Abstract
 Top of page
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
Impressive renal allograft survival improvement between 1988 and 1995 has been described using projections of halflives based on limited actual follow up. We aimed, now with sufficient follow up available to calculate real halflives.
Real halflives calculated from KaplanMeier curves for the overall population as well as subsets of repeat transplants and African Americans recipients were examined.
Real halflives were substantially shorter than projected halflives. As a whole, halflives have improved by about 2 years between 1988 and 1995 as compared to the earlier projected 6 years of improvement. The improvement seems to be driven primarily by the improvement in graft survival of retransplants. First transplants showed a cumulative increase in graft survival of less than 6 months.
Projected halflives are a risky estimation of longterm survival especially when based on short actual follow up. First transplant survival has only marginally improved during the early years of post transplant follow up while no significant improvement in longterm survival could be detected between 1988 and 1995. Redirection of attention from early endpoints towards the process of longterm graft loss may be necessary to sustain early gains in the long term.
Introduction
 Top of page
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
Survival analysis in renal transplantation have traditionally been made by KaplanMeier estimates, because of the necessity to evaluate data on outcomes before all patients reach a certain followup time (1). Each patient contributes to the survival estimate only up to the time they have available follow up, so that with increasing length of follow up progressively fewer patients contribute to the survival estimate. In this sense KaplanMeier survival estimates are only an estimation of the real survival outcome of the entire study population. In fact in order to make somewhat reliable deductions from KaplanMeier survival estimate comparisons, sufficient patients have to be at follow up at the point of analysis.
A calculated halflife is a projection of survival which has many assumptions in itself, most importantly that the slope over time follows a predictable function (2). As opposed to survival estimates by KaplanMeier analysis (where at least one patient has to have reached the last follow up), projected halflives are based on predicted survival probabilities beyond any real observations available. Moreover it is important to remember that in general when halflives are calculated, only a fraction of patients have reached the complete follow up, which the calculated slope is based on. In most cases a halflife uses the estimated survival at a given follow up, as opposed to the real survival (survival based on complete follow up information), and than projects the longterm survival based on this data. Therefore, the resulting projection is based on an estimate with two possible sources of bias and variability; first the incomplete follow up and secondly the variability of the slope. In general, extreme caution is to be advised when making these extrapolations and longterm projections as they result in easy to understand numbers, which for that very reason are easily over interpreted. In other words, when a population is described to have a life expectancy of 14 years, none of the above described potential variability is expressed in that number. As these numbers are simply projections, based on assumptions which may or may not be born out, particular care must be taken when using these statistics. Therefore, without sufficient understanding of the variability incorporated into such projections, these hard numbers can be particularly misleading.
As we now have sufficient follow up to describe actual halflives (based on actual follow up) for cadaveric transplants until 1995, we analyzed the US Scientific Renal Transplant Registry (SRTR) database for renal transplants executed between 1988 and 1995 to compare previously projected halflives (3) and real halflives, and to obtain hard numbers to see how the survival of renal transplants has evolved in that period of time. This analysis should provide a better understanding of longterm results by evaluating if shortterm improvements have really translated into a longterm advantage in renal allograft survival.
Methods
 Top of page
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
We analyzed 57 961 solitary deceased donor and 19 976 living donor renal transplants between 1988 and 1995 from the SRTR database to examine the longterm effects of the era of transplant on patient outcomes. KaplanMeier survival curves were used to assess the unadjusted halflives and survival rates for overall and death censored graft survival in this cohort. Subsequently, we examined these effects restricted to first transplants and various additional strata from this era.
In order to quantify the years of graft life gained over the span of the analysis, area under the KaplanMeier survival curves were calculated (4). Cumulative graft years per patient were calculated using the summation of the trapezoidal areas between each followup time point under the survival curves for the applicable strata. The difference between the plots areas with equivalent follow up were interpreted as the graft years gained between applicable comparisons. Calculations were extended to 8 years posttransplant which represented the maximum period of follow up for the last year of study.
