Patients and methods
NYU SAMPLE Registry and Biorepository
Patients were enrolled in the NYU SAMPLE Bioregistry after signing consent. Inclusion in the SLE cohort required that a patient meets four or more American College of Rheumatology (ACR)15 or systemic lupus international cooperating clinics (SLICC)16 criteria and be 15 years of age or older.
Patient selection
Within the SLE cohort of 763 patients, 325 (43.3%) have history of kidney disease, as defined by the ACR criteria.15 For this study we identified 129 patients with documented renal biopsy. Laboratory values and pathology information were collected at the time of biopsy, and all patients were followed longitudinally from the time of biopsy up to 4 years. Ninety-one patients met the inclusion criteria of having their first annual follow-up visit within a window of ±1 month of 12 months from biopsy, hereafter referred to as 12 months. The objective of the study is to identify the best predictors at 12 months of an adverse renal outcome (ARO) by 48 months from biopsy. ARO is defined as doubling of serum creatinine, as creatinine >4 mg/dL if initial >2.5 mg/dL, or need for renal replacement therapy with either dialysis or transplant. Four patients who were not ARO-free at 12 months were excluded. Patients were required to have at least one additional follow-up visit after their first annual follow-up visit generating 83 patients. The dsDNA samples were measured in different laboratories; values were calculated relative to the upper limit of each laboratory. Since only one patient had an upper limit of 60 at 12 months, the dsDNA could not be standardised within their laboratory and the patient was omitted. Eighty-two patients remained for the final analyses.
Statistical analysis
Distributions of age at biopsy and the continuous markers measured at 12 months (ie, albumin, uPCR, haematuria, dsDNA, C3 and C4) were summarised using medians, ranges and boxplots. The categorical demographic covariates were described using frequency distributions. Scatter plots and Spearman correlation coefficients were used to evaluate the pairwise associations between the continuous markers measured at 12 months. dsDNA at 12 months was standardised for this and subsequent analyses.
Kaplan-Meier (KM) curves were used to estimate ARO-free survival over time in 82 patients. Time to ARO was calculated from 12 months until an ARO occurred or the patient was censored (eg, last data time point at least second follow-up visit, but prior to 48 months). Time to ARO was represented in month units and was rounded to the nearest whole month. A 95% CI for the KM curve was calculated. In addition, the 36-month (from the first annual follow-up) cumulative ARO rate was estimated from the KM curve.
The significance of each of the markers (albumin, uPCR, RBC, C3, C4 and standardised dsDNA at 12 months), as a predictor of ARO, was evaluated both in univariate and exploratory multivariable Cox proportional hazards models. Additionally, albumin and uPCR at 12 months were evaluated in a bivariable model with and without interaction. Hazard ratios (HRs) with corresponding 95% CIs were calculated.
From the Cox models we generated ordered cut-off points for the linear predictor and generated receiver operating characteristic (ROC) curves at 48 months that correspond to the time of interest post biopsy.17 The area under the curve was estimated and the Youden Index was calculated and maximised in order to find the optimal cut-off point of the linear predictor of ARO at 48 months. For a marker with a negative association with ARO, values above the cut-off point predicted a non-ARO at 48 months, whereas marker values below or equal to the cut-off point predicted an ARO at 48 months. For a marker with a positive association, the prediction is in the reverse. This covariate cut-off point corresponded to a particular sensitivity and specificity. PPV and NPV were calculated under an assumed prevalence of ARO at 48 months of 20%.
All analyses were performed using SAS 9.4 and R version 3.0.2 software.