rate of decline in renal length. Although eGFR demonstrated a significant decrease during follow-up, this decrease had no association with the presence or absence of baseline renovascular disease or its progression. These results and the results from other reports call into question the practice of RA intervention for anatomic RVD in the absence of strict clinical indications.9
Pillay et al17 described the changes in blood pressure and serum creatinine among 98 patients with RVD. In this multi-center, non-randomized, observational study, RA stenosis was defined by aortography. Complete data were available for 85 patients with a minimum follow-up of 2 years. Seventy-six patients were managed medically (64 patients with unilateral RVD; 12 with bilateral RVD) while 12 patients with bilateral disease had RA intervention. On follow-up, the 2-year estimated mortality was 32%, and was equivalent among patients treated with intervention and medical management. These authors observed no change in median blood pressure, number of anti-hypertensives, or renal size among survivors. A small but statistically signifi-cant increase in serum creatinine was observed in patients who had intervention, however, RVD treated medically had stable serum creatinine over the follow-up period.
Although our study suffers from limitations inherent to retrospective design, there are prospective studies of RVD that utilize renal duplex sonography scan.18-20 Caps et al21 described 170 patients and 295 kidneys with 5-year follow-up. In part, disease progression was defined as a 100 centimeter per second increase in RA-PSV or progression to RA occlusion. These authors described disease progression in 31% of renal arteries while 3% progressed to occlusion. Our retrospective study describes a similar rate of progression, however, in our patient group, a significant rate of change in PSV was estimated at 21 to 39 centimeters per second at 3 years. The annualized rate of progression to RA occlusion was less than half that observed by Caps et al.
Previous reports from our group have examined RVD in a local cohort of participants in the Cardiovascular Health Study (CHS). The CHS is a multi-center, longitudinal study of cardiovascular risk factors, mortality, and morbidity among free-living, community-dwelling, elderly Americans.22 Cross-sectional examination of RVD among this cohort demonstrated prevalent disease in 6.8% at baseline.11 The presence of baseline disease demonstrated strong associations with blood pressure and renal function.23 However, the relationship between RVD and increased serum creatinine was dependent on a significant interaction with increased blood pressure.23 On mean follow-up of 14 months, RVD demonstrated significant association with adverse cardiovascular events, however, this relationship was independent of increased blood pressure and/or serum creatinine.24 Upon second renal duplex scan at a mean interval of 8 years after the first study, a significant change in RVD was demonstrated in only 14% of kidneys (annualized rate, 1.3% per year).9 In addition to a very low rate of progression for RVD, these results suggested that in the absence of renovascular hypertension or ischemic nephropathy, RVD might be considered analogous to the relationship between infra-inguinal occlusive disease and increased risk of adverse coronary events. In this setting, intervention for neither RVD nor infra-inguinal disease could be expected to improve survival free of adverse cardiovascular events.
The absence of data defining increased dialysis-free survival after percutaneous intervention in combination with the exponential growth of this intervention for RVD prompted the CMS to sponsor a public discussion of these issues.1 In the course of discussion, one speaker questioned the value of open repair of RVD compared with PTRA.1 Although the increased mortality and morbidity associated with open operative repair compared with PTRA is widely recognized, preoperative features and results from open operative repair that have demonstrated increased dialysis-free survival might logically apply to both therapies. In this regard, our operative experience in 500 consecutive athero-sclerotic patients demonstrated that severe hypertension, complete repair of global renovascular disease including RA occlusion, and rapid deterioration in preoperative renal function were associated with the best response to operation.25 Blood pressure cure and early incremental increase in excretory function demonstrated significant and independent associations with survival free from dialysis dependence.26
Renal function response to operation has demonstrated significant and independent association with dialysis-free survival.26 Incremental increase in renal function after RA intervention establishes a proved relationship between the two (ie, ischemic nephropathy). In the absence of increased renal function, patients with preoperative serum creatinine greater than 1.8 mg/dL demonstrated no change in declining renal function after surgery, and increased risk of dialysis and death.25-27 Whether PTRA improves dialysis-free survival or demonstrates similar strong relationships between function improvement and clinical outcome is unknown. To date, meta-analyses of data from three small, randomized trials comparing PTRA and medical management in hypertensive patients failed to demonstrate a treatment benefit on renal function.28
The influence of prevalent RVD on disease progression deserves special comment. In this report, increased rate of progression was observed in kidneys with RVD, and in contralateral nondiseased kidneys in patients with unilateral RA stenosis. This association has been reported by others and is consistent with the most common pathology of atherosclerotic RVD.19,21 The majority of lesions are ostial in nature; ie, the RA stenosis reflects aortic atherosclerosis that “spills over” into the RA ostium. Given the shared nature of the aortic atheroma between RAs, the presence of RVD in one could be expected to favor progression in the contralateral vessel.
