Optimal Cutoff Point for Immunoperoxidase Detection of C4d in the Renal Allograft: Results From a Multicenter Study : Transplantation

Journal Logo

Clinical and Translational Research

Optimal Cutoff Point for Immunoperoxidase Detection of C4d in the Renal Allograft: Results From a Multicenter Study

Crary, Gretchen S.1; Raissian, Yassaman2; Gaston, Robert C.3; Gourishankar, Sita M.4; Leduc, Robert E.5; Mannon, Roslyn B.6; Matas, Arthur J.7; Grande, Joseph P.2,8

Author Information
Transplantation 90(10):p 1099-1105, November 27, 2010. | DOI: 10.1097/TP.0b013e3181f7fec9
  • Free

Abstract

Peritubular capillary deposition of C4d has been identified as a risk factor for the development of progressive graft dysfunction and loss (1–3). Peritubular capillary staining for C4d and presence of donor-specific antibodies have been incorporated into the revised Banff ’97 classification scheme as criteria for the diagnosis of acute humoral rejection (4). Moreover, in patients with chronic graft dysfunction, C4d predicts adverse outcome and is associated with a variety of histopathologic alterations, including transplant glomerulopathy (1, 5). Based on these considerations, it has been recommended that staining for C4d be performed on all renal allograft biopsies (6).

Methods for detection of C4d deposition in peritubular capillaries include immunofluorescence performed on frozen tissue and immunoperoxidase performed on formalin-fixed, paraffin-embedded tissue. There are advantages to both methods for detection of C4d, for example, it is recognized that detection of C4d by immunofluorescence provides the most sensitive and specific marker for C4d positivity (7–9). However, immunofluorescence studies require the acquisition of additional tissue, which may not be practical for protocol biopsies or for retrospective studies. Unfortunately, there is a considerable variation between centers on thresholds for “focal” or “diffuse” positive staining, both by immunofluorescence and immunoperoxidase, rendering it difficult to compare the results for C4d positivity between centers. Furthermore, there are no data in the literature that systematically correlate various cutoff points for the detection of C4d with outcome. At the Banff ’07 Meeting on Allograft Pathology, the need to correlate C4d cutoffs with long-term outcome was emphasized (6).

The purpose of this study was to establish an optimal cutoff for C4d deposition, as assessed by central analysis of formalin-fixed, paraffin-embedded tissues stained by the immunoperoxidase method. Study patients were enrolled in the Deterioration of Kidney Allograft Function Study, a multicenter study to define clinical and histopathologic features that predict adverse outcome in renal biopsies performed for new-onset late graft dysfunction (10). In this cohort of patients, recruited from seven centers in the United States and Canada, we demonstrate that a cutoff of 10% of peritubular capillaries to establish a positive C4d stain provided the best fit for the association of C4d positivity with graft loss.

RESULTS

The study cohort consists of 296 patients who underwent a clinically indicated renal biopsy for new-onset deterioration of function (defined as 25% increase in serum creatinine or new onset of proteinuria) for which both local and central C4d data are available. The percentage of peritubular capillaries staining positively for C4d was assessed in a blinded fashion. A representative photomicrograph showing diffuse positive and negative immunoperoxidase staining for C4d is shown in Figure 1. The percentage of C4d positivity ranged from 0% to 90% with 44% (129/296) having a positive percentage of C4d staining; among those with positive percentage C4d, the median percentage was 25%. Local C4d+ results were reported qualitatively, with 28% (83/296) recorded as positive for C4d. Using a centrally determined cutoff of 10%, the κ score for agreement of local and centrally determined C4d positivity was 0.40 (95% confidence interval [CI] 0.29–0.51). Raising the cutoff for centrally determined C4d values did not appreciably change the κ values—using a centrally determined cutoff of 25%, the κ value was 0.46 (95% CI 0.35–0.57), and using cutoff of 50% for centrally determined positivity, the κ value was 0.42 (95% CI 0.30–0.54). The overall proportion of observed agreement ranged between 69% and 79% for all putative cutoffs examined. The proportion of specific agreement for C4d+ negativity (range 76%–87%) was higher than the proportion of specific agreement for positivity (54%–60%), indicating that observed agreement was superior on cases where the local or central result was negative rather than on cases where at least one result was positive, for each putative cutoff value.

F1-11
FIGURE 1.:
Representative histopathologic section of renal biopsy stained by immunoperoxidase for C4d showing diffuse positive staining within peritubular capillaries (A). Representative section of a renal biopsy showing no significant staining for C4d (B).

