Although the implementation of a new donor organ allocation system in 2002 has improved efficiency in organ allocation and reduced wait-list mortality, a significant gap remains between the number of liver transplantation (LT) candidates waiting for an organ and the number of available organs. The scarcity of donated organs highlights the responsibility of the transplant community as a steward of these lifesaving resources to ensure their best use. In that context, the identification of factors that can be modified to optimize outcomes after LT is critically important.
Numerous recipient and donor characteristics have been studied as determinants of outcomes after LT. For example, the severity of a recipient's hepatic decompensation, as captured by the Model for End-Stage Liver Disease (MELD) score, has a significant impact on the post-LT outcome. Patients with far advanced liver cirrhosis and multiorgan failure are likely to encounter a difficult postoperative course and a higher risk of mortality and graft loss.[2, 3] Furthermore, certain donor characteristics as well as the circumstances of organ donation and transplantation affect the outcome of a graft. These variables have been summarized as the donor risk index (DRI).
Despite the demonstrable statistical significance of donor and recipient variables, those variables account for a relatively limited amount of the variability in post-LT outcomes.[5, 6] This is in part because random, unpredictable events that compromise graft survival occur not uncommonly in the postoperative period. However, the degree to which these events occur depends in part on the center at which the procedure is performed. Transplant centers differ in surgical and medical expertise, in the selection policy for donors and recipients, and in organ availability. In addition, the geographic location of a center may play an important role. Organ availability differs by region, and even within a region, liver procurement and transplantation practices vary among the different donor service areas (DSAs), the units of organ distribution.
Currently, the Centers for Medicare and Medicaid Services (CMS) monitors posttransplant mortality data at each transplant center. In identifying centers that do not meet set performance standards, CMS uses donor and recipient factors to derive estimates for expected center-specific graft and patient survival rates. An underperforming center, if mitigating factors are not adequately addressed, may be subject to citation or termination. However, the extent to which factors beyond the transplant center's control (eg, the region and DSA to which it belongs) contribute to deviations from the expected remains unclear. Furthermore, policy proposals have been made to include outcomes after LT in the consideration for donor organ allocation. The effectiveness of such a policy would be dependent on the accuracy with which post-LT outcomes could be predicted. If the LT center plays an important and quantifiable role in graft survival, a true outcome-based allocation model may take into account where LT is going to be performed in addition to recipient and donor characteristics.
We hypothesized that the center of transplantation has a substantial impact on graft failure after LT even after one accounts for relevant geographic factors and donor-, recipient-, and transplant-related factors. Our present analysis of Organ Procurement and Transplantation Network (OPTN) data under the current allocation system based on the MELD score shows that the LT center has an independent effect on graft survival after adjustments for known predictors. We present the magnitude of the effect associated with the LT center in comparison with that of other important determinants of LT outcomes, including the region and DSA in which the center is located.
PATIENTS AND METHODS
Data Source and Elements
We used data submitted to the OPTN for all adult (age ≥ 18 years) primary LT recipients from January 1, 2005 to December 31, 2008 (n = 18,358). We excluded patients who were missing data for calculating the DRI (n = 1257). Recipients undergoing simultaneous liver-kidney transplantation and retransplantation were excluded. The data set provided by the OPTN included de-identified center information, including the DSA and the region to which it belonged, in addition to a comprehensive array of variables about the recipient and the donor. The investigators were blinded to the identity of each center and DSA in the data. The study was approved by the institutional review board.
The primary outcome variables were patient death and graft failure within 1 year, and the primary predictor variable was the LT center. Graft failure was defined as patient death or retransplantation within 1 year. Patients alive with a functioning graft 1 year after LT were censored. We considered covariates that might confound the association between LT center and graft failure. In the context of assessing the center effect, we selected a priori the recipient's age, sex, and diagnosis and the MELD score and its components at LT as variables representing most of the recipient's condition and the DRI as a variable accounting for transplant- and donor-related factors.[4, 8-16] The DRI consists of donor factors (age, race, height, and cause of death) as well as transplant-related characteristics (eg, donation after cardiac death, receipt of a partial or split liver, cold ischemia time, and regional and national sharing) and has been shown to be a predictor of graft failure after LT that is independent of recipient factors.[4, 17] We also examined the role of geography by considering the effect of DSA and region. Additional sensitivity analyses were conducted through the inclusion of all of the variables in the risk adjustment modeling used by the Scientific Registry of Transplant Recipients in addition to these selected variables.
