- Email: [email protected]
Use of standardized ratios to examine variability in hemodialysis vascular access across facilities
Use of Standardized Ratios to Examine Variability in Hemodialysis Vascular Access Across Facilities Ashwini R. Sehgal, MD, Marcia R. Silver, MD, Kenneth E. Covinsky, MD, Richard Coffin, and Jeannette A. Cain, MSM, for the Medical Review Board of The Renal Network, Inc ● The type of hemodialysis vascular access (catheter, fistula, graft) is an important determinant of patient morbidity and dialysis efficiency. The relative importance of patient versus provider factors in determining type of vascular access is unclear. We sought to develop a quality improvement tool that adjusts for differences in patient characteristics, thereby allowing examination of provider-related variability in types of vascular access used across facilities. We examined 15,339 patients from 216 chronic hemodialysis units in Indiana, Kentucky, Ohio, and Illinois and found that 20% of patients had catheters, 24% had fistulas, and 56% had grafts. Young, male, and white patients were more likely to have fistulas, whereas old, female, and black patients were more likely to have grafts. Diabetics were more likely to have catheters and less likely to have fistulas. New patients were more likely to have catheters and less likely to have grafts. A facility specific standardized catheter ratio (SCR), standardized fistula ratio (SFR), and standardized graft ratio (SGR) were calculated based on the actual number of patients with each type of vascular access divided by the expected number adjusted for patient characteristics. Facility SCRs ranged from 0.00 to 2.87. Of the 216 facilities, 38 (18%) had an SCR significantly less than 1.00, and 32 (15%) had an SCR significantly greater than 1.00. Similar variability was observed in SFRs and SGRs. In conclusion, the type of vascular access varies greatly across facilities. Use of standardized access ratios adjusted for patient characteristics may help providers examine processes of care that contribute to variability in access use. Analogous to the standardized mortality ratio, the SCR, SFR, and SGR should be effective quality improvement tools. 娀 2000 by the National Kidney Foundation, Inc. INDEX WORDS: Hemodialysis; vascular access; fistula; graft; catheter.
Appendix appears only on the web site (www.ajkd.org) electronic pages
T
HE TYPE OF hemodialysis vascular access (catheter, fistula, or graft) strongly influences several important patient outcomes. For example, fistulas are typically associated with (1) increased access longevity, (2) better dialysis adequacy, (3) fewer infectious or thrombotic complications, and (4) lower health care costs. By contrast, catheters are associated with poor outcomes in all four areas, whereas grafts tend to have intermediate results.1-9 Despite the clear advantages of fistulas, only a minority of patients have fistulas. Moreover, catheter and graft use appears to be increasing. There are also striking geographic differences in access type as exemplified by a 13-fold variation in use of grafts versus fistulas across nine census regions.10 The relative importance of patient versus provider factors in determining variability of vascular access type is unclear. We sought to develop a quality improvement tool that adjusts for differences in patient characteristics, thereby allowing examination of provider-related variabil-
ity in vascular access type. As models for this effort, we selected the standardized mortality ratio (SMR) and the standardized hospitalization ratio (SHR), two widely used measures of provider-related variability across dialysis facilities.11-13
From the Division of Nephrology, MetroHealth Medical Center, Cleveland, OH; the Department of Medicine, Case Western Reserve University, Cleveland, OH; the Center for Biomedical Ethics, Case Western Reserve University, Cleveland, OH; the Department of Epidemiology and Biostatistics, Case Western Reserve University, Cleveland, OH; The Renal Network, Inc, Indianapolis, IN; the Divisions of Geriatrics and General Medicine, Veterans Administration Medical Center, San Francisco, CA; and the Department of Medicine, University of California, San Francisco, CA. Received April 29, 1999; accepted in revised form September 10, 1999. Supported by grant number DK51472 from the National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, and contracts 500-97-E027 and 500-97-E028 from the Health Care Financing Administration, Baltimore, MD. Address reprint requests to Ashwini R. Sehgal, MD, Division of Nephrology, 2500 MetroHealth Dr, MetroHealth Medical Center, Cleveland, OH 44109-1998. E-mail: [email protected]
娀 2000 by the National Kidney Foundation, Inc. 0272-6386/00/3502-0013$3.00/0
American Journal of Kidney Diseases, Vol 35, No 2 (February), 2000: pp 275-281
275
276
SEHGAL ET AL
METHODS
RESULTS
Subjects and Facilities
Subject Characteristics
Patients age 18 years or older on chronic hemodialysis for at least 3 months were eligible for inclusion. Our facility sample consisted of all chronic hemodialysis facilities in Illinois, Indiana, Ohio, and Kentucky that had at least 30 eligible patients.
All 307 chronic hemodialysis facilities in Illinois, Indiana, Ohio, and Kentucky reported complete information on a total of 16,832 eligible patients. We excluded 1,403 patients from 91 small facilities, leaving 15,429 subjects from 216 facilities. Of these, 90 patients had more than one access reported (eg, if blood was drawn from a maturing fistula and returned to a catheter). Our remaining analyses focus on the 15,339 subjects with one access. About half the subjects were male, about half were white, and 37% had endstage renal disease (ESRD) secondary to diabetes (Table 1). The mean duration of ESRD was 3.2 years, with a range from 3 months to 27 years.
Data Collection As part of a quality improvement project of End Stage Renal Disease Networks 9 and 10, all facilities reported the type of vascular access used for the first hemodialysis treatment received by each patient during the week of December 25 to 31, 1997. We obtained patient demographic (age, sex, race, height, weight) and medical characteristics (cause of renal failure, time on dialysis) from the Network database for all subjects. Because of a change in reporting requirements in 1995, we were also able to obtain information on select comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug dependence) for all patients beginning dialysis in the last 3 years.