Cox proportional hazards models were used to estimate the unadjusted risk of graft loss by local and central C4d results. Selection of an optimal cutoff for central C4d positivity was performed by a maximum chi-square search (Table 1), or because all models have the same number of degrees of freedom, equivalently by Akaike Information Criterion. The model with a C4d cutoff of 10% provided the best-fitting model by these measures. A Kaplan-Meier graph of centrally determined C4d positivity using a cutoff threshold of 10% is given in Figure 2(A).

T1-11
TABLE 1:
Hazard ratios from univariate models for time-to-graft failure (including death) by central C4d positivity cutoff or local determination
F2-11
FIGURE 2.:
Time to graft failure by C4d status. Immunoperoxidase staining for C4d was centrally performed and read in a blinded fashion. (A) Time to graft failure as a function of C4d status, using a centrally defined cutoff threshold of 10%. (B) Time to graft failure as a function of C4d status, using centrally defined cutoff values of less than 10% (negative), 10% to 24%, or more than or equal to 25%.

To correct for overfitting (“optimism”) in the fitting of the model as a result of multiple comparisons used in the maximum chi-square search, 2-fold cross-validation was used. Cross-validation analysis estimated a statistically significant association of C4d+ positivity (likelihood ratio chi-square test=19.9, P<0.0001) with a hazard ratio [HR] of 2.66 (95% CI 1.68–4.21), compared with a HR of 2.88 for the direct univariate analysis.

Harrell’s c-index corrected for overfitting through bootstrapping shows that the discrimination of C4d positivity is also greatest at a 10% cutoff (c=0.63), whereas the discrimination of local C4d positivity (c=0.53) is comparable with the null model (c=0.50). Calibration of the model with positivity cutoff of 10% was assessed through bootstrapping the difference between predicted model-based and observed Kaplan-Meier estimates of years 1 and 2 survival to estimate the degree of overfitting in the model (11). The direct estimate of survival and its standard error at year 1 postbiopsy was 81.0%±3.0%, with an estimated survival rate corrected for optimism of 81.2%. Likewise, at year 2, the direct estimate of survival was 71.0%±4.4%, whereas the estimate corrected for optimism is 71.1%. Thus, the model seems to be well calibrated in prediction of time to graft loss.

Increasing the centrally determined C4d-positive threshold beyond 10% weakened the association between C4d positivity and outcome—using a centrally determined cutoff of 40%, there was no significant association between C4d positivity and outcome. Of note, there was no significant association between locally determined C4d positivity (cutoff not specified) and outcome (HR 1.47, 95% CI 0.9–2.38, P=0.121; Table 1).

To further demonstrate the risk associated with relatively low levels of C4d positivity, the time to graft failure analyses was compared between the three subgroups defined by central C4d+ results of less than 10%, 10% to 24%, and more than or equal to 25%. A Kaplan-Meier graph showing these three risk groups is given in Figure 2(B). In proportional hazards regression comparing with the group with less than 10% central C4d, the HR for graft failure for the group with central C4d from 10% to 24% was 3.3 (95% CI 1.8–6.0, P<0.001), whereas the estimated HR for those with more than or equal to 25% central C4d was 2.7 (95% CI 1.6–4.5, P<0.001). The two higher risk groups did not differ significantly in risk of graft failure (P=0.534).

Central histopathologic analysis of hematoxylin-eosin-, periodic acid-Schiff-, trichrome-, and silver-stained sections was performed according to the Banff ’07 Classification Scheme modified to include assessment of inflammation and tubulitis in the areas of atrophy and nonatrophic regions (6). Distribution of Banff scores according to C4d status is summarized in Table 2. By the Wilcoxon two-sample test, positive C4d staining correlated with the presence of peritubular capillary infiltrates (P<0.001), acute inflammation (Banff i score, P=0.004), acute tubulitis (t, P=0.026), and a marginal but not significant association with transplant glomerulopathy (Banff cg-score, P=0.051). Positive C4d staining was significantly associated with inflammation in areas of atrophy (P=0.001) and tubulitis in areas of atrophy (P=0.030). Positive C4d staining did not significantly correlate with chronic Banff scores, including interstitial fibrosis (ci), tubular atrophy (ct), vascular sclerosis (cv), arteriolar hyalinosis (ah), or mesangial matrix expansion (mm).