The main tool for assessing the center effect was a mixed effects, proportional hazards regression analysis (also called a hierarchical or random effects model). In our mixed effects model, the LT center was considered a random effect variable, whereas other covariates (ie, recipient and donor variables) were fixed effect variables. In estimating the center effect, we first created a model consisting only of fixed variables and determined the effects of those variables on graft failure. We then added the LT center variable to the model, and the effect of the transplant center on graft failure after an adjustment was determined by a comparison of the log likelihood of separate models with and without the center variable. The amount of variability between centers was expressed in the model as the standard deviation (SD) for the hazard ratio (HR). By definition, the average HR is 1.0, whereas individual centers may have HRs much smaller or larger than 1.0, the distribution of which can be represented by the SD.
In order to gauge the magnitude of the center effect, we compared it to that of known recipient and donor predictors of graft failure such as recipient age, MELD score, and DRI. To establish the relative impact of the center versus known predictors, the effects of all factors were standardized by a comparison of the risks of graft failure from observations at the 75th percentile (high risk) and the 25th percentile (low risk) for each of the variables. Furthermore, we compared the variation in the percentage of graft failure within 1 year across centers. Finally, to determine whether the variation in the rates of graft failure between centers was specific to certain patient subgroups with respect to donor and recipient characteristics, we examined the center effect by the quartiles of the MELD score and DRI.
Further analyses were undertaken to understand the nature of the center effect. First, we explored whether the center effect may in part be attributable to geographic disparities in organ distribution. This was accomplished by the repetition of the analysis, but this time, the model was stratified by the 11 United Network for Organ Sharing regions and then by DSA. The other factor that we considered was the annual number of transplants at the center (ie, the center volume), which has been suggested to play an important role in graft survival.[12, 15, 16] Because centers contributed data for multiple calendar years, for this analysis, each center was characterized by its average number of transplants per year. Finally, we conducted a sensitivity analysis to determine whether the center effect changed after status 1 recipients and recipients diagnosed with hepatocellular carcinoma were excluded so that we could examine a uniform cohort of patients with chronic liver disease.
Between 2005 and 2008, there were 14,654 primary LT procedures performed at 115 unique centers that met the study criteria. The mean annual number of transplants per center was 60 (SD = 47). Ninety-four centers (82%) were present for all 4 years.
The median 1-year graft failure rate was 12.5% [interquartile range (IQR) = 9.5%-17.4%]. The median 1-year patient survival rate was 88.3% (IQR = 84.4%-92.1%). Table 1 shows the characteristics of transplant centers by the quartiles of the incidence of graft failure. The rate of graft loss within 1 year was 5.9% in the first quartile (Q1) and 20.2% in the fourth quartile (Q4). Overall, 66 centers (57%), 33 centers (29%), and 16 centers (14%) annually performed 0 to 49, 50 to 100, and >100 transplants, respectively. There were significant differences in the proportion of non-Caucasian recipients (P < 0.001), MELD score at LT (P < 0.001), diagnosis of chronic hepatitis C (P < 0.001), DRI (P < 0.001), and donor age (P < 0.001) between the quartiles.
Table 1. Center Characteristics by the Quartiles of the Incidence of Graft Loss: 2005-2008
Table 2 addresses the effects of recipient and donor factors that are associated with graft loss. In the fixed effect model, recipient age, MELD score at transplant, and DRI were all significant predictors of graft failure (P < 0.001). The addition of center to the model did not change the effect size or statistical significance of these factors. Recipient age (per 10-year increase), MELD score (per 5-U increase), and DRI (per 0.2-U increase) were associated with 20%, 17%, and 13% increased risks of graft failure, respectively. The incorporation of the individual components of the DRI instead of the composite score did not change the results (data not shown). More importantly, after adjustments for these factors, the LT center was a significant predictor of graft failure at 1 year (P < 0.001). The SD of the HR was 0.29, and this indicates that centers that are 2 SDs away from the mean (HR = 1.0) would have a 58% lower or higher risk of graft failure. Similar results were obtained upon an examination of patient death (SD = 0.29) instead of graft failure. Furthermore, the addition of other donor-, recipient-, and transplant-related factors currently included in the Scientific Registry of Transplant Recipients risk adjustment model did not appreciably change the results (see Supporting Fig. 1).