Statistical Analysis We used two different units of analysis: first patients and then facilities. At a patient level, we used the chi-squared test or analysis of variance to examine the univariate relationship between actual catheter use and the patient demographic and medical characteristics listed above. We used logistic regression to examine the multivariate relationship between actual catheter use and all patient characteristics. To account for possible nonlinear relationships, continuous predictor variables were entered as a series of indicator variables (eg, five categories for age). We then used the odds ratios from this regression analysis to calculate the expected probability of catheter use (as opposed to actual catheter use) for each patient, taking into account the patient’s demographic and medical characteristics. At a facility level, we summed the expected probability of catheter use of all patients at a particular facility to determine the total expected number of patients with catheters. This expected number takes into account the demographic and medical characteristics of the patients at each facility. We then calculated the standardized catheter ratio (SCR) as the actual number of patients with catheters divided by the expected number of patients with catheters. We used the chi-squared test to determine whether the SCR was significantly different from 1.00.12 We calculated the standardized fistula ratios (SFR) and standardized graft ratios (SGR) in a similar manner. We also performed several secondary analyses. First, we calculated standardized ratios for patients from small facilities in aggregate. Next, we determined the impact of additional comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug use) by calculating standardized ratios on the subset of patients for whom this information was available.
Vascular Access Type Of all subjects, 20% had catheters, 24% had fistulas, and 56% had grafts. Among the 3,128 subjects with catheters, 76% had tunneled, cuffed catheters and 24% had temporary catheters. Reported reasons for catheter use were: no vascular sites (37%), no graft/fistula created (25%), temporary interruption in graft/fistula (13%), graft/ fistula maturing (13%), other specific reason (7%), and no reason reported (5%). The proportion of patients with catheters ranged from 0% to 63% (mean, 20%) across the 216 facilities. In 33 (15%) facilities, the proportion of patients with catheters was significantly less than 20%; in 146 (68%) facilities, the proportion was not significantly different from 20%; and in 37 (17%), the proportion was significantly Table 1. Subject Characteristics (n ⴝ 15,339 Patients) Mean age, years (range) Men (%) Race (%) White Black Other Mean height, centimeters (range) Mean weight, kilograms (range) Cause of renal failure (%) Diabetes Hypertension Glomerulonephritis Other Mean time on dialysis, years (range)
61 (18-98) 52 53 44 3 169 (137-213) 74 (32-181) 37 29 15 19 3.2 (0.25-27)
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
greater than 20%. The proportion of patients with fistulas ranged from 2% to 62% (mean, 24%) across facilities. In 28 (13%) facilities, the proportion of patients with fistulas was significantly less than 24%; in 161 (75%) facilities, the proportion was not significantly different from 24%; and in 27 (13%), the proportion was significantly greater than 24%. The proportion of patients with grafts ranged from 16% to 91% (mean, 56%). In 27 (13%) facilities, the proportion of patients with grafts was significantly less than 56%; in 171 (79%) facilities, the proportion was not significantly different from 56%; and in 18 (8%), the proportion was significantly greater than 56%. Correlates of Vascular Access Type On univariate analysis (Table 2), catheter use was most strikingly associated with short height, low weight, and a short time since ESRD developed. For example, 36% of patients with ESRD for 3 to 5.9 months had catheters, compared with 13% of patients with ESRD for 7 or more years. Fistula use was most strikingly associated with younger age, male gender, large height, nondiabetic causes of renal failure, and a long time since ESRD developed. For example, 32% of men had fistulas compared with 14% of women. On multivariate analysis (Table 3), young patients, men, and whites were more likely to have fistulas, whereas old patients, women, and blacks were more likely to have grafts. Heavy patients were more likely to have grafts, and lighter patients were more likely to have catheters. Diabetics were more likely to have catheters and less likely to have fistulas. New patients were more likely to have catheters and less likely to have grafts. The c statistic for the catheter, fistula, and graft analyses were 0.62, 0.68, and 0.63, respectively. Facility Access Ratios Facility SCRs ranged from 0.00 to 2.87. Of the 216 facilities, 38 (18%) had an SCR significantly less than 1.00, 146 (68%) had an SCR not significantly different from 1.00, and 32 (15%) had an SCR significantly greater than 1.00 (Fig 1). Facility SFRs ranged from 0.08 to 2.60. Of the 216 facilities, 25 (12%) had an SFR significantly less than 1.00, 163 (75%) had an SFR not
277 Table 2. Univariate Relationship Between Vascular Access Type and Subject Characteristics Vascular Access Type (%)
Age (y) 18-44 45-54 55-64 65-74 75⫹ Sex Men Women Race White Black Other Height (cm) ⬍155 155-164 165-174 175-184 185⫹ Weight (kg) ⬍50 50-64 65-79 80-94 95⫹ Cause of renal failure Diabetes Hypertension Glomerulonephritis Other Time on dialysis 3-5.9 mo 6-11.9 mo 1-2.9 y 3-6.9 y 7⫹ y
N
Catheter
Fistula
Graft
2,749 2,359 3,068 4,190 2,973
20 19 20 20 22
32 26 22 21 20
47 55 58 59 58
7,998 7,341
18 23
32 14
50 63
8,182 6,679 478
21 20 21
26 21 28
53 59 51
1,098 4,148 5,331 3,596 1,166
27 22 20 18 17
14 16 24 32 34
59 62 56 50 49
1,055 4,153 5,028 3,019 2,084
30 24 18 18 19
18 22 27 23 22
52 54 55 59 60
5,744 4,481
22 18
18 26
60 56
2,261 2,853
18 22
31 25
51 52
1,253 2,453 6,025 3,986 1,622
36 25 20 16 13
21 24 21 24 33
43 51 59 60 54
P*
⬍0.001
⬍0.001 ⬍0.001
⬍0.001
⬍0.001
⬍0.001
⬍0.001
*Chi-squared test or analysis of variance.