This retrospective study suffers from a number of important limitations. The study cohort represents 2% of all patients receiving RDS during this time interval. Subjects were selected by the presence of multiple RDS studies in the absence of interval intervention. The indications for referral and re-referral could not be distinguished with certainty, nor can the impact of this referral bias on these results. Moreover, the number and time intervals for renal duplex sonography scan varied widely. In addition, sonographers and duplex sonography scan were subject to change during the study period. Although 65% of all studies were performed by two of four sonographers, estimates of repeatability between and among sonographers are not available. The variability inherent to multiple sonographers and changing technology over time may have been partially offset, since each sonographer had access to the prior categorical duplex scan results. Issues regarding reproducibility of RDS results have an important impact on estimates of RA disease progression. Besides categorical changes in RA stenosis and occlusion, progression of renovascular disease was defined by an increase in RA peak systolic velocity greater than twice the SD of predicted change. Predicted change in peak systolic velocity was estimated using subjects with three or more duplex scan studies. Compared to patients with three or more studies, patients with two duplex scan studies demonstrated differences in ethnicity, serum creatinine, and prevalent diabetes mellitus (Appendix 1, online only). Patients with two studies only were more commonly African American, had increased serum creatinine, and greater prevalence of diabetes.
Regarding comparability of our hypertensive cohort to the general hypertensive population, blood pressure control among patients included in the cohort and treated with anti-hypertensive medications was poor (16% with systolic blood pressure/diastolic blood pressure [SBP/DBP] <140/90) compared to that of the general population of the United States hypertensive adults on treatment (47% with SBP/DBP <140/90).29 The relative severity and poor blood pressure control of patients referred for duplex sonography scan is probably reflective of the perceived need by referring physicians to find and treat a renovascular source of hypertension. The fact that everyone in this cohort was seen at least twice (even those without evidence of disease at first RDS exam) is probably consistent with hypertension that was unresponsive or difficult to control.
As highlighted at the CMS public hearing, central to the question of RVD progression and its management is whether cessation of RA intervention as currently practiced would favor dialysis-free survival in our country while eliminating the economic burden associated with intervention. The CMS is sensitive to this question, and they should be.1 Almost certainly, the results from ongoing trials will be used to determine CMS reimbursement for RVD intervention in the future.
This is another excellent study from the Wake Forest group that is clinically helpful in the management of patients with renovascular disease and hypertension. It adds another nail in the coffin of prophylactic or “drive-by” renal artery angioplasty.
The 9% incidence of progression may actually be an overestimate, given the threshold criteria of a 21 to 39 cm/second increase over 3 years. Despite the statistical analysis to judge the variance of this parameter, I find these criteria to be well within the variability of repeated testing. This is especially so given the turnover in techs, equipment, and the inherent difficulty of this non-invasive exam. So, my first question is do you have robust measures of variability of the peak systolic velocity with repeated measurements in patients with and without disease over time?
My second question relates to the value of follow-up renal duplex sonography scan. If it doesn't really help in predicting treatable events, is it ever worthwhile to follow patients with renovascular disease with repeated exams? If so, who do you select for follow-up sonography scan?
My final question involves selecting patients for intervention. Despite, the increase in morbidity and mortality, the Wake Forest group and others have documented that surgical renal revascularization in selected patients can lead to improvement in blood pressure, renal function, and dialysis-free survival. Is this also true for renal percutaneous transluminal renal angioplasty (PTRA)? I must say, the data to support this are weak. Among those with documented renovascular disease, who do you select for renal PTRA?
We acknowledge that the utilization of repeated measurements of peak systolic velocity (PSV) to construct a regression model to define progression was not particularly robust. These patients had widely varying numbers of duplex scan examinations over the follow-up time, and our model could well be predominantly influenced by those few patients with multiple scans that did not demonstrate significant progression.
In regards to the second question, we do feel that it is valuable to follow patients with duplex scans and continue to do so in our practice. We recognize that there are few data to support such routine use of follow-up renal duplex sonography scans, however, we do feel that its use, in conjunction with attention to renal function and strict blood pressure control, does enable us to predict that subset of patients with specific clinical indications for revascularization. We do recognize that our particular clinical interest in this disease process may influence our frequency of this follow-up.
