Steroid withdrawal after renal transplantation: a retrospective cohort study

Study design and data sources

We conducted a retrospective open cohort study to investigate the effect of steroid
withdrawal at numerous points in time following kidney transplantation on patient
and graft survival using data from three sources: the OEstereichische (Austrian) Dialysis
and Transplant Registry (OEDTR), the EUROTRANSPLANT database, and the Vienna Kidney
Biopsy Registry, as previously done by our group 26], 27]. The OEDTR was established by the Austrian Society of Nephrology in 1970 and has
almost complete follow-up – only 0.6% of all Austrian residents on renal replacement
therapy have been lost since 1990. The OEDTR contains thoroughly extracted data from
the original medical records in which the original data was assessed at the time of
the follow-up visit by the responsible physician 28]. Data provided by the OEDTR included recipient age and sex, date of transplantation,
primary renal diagnosis, the presence of comorbidities at transplantation and annually
throughout follow-up, panel reactive antibodies, patient and graft survival, and immunosuppression.
Use of immunosuppressive medication was reported quarterly in the first year after
transplantation and annually thereafter. Induction treatment consisted of IL-2 antibodies.
We retrieved data on donor age and type (deceased or living), the number of human
leukocyte antigen mismatches, and cold ischemia time from the EUROTRANSPLANT database,
which was established in 1968 and holds complete entries of organ donor characteristics
from transplants that have been performed in the EUROTRANSPLANT region to which Austria
belongs 29]. Information on biopsy confirmed acute rejection defined according to Banff 97 criteria
were extracted from the Vienna Kidney Biopsy Registry, which is composed of standardized
descriptions of renal histopathology of native and transplant kidney biopsies 30].

All end-stage renal disease patients recorded in the OEDTR who received their first
single-organ, ABO-compatible kidney transplant between January 1, 1990, and December
31, 2012, with an initial steroid-containing immunosuppressive regimen were included
in this study and followed up until November 19, 2014.

The exposure of interest, ‘steroid withdrawal’, is a dichotomous time-dependent variable.
Outcome variables were functional graft loss and all-cause death with functional graft.
We performed cause-specific analyses of either event type. Graft survival time was
defined as the time from transplantation until either permanent return to dialysis
treatment or second transplantation, counting death or end of follow-up as censored
observations. Patient survival time was defined as the time from first kidney transplantation
until death, censored for graft loss, and end of follow-up.

Statistical analyses

Continuous variables are expressed by mean and standard deviation, categorical variables
are presented by frequencies and percentages.

To investigate the long-term effects of steroid withdrawal at various time points
after kidney transplantation, we chose the landmarking approach, by which causal effects
can be inferred under the usual assumptions of propensity score analyses 31]. Specific points in time following engraftment, so called landmark times, were pre-defined
at 3-month intervals until 10 years after engraftment. At each of these landmark times,
study participants were classified as either ‘steroid withdrawal’ or ‘steroid maintenance’
depending on steroid treatment status within the preceding time interval (first day
after previous landmark time until current landmark time). Once patients were classified
as ‘steroid withdrawal’ at a specific landmark time they were excluded from consideration
at subsequent landmark times (Additional file 1: Figure S1).

Confounding by indication, caused by any potential difference in covariates between
patients withdrawn from steroids and patients maintained on steroids that could have
influenced the decision to withdraw or maintain steroids at a given landmark time,
was addressed by introducing a landmark-time-dependent propensity score for matching
steroid-maintenance patients to steroid-withdrawal patients at each landmark time
32]–34]. First, we computed a logistic regression model to calculate the probability of steroid
withdrawal or maintenance for each patient in the risk set at each landmark time based
on the most recent values of confounding covariates (Additional file 1: Figure S2). As a second step, we matched patients withdrawn from steroids to patients
maintained on steroids based on these individual propensity scores at each landmark
time to generate a cohort of steroid withdrawal and steroid maintenance patients whose
only remaining difference, in theory, is the steroid treatment status at a given landmark
time. Using these matched study cohorts, we computed cause-specific cumulative incidence
functions for the competing event type graft loss and death with functional graft
and compared them between steroid treatment groups at specific landmark times. To
summarize differences in graft loss and mortality following steroid withdrawal or
maintenance at different time points, we estimated a landmark-stratified Cox supermodel
using all matched study cohort data from all landmarks. In this supermodel, we included
an interaction of steroid withdrawal status with landmark time, smoothing transitions
between neighboring points in time using restricted cubic splines with knots at 1,
2, and 4 years 35]–37]. This approach yielded the landmark-specific, propensity score-adjusted hazard ratios
and 95% confidence intervals from which the time point with the largest benefit from
discontinuation of steroids could be determined. Assessment of the proportional hazards
assumption was conducted using a log minus log plot based on the cause-specific cumulative
hazard estimated by the Kaplan–Meier method with weights according to the matching
procedure. To deal with missing data in the covariates used for the propensity score,
multiple imputation was employed 38], 39]. For steroid withdrawal status (the exposure of interest), no imputations were necessary.
To determine whether biomarkers of cardiovascular risk improved after steroid withdrawal,
we compared serum cholesterol, fasting glucose, the number of antihypertensive drugs,
and body mass index before and after steroid withdrawal (Additional file 1).

A 95% confidence interval excluding parity or a two-sided P value less than 0.05 was considered as indication for statistical significance. For
all analyses, the software R (version 3.2.1) was used. The study was approved by the
Ethics Committee of the Medical University Vienna (1359/2014) and performed in accordance
with the Declaration of Helsinki. The detailed statistical methods are outlined in
Additional file 1.