significantly different from 1.00, and 28 (13%) had an SFR significantly greater than 1.00. Facility SGRs ranged from 0.29 to 1.70. Of the 216 facilities, 23 (11%) had an SGR significantly less than 1.00, 175 (81%) had an SGR not significantly different from 1.00, and 18 (8%) had an SGR significantly greater than 1.00. In general, facilities were classified similarly based on proportion of patients or on standardized access ratios. For example, 31 facilities
278
SEHGAL ET AL
Table 3. Multivariate Relationship Between Vascular Access Type and Subject Characteristics. Odds Ratio of 1.00 Indicates Reference Group for Each Predictor Variable
more types of vascular access based on proportion of patients versus standardized access ratios. Secondary Analyses
Odds Ratios Catheter
Age (y) 18-44 45-54 55-64 65-74 75⫹ Sex Men Women Race Black White/Other Height (cm) ⬍155 155-164 165-174 175-184 185⫹ Weight (kg) ⬍50 50-64 65-79 80-94 95⫹ Cause of renal failure Diabetes Hypertension Glomerulonephritis Other Time on dialysis 3-5.9 mo 6-11.9 mo 1-2.9 y 3-6.9 y 7⫹ y
Fistula
Graft
1.04 0.98 1.00 0.93 0.97
1.46* 1.16† 1.00 0.92 0.78*
0.72* 0.91 1.00 1.11† 1.19†
0.82* 1.00
2.37* 1.00
0.63* 1.00
1.05 1.00
0.67* 1.00
1.28* 1.00
1.11 0.93 1.00 1.01 0.97
0.79† 0.90 1.00 1.14† 1.32*
1.05 1.12† 1.00 0.89† 0.80†
1.92* 1.41* 1.00 0.96 1.02
0.89 0.93 1.00 0.78* 0.66*
0.68* 0.85* 1.00 1.23* 1.35*
1.17† 1.00 1.03 1.22†
0.72* 1.00 1.09 0.87†
1.13† 1.00 0.91 0.97
2.27* 1.33* 1.00 0.76* 0.56*
0.96 1.14† 1.00 1.15† 1.66*
0.53* 0.74* 1.00 1.08 0.94
*P ⬍ 0.001 by logistic regression analysis. †P ⬍ 0.05 by logistic regression analysis.
were classified as having fewer catheters than expected using either method, 139 were classified as having as many catheters as expected, and 32 were classified as having more catheters than expected (Table 4). However, a total of 14 (6%) facilities were classified differently using SCRs versus proportion of patients with catheters. Results for SFRs and SGRs were similar (7% and 6% classified differently). Altogether, 40 (19%) facilities were classified differently for one or
As noted above, 1,403 patients from 91 small facilities were excluded from the main analyses. When analyzed in aggregate, these patients had an SCR of 0.91 (95% confidence interval [CI] 0.81, 1.02), an SFR of 0.84 (CI 0.74, 0.94), and an SGR of 1.11 (CI 1.04, 1.19). Information on additional comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug use) was available for 9,826 patients who began dialysis in the last 3 years. Including this information in calculations did not appreciably affect the expected probability of catheter, fistula, or graft use. The correlation coefficients between the baseline calculations reported above and these new calculations were 0.96, 0.97, and 0.98 for catheters, fistulas, and grafts, respectively. DISCUSSION
Our study indicates that less than one fourth of patients have fistulas, whereas one fifth have catheters. More importantly, we found that vascular access type varies greatly across providers and that considerable variability persists even after adjusting for the patient characteristics examined (Fig 1). For example, in unadjusted analyses, 17% of facilities had a high proportion of patients with catheters. Adjustment for patient characteristics decreased this only slightly to 15% of facilities with an SCR greater than 1.00. The marked variation in standardized access ratios, which account for patient characteristics, suggests that provider factors play an important role in determining type of vascular access. Although it is possible that unmeasured patient characteristics contribute, it is unlikely that they are sufficient to entirely explain the variation observed. Our large, contemporary sample consists of patients from all chronic hemodialysis facilities in four states. Demographic characteristics of these patients are comparable to those of patients nationally.14 Moreover, the relationships we found between patient characteristics and vascular access type are consistent with those described in a national sample.10 Because creation and matura-
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
279
Fig 1. Distribution of SCRs by facility. Of the 216 facilities, 38 (18%) had an SCR significantly F1.00, 146 (68%) had an SCR not significantly different from 1.00, and 32 (15%) had an SCR significantly G1.00. The boxes show the median as the line across the middle and the 25th and 75th percentiles as the ends. The bars represent the 10th and 90th percentiles.
tion of fistulas and grafts may take anywhere from a couple of weeks to several months, preESRD care is an important determinant of type of vascular access at initiation of dialysis. However, because we excluded patients new to dialysis, the sizable proportion of patients with catheters cannot be ascribed to care individuals received before becoming dialysis patients at specific facilities. Even patients on dialysis for more than 1 year had a substantial prevalence of catheter use. Table 4. Classification of Facilities by Proportion of Patients With Catheters Versus SCR SCR Significantly ⬍1.00
Proportion with catheters : Significantly ⬍20% Not significantly different from 20% Significantly ⬎20%
Not Significantly Different From 1.00
Significantly ⬎1.00
31
2
0
7 0
139 5
0 32
NOTE. The mean proportion of patients with catheters across facilities was 20%. A total of 202 (31 ⫹ 139 ⫹ 32) facilities were classified similarly with both methods, whereas the remaining 14 (7 ⫹ 2 ⫹ 5) were classified differently.