Regarding the third question, unfortunately, the best medical therapy vs best medical therapy plus percutaneous intervention vs best medical therapy plus surgical revascularization just have not been compared in a prospective randomized fashion and without these data it is difficult to know whether the difference in outcome is due to differences in method of intervention, or differences in patient selection, or a combination of those factors. Certainly, the most recent publication of preliminary results from the Angioplasty and STent for Renal Artery Lesions (ASTRAL trial) show no difference in renal function, blood pressure control, or adverse cardiovascular events for percutaneous intervention and it has been difficult to demonstrate any improvement in dialysis-free survival for percutaneous intervention vs surgery. Matt Corriere of our group is going to give a more in-depth discussion of our approach to percutaneous intervention and some of our recent outcomes later this morning, so I will defer a discussion of our selection criteria and utilization of therapy to his comments later in this morning's program.
|3 or more scans||2 scans|
|Age at first scan||170||65.0 ± 12.3||260||64.3 ± 12.1||.558|
|African American||173||33 (19.1)||261||81 (31.0)||.006|
|Female||173||91 (52.6)||261||131 (50.2)||.623|
|Systolic blood pressure||141||160.7 ± 28.3||170||155.8 ± 31.2||.151|
|Diastolic blood pressure||131||87.6 ± 17.5||168||87.2 ± 15.9||.812|
|Mean arterial blood pressure||131||112.0 ± 18.5||168||109.8 ± 18.2||.300|
|History of hypertension||171||166 (97.1)||249||241 (96.8)||.867|
|Serum creatinine||147||1.3 ± 0.6||220||1.6 ± 0.8||<.001|
|MDRD-GFR (mL/min/1.73m2||146||66.7 ± 30.2||219||58.5 ± 29.2||.010|
|Maximum RA-PSV (m/sec)||169||1.8 ± 0.9||256||1.4 ± 0.7||<.001|
|Resistive index (side w/max PSV)||144||0.7 ± 0.1||233||0.8 ± 0.1||.611|
|Kidney length (side w/max PSV)||168||10.7 ± 1.3||257||10.8 ± 1.3||.767|
|Renal insufficiency (Cr ≥1.8)||147||28 (19.0)||220||62 (28.2)||.046|
|Resistive index ≥1.8||144||37 (25.7)||233||76 (32.6)||.154|
|History of TIA/CVA||162||64 (39.5)||240||90 (37.5)||.685|
|History of diabetes||165||60 (36.4)||243||116 (47.7)||.023|
|History of CAD||166||109 (65.7)||243||173 (71.2)||.235|
|On lipid-lowering agent||163||90 (55.2)||243||119 (49.0)||.217|
|History of PVD||164||37 (22.6)||241||47 (19.5)||.456|
|Current/former smoker||158||113 (71.5)||230||157 (68.3)||.493|
MDRD-GFR, Modification of diet in renal disease-glomerular filtration rate; RA-PSV, renal artery peak systolic velocity; TIA/CVA, transient ischemic attack/cerebrovascular accident; CAD, coronary artery disease; PVD, peripheral vascular disease.
The model chosen to estimate both linear change in PSV and variability of that change is the random coefficients regression model of Laird and Ware.1 The model set-up for linear growth is as follows:2 let yi = (yi1, yi2, . . . , yip)T denote the vector of observed PSVs for the i-th subject (note that the dimension p may vary from subject to subject); β= (β0, β1)T be a vector of fixed regression effects; bi = (bi0, bi1)T be a vector of random regression effects for the i-th subject; ei = (ei1, ei2, . . . , eip)T be a vector of residuals for the i-th subject; and Xi and Zi pi×2 design matrices for the fixed and random effects, respectively. The model is then
with the following specifications for the random terms:
Here E(·) denotes statistical expectation, V(·) denotes variance, Cov(·) denotes covariance, and Ipi is an identity matrix of dimension pi. We assume that both ei and bi are normally distributed (ie, that the distribution of yi is where . In this model, we have
a 2 × 2 covariance matrix for the random regression lines where is the variance of the intercept, σb0,b1 is the covariance between the intercept and slope, and is the variance of the slope and the parameter of interest for classifying significant change. This model can be easily generalized to include higher-order terms for both random and fixed effects to evaluate curvilinear relationship with time (ie, quadratic growth).
Empirical investigation of the distribution of PSV change was performed in a multi-step process involving: (1) evaluation of the statistical distribution of PSV for normality; (2) examination of longitudinal change for non-linearity and uniformity of slopes over the various lengths of follow-up; (3) estimation of the distribution of PSV slopes; and (4) definition of “significant change” and evaluation of individual subjects and time points to categorize PSV values as having significantly changed from the previous observation.
First, the distribution of individual PSVs was examined to assess normality. To do this, individual measurements were divided into groups based on time in years from index exam. Stem-leaf plots of the PSVs (ignoring within-subject correlation) were examined. As expected, the empirical distributions tended to be positively skewed, however, each distribution was mound shaped, uni-modal, and though skewed, the skewness did not appear to be so severe as to make abandonment of the normality imperative, particularly when weighed against the desire to preserve measurements in their original units.