T2-11
TABLE 2:
Selected centrally determined Banff scores as a function of centrally determined C4d positivity (≥10%)

The number of centrally defined C4d-positive biopsies, using a cutoff threshold of 10%, as a function of locally determined primary or secondary clinicopathologic diagnosis is shown in Table 3. By Fisher’s exact test, C4d positivity was associated with a local diagnosis of acute cellular rejection (P<0.001), antibody-mediated rejection (P=0.024), and chronic allograft nephropathy (P=0.039). Acute tubular necrosis was significantly associated with a negative C4d status (P=0.016).

T3-11
TABLE 3:
Local primary or secondary diagnoses by centrally determined C4d status

Twenty-three percent of the 296 biopsies studied were C4d-positive (centrally determined), 12% had acute rejection (Banff i2 t2 or greater), and 12.5% had both acute rejection and positive C4d staining. The presence of positive C4d staining was at least as strong a predictor of adverse outcome as the presence of acute rejection (Fig. 3). Compared with patients with biopsies that were C4d-negative and had no evidence of rejection (neither), HRs for patients with biopsies that were C4d positive, acute rejection positive, or both ranged from 2.4 to 4.0. However, these differences in HRs for the latter three groups were not significantly different.

F3-11
FIGURE 3.:
Time to graft failure in patients (n=296) whose biopsies show positive C4d staining (n=68), acute rejection (n=35), both (n=37), or neither (n=156).

DISCUSSION

Although C4d has been touted as a marker of humoral immunity and adverse outcome (12, 13), interpretation of results is method dependent (immunofluorescence vs. immunoperoxidase, and cutoffs for positivity ranging from “focal” to “diffuse”). Studies have suggested that immunofluorescence detection of C4d is more sensitive than peroxidase-based methods, but it may be impractical to obtain an additional biopsy core for immunofluorescence studies at some centers. Furthermore, the clinical utility of various cutoff points for positive C4d staining, particularly by immunoperoxidase, has not been established previously.

In this study, κ values for agreement between central and local determination of C4d were fair (0.40 for a cutoff of 10%) and did not appreciably change when higher cutoff values for centrally determined C4d positivity were used. These κ values are similar to those obtained in a study comparing immunofluorescence and immunoperoxidase detection of C4d, which were performed on the same biopsy cases (κ 0.3) (9). Based on a maximum chi-square search, a cutoff of 10% for central C4d positivity best predicted graft failure or death. The cross-validated estimate of the HR of central C4d positivity using a 10% cutoff was 2.66 (95% CI 1.68–4.21, likelihood ratio chi-square=19.9, 1 df, P<0.001). The association of C4d positivity with graft failure or death became less significant as higher cutoff thresholds for centrally determined C4d were used; there was no significant association between C4d positivity and graft loss when a threshold of 50% was used. Those with central C4d between 10% and 24% (inclusive) had a significantly higher risk of failure than those with less than 10% C4d and furthermore had similar risk of graft failure compared with those with more than or equal to 25% C4d, strengthening the assertion that relatively low levels of C4d are associated with a higher risk of graft failure. There was no significant association between local C4d determinations and outcome, as defined by time to graft failure or death (P=0.121). This may be due to the center-to-center variation in method to detect C4d and the lack of a validated threshold for determination of C4d positivity.

One limitation of our study is an inability to investigate whether a lower cutoff than 10% would have significantly more or less predictive power than the currently selected cutoff of 10% because of insufficient numbers of events among individuals with central C4d more than 0% and less than 10%. This question may become open to analysis as additional events accrue with continued follow-up.

By central histopathologic analysis, positive C4d scores correlated with the presence of peritubular capillary infiltrates, acute inflammation, acute tubulitis, and inflammation in areas of atrophy. Other investigators have found C4d positivity in approximately 30% to 40% of biopsies with acute cellular rejection and an association with steroid-resistant acute rejection (12–15). Our data confirm this finding, because the presence of C4d in biopsies demonstrating acute cellular rejection is associated with adverse outcome (3, 14–17). Widespread peritubular capillary staining for C4d in association with peritubular capillary inflammatory infiltrates and glomerulitis is believed to be representative of acute humoral rejection (13, 18). Positive C4d staining has also been identified in 25% to 40% of patients with chronic antibody-mediated rejection, as defined by presence of donor-specific antibody, double contouring of glomerular capillary loop basement membranes (transplant glomerulopathy), mononuclear inflammatory cell infiltrates within peritubular capillaries, and arterial intimal thickening with mononuclear cell infiltrates (5, 17, 19–21). In patients with transplant glomerulopathy, the presence of peritubular capillary staining for C4d has been associated with adverse outcome (22). However, the prevalence of C4d staining in the presence of transplant glomerulopathy is only 25% to 40% (22, 23).