Retransplants were determined using the indicator variables supplied by the SRTR in the registry database. First year acute rejection was determined by any indication of rejection or treatment for rejection within the initial transplant hospitalization, 6month, or 1year followup records for transplant recipients. Last followup information was available through July 2003.
Cox proportional hazard models were used to estimate the effect of year of transplant adjusted for potential confounding factors; models were corrected for donor and recipient demographic factors, length of dialysis time, donor cause of death, primary diagnosis of recipient, panel reactive antibody (PRA) level, cytomegalovirus (CMV) matching, cold ischemia time, and human leukocyte antigen (HLA)A, B, and DR matching.
Analyses were performed on SAS software (v9.0, Cary, NC, USA).
Results
 Top of page
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
The overall graft survival KaplanMeier plots for first deceased donor transplant recipients are displayed in Figure 1. The curves displayed separation in the early posttransplant period, but seemed to narrow with additional followup time resulting in relatively similar halflives. Graft loss not due to patient death also displayed relatively minor variation in this period. The KaplanMeier survival curves for death censored graft survival with the applicable 8year estimated survival rates are displayed in Figure 2. For all transplants (including retransplants), we found the real halflives of the univariate survival curves by the year of transplant and compared them to previously published (3) projected halflives (Figure 3). The actual halflives were significantly lower, particularly for more recent years as the trend for improvement in the projections did not appear to correspond with extended amount of follow up for this patient population. Of note, the actual halflife for 1995 was estimated as this cohort only reached 51% survival for the latest followup data at the time of this analysis. The total per patient average years with a functioning transplant, as calculated by the area under the survival curve, by year of transplant through 8 years of follow up for first living and deceased donor recipients are illustrated in Table 1. The difference in graft years between the extremes of 1988–1995 accounted for a cumulative impact in the 8year follow up of 4.7 months per patient (Figure 4). Repeating the area under the curve calculations for the multivariate models revealed a wider disparity from 1988 to 1995 relative to the univariate equivalent (Figure 5), the adjusted impact of the year of transplant was approximately 8.4 months cumulatively for the 8year follow up. Retransplant rates significantly decreased over this era; from 1988 to 1995 the retransplant rates, respectively, were 23.8%, 20.9%, 19.5%, 19.2%, 17.3%, 14.2%, 14.2%, and 14.2%. Survival rates for retransplants did appear to significantly improve associated with the more proximate year of transplant (Figure 6).
Table 1. Eightyear univariate and multivariate cumulative graft years per patient in deceased donor transplant recipients by year of transplant* Year of Transplant  Univariate Estimates  Multivariate Estimates 


1988  5.3  5.2 
1989  5.4  5.4 
1990  5.5  5.5 
1991  5.7  5.7 
1992  5.6  5.6 
1993  5.6  5.7 
1994  5.6  5.8 
1995  5.7  5.9 
The multivariate model effects of the year of transplant did display a mild risk associated with transplants in earlier years. The adjusted relative risks and 95% confidence intervals for overall graft loss by year of transplant for both deceased donor and living transplants are displayed in Table 2. Significant covariates in the deceased donor transplant recipient model are displayed in Table 3. Our examination of the association of the year of transplant in the presence or absence of first year acute rejection did not reveal any specific trend by year of transplant in either strata for first transplant recipients. In a subset analysis we found that the impact of the year of transplant in this era on overall graft survival appeared more impressive in African American recipients relative to Caucasians (Figure 7).