After assessing the appropriateness of the normality assumption, data were examined to assess both linearity and uniformity of slopes over the varying follow-up times (ie, whether patients with longer vs shorter follow-up had different slopes over time). Linearity was assessed by regressing PSV vs time from initial scan (in years) and fitting a second-degree polynomial for time in both the fixed and random effects parts of the linear model. Due to estimation problems when fitting higher order random coefficients models to data pooled across both kidneys, separate models were fitted for each side. There was no evidence of signifi-cant curvature in the lines fitted to either side (P = .371 for quadratic term when fitted to left side data; P = .631 when fitted to the right side).
Another important assumption of an analysis pooling data across all patients is that the linear slope is the same regardless of the length of follow-up (ie, that a single slope model would be appropriate). The extreme irregularity of measurements, the large individual differences in time course examined, and the number of individual data points made graphical examination of all points combined impractical; therefore, patients were divided into groups with varying lengths of follow-up. Fig A1,, online only, shows estimated mean slopes and associated 95% confidence intervals (CIs) after dividing patients into six groups based on the following categorization of follow-up time: ≤ 1.5 years, 1.5 to 2.5 years, 2.5 to 3.5 years, 3.5 to 4.5 years, 4.5 to 7.5 years, and >7.5 years. Estimated mean slopes are shown for each side (left and right). There are considerable differences in the variability of the slope estimates, however, there is little evidence of a systematic change in slopes among patients with longer follow-up times. Note that the 95% CIs for each slope includes the null value (zero).
The final challenge in estimating kidney-specific slopes is how to deal with multiple kidneys in individual patients. Since each renal duplex sonography (RDS) exam typically includes PSV estimates from renal arteries (RAs) on both the right and left sides, ignoring this second level of clustering when pooling data from both sides potentially ignores an important source of correlation, as well as ignoring the potential for differential (systematic) variability between the two sides. Therefore, an expanded model was used where individual growth curves were fitted for each side.3 The variance-covariance matrix for the random effects under this model is
where the parameters of interest are (R) and (L), the estimated variances of the slopes for the right and left side PSVs, respectively. Note that the variance-covariance matrix includes the parameters σb0(R, L), σb0,b1(L, R), σb0,b1(R, L), and σb1(R, L) to estimate the covariance between fitted lines from opposite sides within the same patient. A likelihood-ratio test comparing models A2 and A3 was performed and found to be highly significant (P < .001), indicating that the additional variance-covariance parameters included in model A3 are important.
To calculate the side-specific variance estimates, data from all patients with three or more observations were pooled. Patients with only two observations were omitted from variance calculations since their growth curves did not provide any information on within-subject variability. Significant change in PSV between any two measurements was defined as any annualized difference between two points greater than two times the estimated standard error of the side-specific regression slope. In patients with only two RDS exams, side-specific variability estimates from all patients with three or more exams were used when assessing significant change. In patients with three or more RDS exams, a jackknife approach4 was used where the patient to be classified was removed from analysis, the variance components re-estimated, and significant change in PSV for the removed subject then classified using the updated estimates.
Individual follow-up PSV values (ie, all PSVs subsequent to the first RDS) were classified as exhibiting significant progression (assigned a value of “1”) or no progression (assigned a value of “0”) by comparing the current PSV to the most recent prior value. In subjects with more than two observations, if a renal artery exhibiting significant progression at a prior time point had a current PSV value that was significantly smaller (ie, a negative difference that was greater than two standard errors of the mean slope in magnitude), the response at that time point was set back to “0” indicating no progression. In this way, response profiles of zeros and ones were created for each renal artery with serial exams. Furthermore, these vectors of responses allowed for the potential “regression” of previously observed PSV increase which is consistent with the observed variability in the technique.
Risk factors for PSV progression (ie, the profiles of repeated assessment of PSV progression) were examined using generalized estimation equations.5,6 In this model, the usual logistic link to characterize a dichotomous outcome was employed. Let πt be the underlying probability of observed progression at time t, the model fitted was
where X and β are the usual design matrix and vector of (fixed) regression parameters, respectively. Clustering (ie, correlation) between response profiles on the same patient was accounted for using “R-side” random effects.7 The empirical sandwich estimator of the variance of β was used because of its robustness to misspecification of the true underlying correlation.
Competition of interest: none.
Presented at the Thirty-third Annual Meeting of the Southern Association for Vascular Surgery, Jan 15, 2009.
Additional material for this article may be found online at www.jvascsurg.org.