Histopathologic correlates of positive C4d staining did not correlate with chronic Banff scores, including interstitial fibrosis, tubular atrophy, vascular sclerosis, arteriolar hyalinosis, or mesangial matrix expansion. One possible explanation for the lack of correlation of C4d staining with chronic lesions is that there is a considerable variability of C4d staining in sequential biopsies, with both late appearance and disappearance of C4d staining observed in serial biopsies obtained from the same patient (19). Nevertheless, a positive C4d stain in patients with chronic allograft dysfunction, when present, is a strong marker of adverse outcome. In a retrospective analysis of 80 patients with chronic allograft nephropathy but no evidence of transplant glomerulopathy, peritubular capillary staining for C4d was associated with a worse outcome than C4d-negative biopsies (24).

On the basis of these considerations, we conclude that immunoperoxidase detection of C4d, with a cutoff threshold of 10%, predicts adverse graft outcome in this cohort of patients with graft dysfunction. In patients with acute cellular rejection, positive C4d staining may predict adverse outcome and may provide the basis for alternative therapeutic strategies to target a humoral immune response. Although a positive C4d predicted adverse outcome in patients with chronic histopathologic alterations, C4d was not a sensitive marker for the presence of transplant glomerulopathy. Future studies are needed to develop an optimal model that predicts graft failure, using a cutoff value of 10% for C4d positivity in association with other biomarkers of chronic graft dysfunction. Nevertheless, validation of C4d as a definitive marker using paraffin-embedded material is an important first step in the development of therapeutic protocols to improve long-term graft function.

MATERIALS AND METHODS

Patients and Enrollment

Patients were enrolled in the cross-sectional cohort of the Deterioration of Kidney Allograft Function Study, a multicenter study to define clinical and histopathologic features that predict adverse outcome in renal biopsies performed for new-onset late graft dysfunction (10). Recipients were eligible for enrollment if transplanted before October 1, 2005, having a baseline serum creatinine less than 2.0 mg/dL as of January 1, 2006, and subsequently developing new-onset dysfunction, defined as a more than or equal to percent increase in serum creatinine, or new-onset proteinuria (albumin/Cr ratio >0.2 or protein/creatinine ratio >0.5) leading to an allograft biopsy. Enrollment occurred at the time of the biopsy. The study is registered at www.clinicaltrials.gov (NCT00270712). An institutional review board approval was obtained at all participating sites.

This study included patients with late-onset graft dysfunction, as defined earlier, for which both local and central C4d data were available. Local C4d determinations were performed in conjunction with standard histopathologic analysis of the clinically indicated renal biopsy and were reported as negative or positive using locally established cutoff points. Unstained slides were submitted for central C4d staining. Serum creatinine, urine protein and creatinine, demographic information, current immunosuppression therapy, intervening illnesses, and date and cause of graft loss were collected from all enrolled patients every 6 months.

Immunoperoxidase Stain for C4d

Central C4d staining of formalin-fixed, paraffin-embedded tissues from the clinically indicated renal biopsies was performed by immunoperoxidase using a polyclonal anti-C4d antibody (Alpco Diagnostics, Salem, NH) after antigen retrieval (25). To facilitate consistency, slides were batched and stained on a Dako autostainer (Dako, Carpinteria, CA). C4d stains were read blindly, without clinical or pathologic information. The percentage of peritubular capillaries staining positively for C4d was recorded.

Histopathologic Analysis

Allograft biopsies were read by the local pathologist, and the local pathologic diagnosis was used to guide clinical care and immunosuppressive management. Representative sections were submitted to a central laboratory for analysis. All biopsies were interpreted by the central pathologist in a blinded fashion, without knowledge of C4d status. In addition to the “standard” Banff classification scheme (6), a simplified semiquantitative estimate of inflammation in areas of atrophy was developed and scored as follows: 0=inflammation in less than 10% of atrophic regions; 1=inflammation in 10% to 25% of atrophic regions; 2=inflammation in 26% to 50% of atrophic regions; and 3=inflammation in more than 50% of atrophic regions.