The standardized access ratios we calculated are interpreted in a similar manner as the SMR and the SHR.11-13 For example, the SCR compares catheter use among patients at a specific facility with patients across all four states of the same mix of age, sex, race, height, weight, cause of renal failure, and duration of ESRD. An SCR of 1.30 indicates that a facility has 30% more patients with catheters than would be expected based on the characteristics of patients at that facility. An SCR not significantly different from 1.00 indicates the facility has as many patients with catheters as would be expected. An alternate approach would focus on the proportion of patients at each facility with a specific vascular access type (eg, if 30% of a facility’s patients have catheters compared with 20% of patients in all four states). This is especially appealing because most facilities are classified similarly by proportion of patients or by a standardized ratio (Table 4). However, this method does not account for patient characteristics and therefore cannot be used to examine provider-related variability in vascular access use. Nearly one fifth of facilities will be incorrectly classified for one or more types of vascular access by focusing solely on the proportion of patients. Adjusting for patient characteristics may
280
also enhance the willingness of providers to use these data as quality improvement tools. We caution against using standardized access ratios to rank facilities because of several limitations, many of which are common to all standardized ratios such as SMRs and SHRs.11-13 First, it is difficult to account for patients moving from one facility to another. Second, factors beyond the control of facilities such as surgical technique may play a role. Third, it is often impractical to obtain data on all relevant patient characteristics. For example, the presence and severity of peripheral vascular disease influences the type of vascular access. Although we had information on the presence of vascular disease for a subset of patients, we had no knowledge of its severity. Fourth, it is important to adjust only for preexisting patient characteristics and not for factors that may reflect the process of care being examined. For example, the number of prior vascular accesses or vascular access interventions is an important determinant of subsequent vascular access type, but number of prior vascular accesses or interventions also may reflect the process of care at a particular facility. Excluding such variables from the adjustment process is not always straightforward because it is possible for a patient characteristic (such as race) to reflect both biological differences and process of care. In this case, including race (as we did) risks the possibility of inappropriately adjusting for process of care, but excluding race risks the possibility of not adjusting for biological differences that may account for differences in outcomes. Fifth, using standardized ratios to rank facilities may inadvertently penalize attempts to increase fistula use. For example, a facility that emphasizes fistula placement may have a high SCR because many patients are using catheters while awaiting fistula maturation. Sixth, excluding patients on dialysis for less than 3 months may not be sufficient to eliminate the effect of predialysis care (or lack thereof) on a facility’s standardized access ratio. For example, a patient who first presents to a nephrologist with ESRD may be started acutely on dialysis with a catheter, referred for a fistula only after a period of stabilization, and require a prolonged period for fistula maturation. We recommend that providers view standardized access ratios not as a final ranking of their
SEHGAL ET AL
performance, but rather as a starting point for examining a number of vascular access–related issues. Specifically, providers may wish to (1) examine processes of care such as needle placement or delayed surgical referral and (2) explore additional patient factors such as previous central vein catheter placement, prior axillary/ thoracic surgery, and lymph node dissection. Studying practice patterns at facilities with low SCRs and high SFRs also may guide quality improvement efforts at other facilities. Our formulas for calculating standardized access ratios are included in the Appendix (the Appendix appears only on the web site [www.ajkd. org] electronic pages). However, because vascular access differs among the various ESRD Networks, we recommend that other Networks develop formulas based on their patient and provider population.10 Collecting uniform data and including patient characteristics unavailable to us such as severity of peripheral vascular disease will improve the quality of adjustment. Although it is not possible to derive statistically meaningful results for small facilities, aggregating patients across multiple years or aggregating all of the facilities in a chain may provide useful information. In this sample, patients at small facilities were slightly more likely to have grafts and less likely to have fistulas. In conclusion, providers should not attribute suboptimal access types to patient case mix. Standardized access ratios should be useful screening tests that prompt providers to examine processes of care that contribute to variability in access use. ACKNOWLEDGMENT The authors thank the staff of The Renal Network (End Stage Renal Disease Networks 9 and 10) for making this project possible. We also thank the providers who collected and submitted information on their patients.