Table 2. Cox proportional hazard model estimates for association of year of transplant and overall graft loss: adult first solitary transplants 1988–1995 Year of transplant  Deceased donor transplants  Living transplants 

Hazard estimate (1995 as reference)  95% confidence interval  Hazard estimate (1995 as reference)  95% confidence interval 

1988  1.359  (1.286, 1.436)  1.097  (0.977, 1.233) 
1989  1.287  (1.218, 1.359)  1.091  (0.976, 1.220) 
1990  1.238  (1.174, 1.306)  1.067  (0.956, 1.192) 
1991  1.176  (1.114, 1.240)  1.074  (0.964, 1.197) 
1992  1.213  (1.150, 1.279)  1.098  (0.987, 1.220) 
1993  1.125  (1.067, 1187)  1.087  (0.981, 1.205) 
1994  1.063  (1.008, 1.121)  0.984  (0.889, 1.090) 
Table 3. Cox proportional hazard model estimates for significant covariates associated with overall graft loss in deceased donor transplant recipients (1988–95) Parameter  Hazard ratio  Confidence interval  Pr > chi square 

CMV D+/R (ref: D/R)  1.074  (1.035, 1.114)  0.0002 
Donor race African American (ref: Caucasian)  1.132  (1.088, 1.178)  <0.0001 
Donor age 60–70 (ref: 17–29)  1.806  (1.708, 1.910)  <0.0001 
HLAA 2 MM (ref: 0 MM)  1.070  (1.029, 1.112)  0.0006 
HLAB 2 MM (ref: 0 MM)  1.067  (1.024, 1.111)  0.0020 
HLADR 2 MM (ref: 0 MM)  1.144  (1.104, 1.185)  <0.0001 
Recipient age 65 + (ref: 17–29)  1.686  (1.572, 1.807)  <0.0001 
Diabetes as primary diagnosis (ref: GN)  1.564  (1.510, 1.619)  <0.0001 
Recipient Female  0.939  (0.915, 0.963)  <0.0001 
Recipient race African American (ref: Caucasian)  1.377  (1.338, 1.419)  <0.0001 
Dialysis time 36 + months (ref: preemptive)  1.137  (1.091, 1.184)  <0.0001 
Donor CV COD  1.041  (1.010, 1.073)  0.0094 
Cold ischemia time (h)  1.027  (1.014, 1.040)  <0.0001 
Discussion
 Top of page
 Abstract
 Introduction
 Methods
 Results
 Discussion
 Acknowledgements
 References
Considerations regarding longterm results after kidney transplantation are often based on projections of early results. Now, with sufficient follow up available to evaluate real halflives in patients transplanted between 1988 and 1995, we compared actual halflives to previously used projected halflives (3).
Based on projected half life calculations, it was assumed and has become part of the current thinking in the transplant community, that the halflives of transplanted kidneys had close to doubled from 1988 to 1995 (3). This effect seems to be largely due to an overestimation by the longterm projection of the survival data available at the time of publication. As evident from the data shown in Figure 3, with now sufficient data available to describe real halflives, between 1988 and 1995 there has been merely a 2year increase in the halflife of transplanted kidneys. We did this analysis combining first and retransplants in order to most closely replicate the earlier halflives estimated as published in the New England Journal of Medicine (3). On closer examination we found that in fact almost the entire improvement in halflives during the mentioned time period was due to the improvement in retransplants. As it is evident from the data displayed in Figure 1 and Table 4 the real halflife for first transplants has only marginally changed between 1988 and 1995. The overall trend toward improved halflives is therefore driven by the improvement in retransplants and by the lower percentage of retransplants in more recent era. In fact the impressive improvement in the halflife of retransplants during the same time period is depicted in Figure 6.
Table 4. Univariate and multivariate halflives (hl) by year of transplant (first deceased donor transplants 1988–1994*) Year of transplant  KaplanMeier HL  Cox multivariate HL 


1988  7.5  7.0 
1989  7.9  7.7 
1990  7.9  7.7 
1991  8.0  7.9 
1992  7.7  7.7 
1993  7.9  8.2 
1994  7.9  8.8 
This is not to say that primary renal transplant survival has not increased at all during this time period. In fact there is a significant yet modest survival benefit early on after transplantation which over time translates into an approximate 5month average gain in first graft survival between 1988 and 1995 as presented in Figure 4. This improvement is far more modest than the originally proposed and widely cited almost doubling in transplant halflives.
We calculated this modest overall survival benefit by comparing the area under the curve of the survival curves in order to get a cumulative estimate as opposed to an assessment at a single point in time. When survival curves diverge early on and subsequently converge as in this case, comparison of the area under the survival curve may be the most accurate estimate of a cumulative benefit or change in outcome.
