Table 3.

Multivariable Logistic Regression Models Predicting Graft Failure

A. Three variable model for predicting graft failure in BKVN patientsa.
Variable	Coefficient	Odds Ratio (95% Confidence Intervals)	P Value
Constant	−10.59
Urinary cell PAI-1 mRNA	1.01	2.8 (1.12 – 6.78)	0.03
Serum Creatinine	0.59	1.8 (0.55 – 6.02)	0.33
Biopsy B/C	1.65	5.2 (0.38 – 70.38)	0.22
Model pseudo-R2 valueb: 0.4142, P=0.0015.
Composite Score Equation: −10.59 + 1.01Ln(PAI-1/18S) + 0.59(serum creatinine mg/dL) + 1.65(biopsy BKVN stage B/C[1] or A[0]).

B. Two variable model for predicting graft failure in BKVN patientsc.
Variable	Coefficient	Odds Ratio (95% Confidence Interval)	P value
Constant	−10.61
Urinary cell PAI-1 mRNA	1.14	3.1 (1.28 – 7.54)	0.01
Serum Creatinine	0.72	2.1 (0.63 – 6.77)	0.23
Model pseudo-R2 valueb: 0.3682, P=0.001.
Composite Score Equation: −10.61 + 1.14Ln(PAI-1/18S) + 0.72(serum creatinine mg/dL).

^aMultivariable logistic regression models were constructed including variables associated (P<0.1) with graft failure by bivariate analysis (SDC Table S2). Since Spearman’s rank-order correlation analysis showed that the mRNAs are highly related with each other (SDC, Table S4), only one mRNA measure was added to each logistic model that included serum creatinine level and BKVN biopsy stage. Among the 7 models created with the addition of different mRNAs to serum creatinine and biopsy stage (SDC Table S5), the combination of creatinine, biopsy stage and PAI-1mRNA was the best predictor. The coefficient, odds ratio and the P value for each variable included in this optimum model are provided in Table 3A.

^bR² value indicates the percent of variation in the dependent variable (graft failure) that can be explained by the predictor variables (e.g., PAI-mRNA level) in the model. It ranges from 0 to 1.0 with a value of 1.0 indicating the perfect predictor. The equivalent terminology for the term R² used in the linear regression model is pseudo-R² in the logistic regression model;

^cTable 3B shows the coefficient, odd ratios and P values for the noninvasive prediction model comprised of serum creatinine level and urinary cell PAI-1 mRNA level and without the biopsy results. None of the other mRNAs included in the prediction models shown in SDC Table S5 was an independent predictor (P<0.05) in the multivariable logistic regression analysis.