REFERENCES 1. National Kidney Foundation: DOQI Clinical Practice Guidelines for Vascular Access. Am J Kidney Dis 30:S150S191, 1997 (suppl 2) 2. Fan PY, Schwab SJ: Vascular access: Concepts for the 1990s. J Am Soc Nephrol 3:1-11, 1992 3. Feldman H, Held PJ, Hutchinson JT, Stoiber E, Hartigan MF, Berlin JA: Hemodialysis vascular access morbidity in the United States. Kidney Int 43:1091-1096, 1993 4. Feldman HI, Kobrin S, Wasserstein A: Hemodialysis vascular access morbidity. J Am Soc Nephrol 7:523-535, 1996
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
5. Hakim R, Himmelfarb J: Hemodialysis access failure: A call to action. Kidney Int 54:1029-1040, 1998 6. Windus DW: Permanent vascular access: A nephrologist’s view. Am J Kidney Dis 21:457-471, 1993 7. Woods JD, Turenne MN, Strawderman RL, Young EW, Hirth RA, Port FK, Held PJ: Vascular access survival among incident hemodialysis patients in the United States. Am J Kidney Dis 30:50-57, 1997 8. Rocco MV, Bleyer AJ, Burkart JM: Utilization of inpatient and outpatient resources for the management of hemodialysis access complications. Am J Kidney Dis 28:250256, 1996 9. Sehgal AR, Snow RJ, Singer ME, Amini SB, DeOreo PB, Silver MR, Cebul RD: Barriers to adequate delivery of hemodialysis. Am J Kidney Dis 31:593-601, 1998 10. Hirth RA, Turenne MN, Woods JD, Young EW, Port FK, Pauly MV, Held PJ: Predictors of type of vascular
281
access in hemodialysis patients. J Am Med Assoc 276:13031308, 1996 11. Wolfe RA, Gaylin DS, Port FK, Held PJ, Wood CL: Using USRDS generated mortality tables to compare local ESRD mortality rates to national rates. Kidney Int 42:991996, 1992 12. Wolfe RA: The standardized mortality ratio revisited: Improvements, innovations, and limitations. Am J Kidney Dis 24:290-297, 1994 13. Strawderman RL, Levine G, Hirth RA, Port FK, Held PJ: Using USRDS generated hospitalization tables to compare local dialysis patient hospitalization rates to national rates. Kidney Int 50:571-578, 1996 14. US Renal Data System: USRDS 1998 Annual Report. The National Institutes of Health, National Institute of Diabetics and Digestive and Kidney Diseases. Bethesda, MD, 1998
Appendix appears only on the web site (www.ajkd.org) electronic pages
T
HE TYPE OF hemodialysis vascular access (catheter, fistula, or graft) strongly influences several important patient outcomes. For example, fistulas are typically associated with (1) increased access longevity, (2) better dialysis adequacy, (3) fewer infectious or thrombotic complications, and (4) lower health care costs. By contrast, catheters are associated with poor outcomes in all four areas, whereas grafts tend to have intermediate results.1-9 Despite the clear advantages of fistulas, only a minority of patients have fistulas. Moreover, catheter and graft use appears to be increasing. There are also striking geographic differences in access type as exemplified by a 13-fold variation in use of grafts versus fistulas across nine census regions.10 The relative importance of patient versus provider factors in determining variability of vascular access type is unclear. We sought to develop a quality improvement tool that adjusts for differences in patient characteristics, thereby allowing examination of provider-related variabil-
ity in vascular access type. As models for this effort, we selected the standardized mortality ratio (SMR) and the standardized hospitalization ratio (SHR), two widely used measures of provider-related variability across dialysis facilities.11-13
From the Division of Nephrology, MetroHealth Medical Center, Cleveland, OH; the Department of Medicine, Case Western Reserve University, Cleveland, OH; the Center for Biomedical Ethics, Case Western Reserve University, Cleveland, OH; the Department of Epidemiology and Biostatistics, Case Western Reserve University, Cleveland, OH; The Renal Network, Inc, Indianapolis, IN; the Divisions of Geriatrics and General Medicine, Veterans Administration Medical Center, San Francisco, CA; and the Department of Medicine, University of California, San Francisco, CA. Received April 29, 1999; accepted in revised form September 10, 1999. Supported by grant number DK51472 from the National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, MD, and contracts 500-97-E027 and 500-97-E028 from the Health Care Financing Administration, Baltimore, MD. Address reprint requests to Ashwini R. Sehgal, MD, Division of Nephrology, 2500 MetroHealth Dr, MetroHealth Medical Center, Cleveland, OH 44109-1998. E-mail: [email protected]
娀 2000 by the National Kidney Foundation, Inc. 0272-6386/00/3502-0013$3.00/0
American Journal of Kidney Diseases, Vol 35, No 2 (February), 2000: pp 275-281
275
276
SEHGAL ET AL
METHODS
RESULTS
Subjects and Facilities
Subject Characteristics
Patients age 18 years or older on chronic hemodialysis for at least 3 months were eligible for inclusion. Our facility sample consisted of all chronic hemodialysis facilities in Illinois, Indiana, Ohio, and Kentucky that had at least 30 eligible patients.
All 307 chronic hemodialysis facilities in Illinois, Indiana, Ohio, and Kentucky reported complete information on a total of 16,832 eligible patients. We excluded 1,403 patients from 91 small facilities, leaving 15,429 subjects from 216 facilities. Of these, 90 patients had more than one access reported (eg, if blood was drawn from a maturing fistula and returned to a catheter). Our remaining analyses focus on the 15,339 subjects with one access. About half the subjects were male, about half were white, and 37% had endstage renal disease (ESRD) secondary to diabetes (Table 1). The mean duration of ESRD was 3.2 years, with a range from 3 months to 27 years.
Data Collection As part of a quality improvement project of End Stage Renal Disease Networks 9 and 10, all facilities reported the type of vascular access used for the first hemodialysis treatment received by each patient during the week of December 25 to 31, 1997. We obtained patient demographic (age, sex, race, height, weight) and medical characteristics (cause of renal failure, time on dialysis) from the Network database for all subjects. Because of a change in reporting requirements in 1995, we were also able to obtain information on select comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug dependence) for all patients beginning dialysis in the last 3 years.
Statistical Analysis We used two different units of analysis: first patients and then facilities. At a patient level, we used the chi-squared test or analysis of variance to examine the univariate relationship between actual catheter use and the patient demographic and medical characteristics listed above. We used logistic regression to examine the multivariate relationship between actual catheter use and all patient characteristics. To account for possible nonlinear relationships, continuous predictor variables were entered as a series of indicator variables (eg, five categories for age). We then used the odds ratios from this regression analysis to calculate the expected probability of catheter use (as opposed to actual catheter use) for each patient, taking into account the patient’s demographic and medical characteristics. At a facility level, we summed the expected probability of catheter use of all patients at a particular facility to determine the total expected number of patients with catheters. This expected number takes into account the demographic and medical characteristics of the patients at each facility. We then calculated the standardized catheter ratio (SCR) as the actual number of patients with catheters divided by the expected number of patients with catheters. We used the chi-squared test to determine whether the SCR was significantly different from 1.00.12 We calculated the standardized fistula ratios (SFR) and standardized graft ratios (SGR) in a similar manner. We also performed several secondary analyses. First, we calculated standardized ratios for patients from small facilities in aggregate. Next, we determined the impact of additional comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug use) by calculating standardized ratios on the subset of patients for whom this information was available.