Statistical Analysis

Agreement between local and central C4d positivity was assessed by Kappa analysis, proportion of observed agreement, and proportions of specific agreement for C4d+ positivity and for C4d+ negativity (26). Methods of survival analysis were used to analyze time-to-event outcomes. Nonparametric assessment (Kaplan-Meier) was used to analyze time to graft failure—as defined by return to dialysis, retransplantation, or death—for biopsies staining positively or negatively for C4d. Cox proportional hazard models were used to estimate unadjusted HRs for graft loss by local or central C4d results.

For both the survival methods and the agreement analyses, all observed values of percentage C4d up to 50% were assessed as putative cutoffs for determination of central C4d positivity. Values greater than 50% were not considered because of the paucity of observations with C4d above this value (41/296).

Selection of the optimal cutoff for determination of central C4d positivity was made using the maximum chi-square search, following Mazumdar et al. (27). Briefly, the optimal cutoff for C4d positivity is determined by the largest chi-square value (smallest P value) for the likelihood ratio test in a univariate model of C4d positivity determined by the putative cutoff. Twofold cross-validation was used to obtain a HR estimate and P value for the optimal cutoff to correct for the multiple testing used in the maximum chi-square search. The predictive performance of the final model was assessed in terms of discrimination by Harrell’s c-index (11). The calibration of the final model was assessed by bootstrap estimation of the bias because of overfitting at 1- and 2-year postbiopsy. Both the c-index and the calibration were corrected for overfitting (“optimism”) through bootstrap estimation with 200 replications.

Associations between C4d staining and histologic features were assessed by the two-sample Wilcoxon test. Associations between C4d staining and local diagnoses were assessed by Fisher’s exact test. Two-sample cross-validation, computation of the c-index, and calibration estimates were performed using the R (28), using the RMS package of Harrell (29). All other analyses were performed using SAS/STAT software, version 9.2 of the SAS System for Unix.

ACKNOWLEDGMENTS

The DeKAF Team consists of Arthur Matas, M.D., University of Minnesota; Fernando Cosio, M.D., Mayo Clinic; Joseph Grande, M.D., Mayo Clinic; Sita Gourishankar, M.D., University of Alberta; Philip Halloran, M.D., University of Alberta; J. Michael Cecka, Ph.D., University of California (UCLA); Lawrence Hunsicker, University of Iowa; David Rush, M.D., University of Manitoba; Robert Gaston, M.D., University of Alabama; Roslyn Mannon, M.D., University of Alabama; Bertram Kasiske, M.D., Hennepin County Medical Center; John Connett, Ph.D., University of Minnesota; Robert Leduc, Ph.D., University of Minnesota; Behzad Najafian, M.D., University of Minnesota; Gretchen Crary, M.D., Hennepin County Medical Center; Kim Solez, M.D., University of Alberta; Ramesh Nair, M.D., University of Iowa; Ian Gibson, M.D., University of Manitoba; and William Cook, M.D., University of Alabama.