The very fact that the real yet modest improvement in first renal transplants has occurred during the earlier follow up after transplantation, while halflives are close to identical, explains why extrapolations and projections from early data can lead to significant over estimations. This example documents very well how misleading projected halflives are when based on short follow up.
The other subgroup of patients, who in addition to the retransplants have had a significant improvement over the time period in question, is African Americans (Figure 7). We believe that the reason for this is that in these 2 groups, African Americans and retransplants, there was more room for improvement especially in preventing graft loss secondary to acute rejection. In fact it has been shown repeatedly that improvements in immunosuppressive regimens have a proportionally higher beneficial impact in high risk recipients like African Americans (5–7). Now that sufficient data has accumulated to calculate halflives of transplant kidneys based on real data, it becomes clear that earlier projections of halflives were overestimating the halflives but also implying a trend which is really only true for a fraction of patients.
Overall graft survival has clearly improved significantly, even though only marginally (4.7 months between 1988 and 1995), mainly driven by improvements in early post transplant outcomes. Longterm results in first renal transplant remain largely unchanged between 1988 and 1995. In part this is likely secondary to the increasing risk constellation with higher risk donors and recipients in more recent era. Both the donor and recipient populations are ageing and higher risk patients are being transplanted. In fact in the multivariate analysis correcting for the higher risk of the transplant population in more recent years, we found a slightly more prominent, yet still modest effect on graft survival in terms of halflife (Table 4) and cumulative graft years gained adjusted for risk factors (Figure 5). Given the higher risk recipients and the declining quality of the transplant kidneys over the period of study, the small improvement in halflives can still be seen as a progress. In fact given the fact that correcting for potential confounding variables the increase in halflife is more sustained and also the increment in average graft years gained during the whole period is up to 8.4 months, the increasing risk constellation of the donor recipient population would have lead to worse results if there had not been an advance in the management of transplant patients. This is clearly though a much more modest success as compared to the previously reported radically changing longterm results (3).
It is important to realize that it is difficult to extrapolate late outcomes based on early followup data. As powering clinical trials for differences in longterm graft loss has become more difficult because of the excellent outcomes in kidney transplantation (8), investigators are pushed to use early surrogates for longterm outcomes, like early graft survival as in our example, or even less hard endpoints like renal function and acute rejections (9–11). It is important to remember that it is almost impossible to project longterm outcomes in kidney transplantation based on early trends. In fact trends might not be constant over time and past trends might not apply under changing immunosuppression. Calculated halflives are therefore often misleading especially when long halflives are calculated based on short actual follow up.
A good approximation of a potential benefit in graft survival can be obtained by the method used in this paper, which compares the area under the KaplanMeier survival curve between two groups and calculates the cumulative life years gained over the time of available follow up. Interestingly this method was already described in the original KaplanMeier paper, but has received in kidney transplantation outcomes research relatively little attention (1). Even though this method contains some uncertainty due to the fact that not all patients have reached the end of followup period, at least this method is based on real data, and if used with circumspection, i.e. enough patients who have completed the followup time, can yield useful estimations of differences in cumulative life years saved.
The other important point which is uncovered now with sufficient followup time available is that despite important improvements in the short term we have not really made an impact on the longterm outcomes. Clearly there seem to be two different phases in the follow up of a transplant population, an early phase where acute rejection and technical problems have the major impact on graft survival. This phase has clearly been positively affected and in return the average survival of a transplant kidney has improved to a certain extent. However, there seems to be a second phase where the survival slope is affected by different events, e.g. subacute or chronic rejection, calcineurin inhibitor toxicity, or also recurrent disease.
A shift in attention toward the longterm followup of renal transplant patients and the prevention of the most common causes of late graft loss, like cardiovascular death and chronic allograft nephropathy will have to occur to really make an impact on longterm outcomes.
In summary longterm renal allograft survival was only marginally improved between 1988 and 1995. The majority of this modest improvement was secondary to improvements in the high risk subsets of African American and repeat transplants. Greater attention will need to be focused on longterm events to further improve longterm graft survival.