Vascular Access Type Of all subjects, 20% had catheters, 24% had fistulas, and 56% had grafts. Among the 3,128 subjects with catheters, 76% had tunneled, cuffed catheters and 24% had temporary catheters. Reported reasons for catheter use were: no vascular sites (37%), no graft/fistula created (25%), temporary interruption in graft/fistula (13%), graft/ fistula maturing (13%), other specific reason (7%), and no reason reported (5%). The proportion of patients with catheters ranged from 0% to 63% (mean, 20%) across the 216 facilities. In 33 (15%) facilities, the proportion of patients with catheters was significantly less than 20%; in 146 (68%) facilities, the proportion was not significantly different from 20%; and in 37 (17%), the proportion was significantly Table 1. Subject Characteristics (n ⴝ 15,339 Patients) Mean age, years (range) Men (%) Race (%) White Black Other Mean height, centimeters (range) Mean weight, kilograms (range) Cause of renal failure (%) Diabetes Hypertension Glomerulonephritis Other Mean time on dialysis, years (range)
61 (18-98) 52 53 44 3 169 (137-213) 74 (32-181) 37 29 15 19 3.2 (0.25-27)
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
greater than 20%. The proportion of patients with fistulas ranged from 2% to 62% (mean, 24%) across facilities. In 28 (13%) facilities, the proportion of patients with fistulas was significantly less than 24%; in 161 (75%) facilities, the proportion was not significantly different from 24%; and in 27 (13%), the proportion was significantly greater than 24%. The proportion of patients with grafts ranged from 16% to 91% (mean, 56%). In 27 (13%) facilities, the proportion of patients with grafts was significantly less than 56%; in 171 (79%) facilities, the proportion was not significantly different from 56%; and in 18 (8%), the proportion was significantly greater than 56%. Correlates of Vascular Access Type On univariate analysis (Table 2), catheter use was most strikingly associated with short height, low weight, and a short time since ESRD developed. For example, 36% of patients with ESRD for 3 to 5.9 months had catheters, compared with 13% of patients with ESRD for 7 or more years. Fistula use was most strikingly associated with younger age, male gender, large height, nondiabetic causes of renal failure, and a long time since ESRD developed. For example, 32% of men had fistulas compared with 14% of women. On multivariate analysis (Table 3), young patients, men, and whites were more likely to have fistulas, whereas old patients, women, and blacks were more likely to have grafts. Heavy patients were more likely to have grafts, and lighter patients were more likely to have catheters. Diabetics were more likely to have catheters and less likely to have fistulas. New patients were more likely to have catheters and less likely to have grafts. The c statistic for the catheter, fistula, and graft analyses were 0.62, 0.68, and 0.63, respectively. Facility Access Ratios Facility SCRs ranged from 0.00 to 2.87. Of the 216 facilities, 38 (18%) had an SCR significantly less than 1.00, 146 (68%) had an SCR not significantly different from 1.00, and 32 (15%) had an SCR significantly greater than 1.00 (Fig 1). Facility SFRs ranged from 0.08 to 2.60. Of the 216 facilities, 25 (12%) had an SFR significantly less than 1.00, 163 (75%) had an SFR not
277 Table 2. Univariate Relationship Between Vascular Access Type and Subject Characteristics Vascular Access Type (%)
Age (y) 18-44 45-54 55-64 65-74 75⫹ Sex Men Women Race White Black Other Height (cm) ⬍155 155-164 165-174 175-184 185⫹ Weight (kg) ⬍50 50-64 65-79 80-94 95⫹ Cause of renal failure Diabetes Hypertension Glomerulonephritis Other Time on dialysis 3-5.9 mo 6-11.9 mo 1-2.9 y 3-6.9 y 7⫹ y
N
Catheter
Fistula
Graft
2,749 2,359 3,068 4,190 2,973
20 19 20 20 22
32 26 22 21 20
47 55 58 59 58
7,998 7,341
18 23
32 14
50 63
8,182 6,679 478
21 20 21
26 21 28
53 59 51
1,098 4,148 5,331 3,596 1,166
27 22 20 18 17
14 16 24 32 34
59 62 56 50 49
1,055 4,153 5,028 3,019 2,084
30 24 18 18 19
18 22 27 23 22
52 54 55 59 60
5,744 4,481
22 18
18 26
60 56
2,261 2,853
18 22
31 25
51 52
1,253 2,453 6,025 3,986 1,622
36 25 20 16 13
21 24 21 24 33
43 51 59 60 54
P*
⬍0.001
⬍0.001 ⬍0.001
⬍0.001
⬍0.001
⬍0.001
⬍0.001
*Chi-squared test or analysis of variance.