REFERENCES

1. Feucht HE, Mihatsch MJ. Diagnostic value of C4d in renal biopsies. Curr Opin Nephrol Hypertens 2005; 14: 592.
2. Wang R, Wang H, Chen J, et al. C4d deposition in allograft renal biopsies is an independent risk factor for graft failure. Nephrology 2009; 14: 527.
3. Herzenberg AM, Gill JS, Djurdjev O, et al. C4d deposition in acute rejection: An independent long-term prognostic factor. J Am Soc Nephrol 2002; 13: 234.
4. Racusen LC, Colvin RB, Solez K, et al. Antibody-mediated rejection criteria—an addition to the Banff 97 classification of renal allograft rejection. Am J Transplant 2003; 3: 708.
5. Mauiyyedi S, Pelle PD, Saidman S, et al. Chronic humoral rejection: Identification of antibody-mediated chronic renal allograft rejection by C4d deposits in peritubular capillaries. J Am Soc Nephrol 2001; 12: 574.
6. Solez K, Colvin RB, Racusen LC, et al. Banff 07 classification of renal allograft pathology: Updates and future directions. Am J Transplant 2008; 8: 753.
7. Ludovico-Martins H, Silva C, Teodoro WR, et al. Analysis of different staining techniques for c4d detection in renal allograft biopsies. Transplant Proc 2009; 41: 862.
8. Troxell ML, Weintraub LA, Higgins JP, et al. Comparison of C4d immunostaining methods in renal allograft biopsies. Clin J Am Soc Nephrol 2006; 1: 583.
9. Seemayer CA, Gaspert A, Nickeleit V, et al. C4d staining of renal allograft biopsies: A comparative analysis of different staining techniques. Nephrol Dial Transplant 2007; 22: 568.
10. Gourishankar S, Leduc R, Connett J, et al. Pathological and clinical characterization of the ‘troubled transplant’: Data from the DeKAF Study. Am J Transplant 2010; 10: 324.
11. Harrell FE Jr, Lee KL, Califf RM, et al. Regression modelling strategies for improved prognostic prediction. Stat Med 1984; 3: 143.
12. Crespo M, Pascual M, Tolkoff-Rubin N, et al. Acute humoral rejection in renal allograft recipients. I. Incidence, serology and clinical characteristics. Transplantation 2001; 71: 652.
13. Mauiyyedi S, Crespo M, Collins AB, et al. Acute humoral rejection in kidney transplantation. II. Morphology, immunopathology, and pathologic classification. J Am Soc Nephrol 2002; 13: 779.
14. Poduval RD, Kadambi PV, Josephson MA, et al. Implications of immunohistochemical detection of C4d along peritubular capillaries in late acute renal allograft rejection. Transplantation 2005; 79: 228.
15. Banasik M, Boratynska M, Nowakowska B, et al. C4D deposition and positive posttransplant crossmatch are not necessarily markers of antibody-mediated rejection in renal allograft recipients. Transplant Proc 2007; 39: 2718.
16. Kawamura N, Tomita M, Hasegawa M, et al. Complement C4d deposition in transplanted kidneys: Preliminary report on long-term graft survival. Clin Transplant 2005; 19(suppl 14): 27.
17. Ranjan P, Nada R, Jha V, et al. The role of C4d immunostaining in the evaluation of the causes of renal allograft dysfunction. Nephrol Dial Transplant 2008; 23: 1735.
18. Gibson IW, Gwinner W, Brocker V, et al. Peritubular capillaritis in renal allografts: Prevalence, scoring system, reproducibility and clinicopathological correlates. Am J Transplant 2008; 8: 819.
19. Regele H, Bohmig GA, Habicht A, et al. Capillary deposition of complement split product C4d in renal allografts is associated with basement membrane injury in peritubular and glomerular capillaries: A contribution of humoral immunity to chronic allograft rejection. J Am Soc Nephrol 2002; 13: 2371.
20. Vongwiwatana A, Gourishankar S, Campbell PM, et al. Peritubular capillary changes and C4d deposits are associated with transplant glomerulopathy but not IgA nephropathy. Am J Transplant 2004; 4: 124.
21. Sis B, Campbell PM, Mueller T, et al. Transplant glomerulopathy, late antibody-mediated rejection and the ABCD tetrad in kidney allograft biopsies for cause. Am J Transplant 2007; 7: 1743.
22. Kieran N, Wang X, Perkins J, et al. Combination of peritubular c4d and transplant glomerulopathy predicts late renal allograft failure. J Am Soc Nephrol 2009; 20: 2260.
23. Shimizu T, Ishida H, Shirakawa H, et al. Clinicopathological analysis of transplant glomerulopathy cases. Clin Transplant 2009; 23(suppl 20): 39.
24. David-Neto E, Prado E, Beutel A, et al. C4d-positive chronic rejection: A frequent entity with a poor outcome. Transplantation 2007; 84: 1391.
25. Matas A, Leduc R, Rush D, et al. Histopathologic clusters differentiate subgroups within the nonspecific diagnoses of CAN or CR: Preliminary data from the DeKAF Study. Am J Transplant 2010; 10: 315.
26. Fleiss J, Levin B, Paik MC. Statistical methods for rates and proportions [ed. 3]. Hoboken, John Wiley & Sons 2003.
27. Mazumdar M, Smith A, Bacik J. Methods for categorizing a prognostic variable in a multivariable setting. Stat Med 2003; 22: 559.
28. The R Development Core Team. R: A language and environment for statistical computing version 2.11.0. Vienna, Austria, R Foundation for Statistical Computing 2010. Available at: http://cran.r-project.org/doc/manuals/refman.pdf.
29. Harrell FE Jr. RMS: Regression Modeling Strategies, R package verson 2.1-0. Available at: http://CRAN.R-project.org/package=rms. Accessed April 22, 2009.
Keywords:

C4d; Graft survival; Immunoperoxidase

© 2010 Lippincott Williams & Wilkins, Inc.