significantly different from 1.00, and 28 (13%) had an SFR significantly greater than 1.00. Facility SGRs ranged from 0.29 to 1.70. Of the 216 facilities, 23 (11%) had an SGR significantly less than 1.00, 175 (81%) had an SGR not significantly different from 1.00, and 18 (8%) had an SGR significantly greater than 1.00. In general, facilities were classified similarly based on proportion of patients or on standardized access ratios. For example, 31 facilities
278
SEHGAL ET AL
Table 3. Multivariate Relationship Between Vascular Access Type and Subject Characteristics. Odds Ratio of 1.00 Indicates Reference Group for Each Predictor Variable
more types of vascular access based on proportion of patients versus standardized access ratios. Secondary Analyses
Odds Ratios Catheter
Age (y) 18-44 45-54 55-64 65-74 75⫹ Sex Men Women Race Black White/Other Height (cm) ⬍155 155-164 165-174 175-184 185⫹ Weight (kg) ⬍50 50-64 65-79 80-94 95⫹ Cause of renal failure Diabetes Hypertension Glomerulonephritis Other Time on dialysis 3-5.9 mo 6-11.9 mo 1-2.9 y 3-6.9 y 7⫹ y
Fistula
Graft
1.04 0.98 1.00 0.93 0.97
1.46* 1.16† 1.00 0.92 0.78*
0.72* 0.91 1.00 1.11† 1.19†
0.82* 1.00
2.37* 1.00
0.63* 1.00
1.05 1.00
0.67* 1.00
1.28* 1.00
1.11 0.93 1.00 1.01 0.97
0.79† 0.90 1.00 1.14† 1.32*
1.05 1.12† 1.00 0.89† 0.80†
1.92* 1.41* 1.00 0.96 1.02
0.89 0.93 1.00 0.78* 0.66*
0.68* 0.85* 1.00 1.23* 1.35*
1.17† 1.00 1.03 1.22†
0.72* 1.00 1.09 0.87†
1.13† 1.00 0.91 0.97
2.27* 1.33* 1.00 0.76* 0.56*
0.96 1.14† 1.00 1.15† 1.66*
0.53* 0.74* 1.00 1.08 0.94
*P ⬍ 0.001 by logistic regression analysis. †P ⬍ 0.05 by logistic regression analysis.
were classified as having fewer catheters than expected using either method, 139 were classified as having as many catheters as expected, and 32 were classified as having more catheters than expected (Table 4). However, a total of 14 (6%) facilities were classified differently using SCRs versus proportion of patients with catheters. Results for SFRs and SGRs were similar (7% and 6% classified differently). Altogether, 40 (19%) facilities were classified differently for one or
As noted above, 1,403 patients from 91 small facilities were excluded from the main analyses. When analyzed in aggregate, these patients had an SCR of 0.91 (95% confidence interval [CI] 0.81, 1.02), an SFR of 0.84 (CI 0.74, 0.94), and an SGR of 1.11 (CI 1.04, 1.19). Information on additional comorbid conditions (peripheral vascular disease, diabetes, tobacco use, illicit drug use) was available for 9,826 patients who began dialysis in the last 3 years. Including this information in calculations did not appreciably affect the expected probability of catheter, fistula, or graft use. The correlation coefficients between the baseline calculations reported above and these new calculations were 0.96, 0.97, and 0.98 for catheters, fistulas, and grafts, respectively. DISCUSSION
Our study indicates that less than one fourth of patients have fistulas, whereas one fifth have catheters. More importantly, we found that vascular access type varies greatly across providers and that considerable variability persists even after adjusting for the patient characteristics examined (Fig 1). For example, in unadjusted analyses, 17% of facilities had a high proportion of patients with catheters. Adjustment for patient characteristics decreased this only slightly to 15% of facilities with an SCR greater than 1.00. The marked variation in standardized access ratios, which account for patient characteristics, suggests that provider factors play an important role in determining type of vascular access. Although it is possible that unmeasured patient characteristics contribute, it is unlikely that they are sufficient to entirely explain the variation observed. Our large, contemporary sample consists of patients from all chronic hemodialysis facilities in four states. Demographic characteristics of these patients are comparable to those of patients nationally.14 Moreover, the relationships we found between patient characteristics and vascular access type are consistent with those described in a national sample.10 Because creation and matura-
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
279
Fig 1. Distribution of SCRs by facility. Of the 216 facilities, 38 (18%) had an SCR significantly F1.00, 146 (68%) had an SCR not significantly different from 1.00, and 32 (15%) had an SCR significantly G1.00. The boxes show the median as the line across the middle and the 25th and 75th percentiles as the ends. The bars represent the 10th and 90th percentiles.
tion of fistulas and grafts may take anywhere from a couple of weeks to several months, preESRD care is an important determinant of type of vascular access at initiation of dialysis. However, because we excluded patients new to dialysis, the sizable proportion of patients with catheters cannot be ascribed to care individuals received before becoming dialysis patients at specific facilities. Even patients on dialysis for more than 1 year had a substantial prevalence of catheter use. Table 4. Classification of Facilities by Proportion of Patients With Catheters Versus SCR SCR Significantly ⬍1.00
Proportion with catheters : Significantly ⬍20% Not significantly different from 20% Significantly ⬎20%
Not Significantly Different From 1.00
Significantly ⬎1.00
31
2
0
7 0
139 5
0 32
NOTE. The mean proportion of patients with catheters across facilities was 20%. A total of 202 (31 ⫹ 139 ⫹ 32) facilities were classified similarly with both methods, whereas the remaining 14 (7 ⫹ 2 ⫹ 5) were classified differently.
The standardized access ratios we calculated are interpreted in a similar manner as the SMR and the SHR.11-13 For example, the SCR compares catheter use among patients at a specific facility with patients across all four states of the same mix of age, sex, race, height, weight, cause of renal failure, and duration of ESRD. An SCR of 1.30 indicates that a facility has 30% more patients with catheters than would be expected based on the characteristics of patients at that facility. An SCR not significantly different from 1.00 indicates the facility has as many patients with catheters as would be expected. An alternate approach would focus on the proportion of patients at each facility with a specific vascular access type (eg, if 30% of a facility’s patients have catheters compared with 20% of patients in all four states). This is especially appealing because most facilities are classified similarly by proportion of patients or by a standardized ratio (Table 4). However, this method does not account for patient characteristics and therefore cannot be used to examine provider-related variability in vascular access use. Nearly one fifth of facilities will be incorrectly classified for one or more types of vascular access by focusing solely on the proportion of patients. Adjusting for patient characteristics may
280
also enhance the willingness of providers to use these data as quality improvement tools. We caution against using standardized access ratios to rank facilities because of several limitations, many of which are common to all standardized ratios such as SMRs and SHRs.11-13 First, it is difficult to account for patients moving from one facility to another. Second, factors beyond the control of facilities such as surgical technique may play a role. Third, it is often impractical to obtain data on all relevant patient characteristics. For example, the presence and severity of peripheral vascular disease influences the type of vascular access. Although we had information on the presence of vascular disease for a subset of patients, we had no knowledge of its severity. Fourth, it is important to adjust only for preexisting patient characteristics and not for factors that may reflect the process of care being examined. For example, the number of prior vascular accesses or vascular access interventions is an important determinant of subsequent vascular access type, but number of prior vascular accesses or interventions also may reflect the process of care at a particular facility. Excluding such variables from the adjustment process is not always straightforward because it is possible for a patient characteristic (such as race) to reflect both biological differences and process of care. In this case, including race (as we did) risks the possibility of inappropriately adjusting for process of care, but excluding race risks the possibility of not adjusting for biological differences that may account for differences in outcomes. Fifth, using standardized ratios to rank facilities may inadvertently penalize attempts to increase fistula use. For example, a facility that emphasizes fistula placement may have a high SCR because many patients are using catheters while awaiting fistula maturation. Sixth, excluding patients on dialysis for less than 3 months may not be sufficient to eliminate the effect of predialysis care (or lack thereof) on a facility’s standardized access ratio. For example, a patient who first presents to a nephrologist with ESRD may be started acutely on dialysis with a catheter, referred for a fistula only after a period of stabilization, and require a prolonged period for fistula maturation. We recommend that providers view standardized access ratios not as a final ranking of their
SEHGAL ET AL
performance, but rather as a starting point for examining a number of vascular access–related issues. Specifically, providers may wish to (1) examine processes of care such as needle placement or delayed surgical referral and (2) explore additional patient factors such as previous central vein catheter placement, prior axillary/ thoracic surgery, and lymph node dissection. Studying practice patterns at facilities with low SCRs and high SFRs also may guide quality improvement efforts at other facilities. Our formulas for calculating standardized access ratios are included in the Appendix (the Appendix appears only on the web site [www.ajkd. org] electronic pages). However, because vascular access differs among the various ESRD Networks, we recommend that other Networks develop formulas based on their patient and provider population.10 Collecting uniform data and including patient characteristics unavailable to us such as severity of peripheral vascular disease will improve the quality of adjustment. Although it is not possible to derive statistically meaningful results for small facilities, aggregating patients across multiple years or aggregating all of the facilities in a chain may provide useful information. In this sample, patients at small facilities were slightly more likely to have grafts and less likely to have fistulas. In conclusion, providers should not attribute suboptimal access types to patient case mix. Standardized access ratios should be useful screening tests that prompt providers to examine processes of care that contribute to variability in access use. ACKNOWLEDGMENT The authors thank the staff of The Renal Network (End Stage Renal Disease Networks 9 and 10) for making this project possible. We also thank the providers who collected and submitted information on their patients.
REFERENCES 1. National Kidney Foundation: DOQI Clinical Practice Guidelines for Vascular Access. Am J Kidney Dis 30:S150S191, 1997 (suppl 2) 2. Fan PY, Schwab SJ: Vascular access: Concepts for the 1990s. J Am Soc Nephrol 3:1-11, 1992 3. Feldman H, Held PJ, Hutchinson JT, Stoiber E, Hartigan MF, Berlin JA: Hemodialysis vascular access morbidity in the United States. Kidney Int 43:1091-1096, 1993 4. Feldman HI, Kobrin S, Wasserstein A: Hemodialysis vascular access morbidity. J Am Soc Nephrol 7:523-535, 1996
STANDARDIZED RATIOS TO EXAMINE VASCULAR ACCESS
5. Hakim R, Himmelfarb J: Hemodialysis access failure: A call to action. Kidney Int 54:1029-1040, 1998 6. Windus DW: Permanent vascular access: A nephrologist’s view. Am J Kidney Dis 21:457-471, 1993 7. Woods JD, Turenne MN, Strawderman RL, Young EW, Hirth RA, Port FK, Held PJ: Vascular access survival among incident hemodialysis patients in the United States. Am J Kidney Dis 30:50-57, 1997 8. Rocco MV, Bleyer AJ, Burkart JM: Utilization of inpatient and outpatient resources for the management of hemodialysis access complications. Am J Kidney Dis 28:250256, 1996 9. Sehgal AR, Snow RJ, Singer ME, Amini SB, DeOreo PB, Silver MR, Cebul RD: Barriers to adequate delivery of hemodialysis. Am J Kidney Dis 31:593-601, 1998 10. Hirth RA, Turenne MN, Woods JD, Young EW, Port FK, Pauly MV, Held PJ: Predictors of type of vascular
281
access in hemodialysis patients. J Am Med Assoc 276:13031308, 1996 11. Wolfe RA, Gaylin DS, Port FK, Held PJ, Wood CL: Using USRDS generated mortality tables to compare local ESRD mortality rates to national rates. Kidney Int 42:991996, 1992 12. Wolfe RA: The standardized mortality ratio revisited: Improvements, innovations, and limitations. Am J Kidney Dis 24:290-297, 1994 13. Strawderman RL, Levine G, Hirth RA, Port FK, Held PJ: Using USRDS generated hospitalization tables to compare local dialysis patient hospitalization rates to national rates. Kidney Int 50:571-578, 1996 14. US Renal Data System: USRDS 1998 Annual Report. The National Institutes of Health, National Institute of Diabetics and Digestive and Kidney Diseases. Bethesda, MD, 1998