Table 2. Center of Transplantation and Graft Failure Within 1 Year With Adjustments for Donor-, Recipient-, and Transplant-Related Characteristics
In Fig. 1A, the magnitude of the center effect is compared with the effects of other predictors of graft failure. The vertical axis of the figure represents the relative risk (RR) of graft failure between the 75th and 25th percentiles of each variable after adjustments for other covariates. For example, the RR of graft loss for recipients at the 75th percentile (MELD score = 28) was 1.47-fold higher than that for recipients at the 25th percentile [MELD score = 16, 95% confidence interval (CI) = 1.36-1.59]. This may be contrasted to the RR of graft failure at a high-risk center at the 75th percentile, which was 1.49 times that at a low-risk center at the 25th percentile (95% CI = 1.36-1.65). Thus, when the center effect is assessed in this fashion, it is as large as the effect of the recipient MELD score or DRI and is larger than the effect of recipient age (Fig. 1A). In comparison, the magnitude of the DSA effect was 1.44. A similar trend was observed when we compared values at the 90th and 10th percentiles (see Supporting Fig. 2), although, as expected, the RR was higher for all categories.
Figure 2 provides further insight into the magnitude of the center effect. In this figure, the variation in the percentage of graft failure within 1 year across the centers is presented after adjustments for the covariates shown in Table 2. For an average recipient receiving an average donor organ, the risk of graft failure within 1 year ranged from 8.3% at the lowest risk center to 24.4% at the highest risk center (an approximately 3-fold difference between the 2 extreme ends). Because of the skewed distribution, more than 50% of the centers (n = 69) had a lower risk of graft failure than the mean (1-year graft failure = 14.0%). In contrast, 3.5% of the centers (4/115) had an expected rate of 1-year graft failure higher than 20%. The exclusion of patients with hepatocellular carcinoma and status 1 did not change the results (Supporting Fig. 3). Similar results were obtained from an analysis of 1-year patient survival, with the risk of patient death ranging from 7.6% at the lowest risk center to 22.1% at the highest risk center (Supporting Fig. 4).
We examined whether the center effect varied with recipient and donor characteristics (Fig. 3A,B). The center effect on the risk of graft failure was relatively uniform across all quartiles of the MELD score. The variability in graft failure seemed to be larger in the highest DRI quartile, but this was within chance variation (P = 0.99). For comparison, the DSA effect on the risk of graft failure is also provided (Fig. 3C,D).
Having demonstrated that the LT center has an important effect on graft survival, we then considered what might potentially determine the center effect. Figure 4 shows the correlation between the center volume and the risk of graft failure. In contrast to reports from a previous era, the center volume had no apparent impact on graft failure once donor, transplant, and recipient characteristics were taken into account. We further examined whether there was a differential risk in the early postoperative period. The variations in the risk of graft failure within the first 3 months (SD = 0.34) and in the period more than 3 months after LT (SD = 0.29) were similar (Supporting Fig. 5).
Finally, because of its pervasive impact on donor availability and quality as well as overall transplant practice, we considered the region and the DSA as important sources of center-specific LT outcomes. First, region was included in the model as a stratification variable: the center remained a significant predictor of graft failure (P < 0.001), although the effect became slightly smaller (SD = 0.25), and this indicated that region explained a small part of the center effect. Next, DSA was incorporated as a stratification variable, and this further reduced the center effect; however, the residual center effect remained substantial (SD = 0.20) and statistically significant (P < 0.001). After stratification by DSA, the RR of graft failure at a high-risk center at the 75th percentile was 1.30 times that at a low-risk center at the 25th percentile (95% CI = 1.16-1.48; Fig. 1B).
In this work, we show that there is a large degree of variability in graft outcomes depending on the LT center and that the center effect is independent of known predictors of graft loss. Because of the number of centers performing LT in the United States, it is difficult to succinctly summarize the center effect. One way of describing the data is shown in Fig. 1A, according to which the increase in the risk of graft failure at a center at the 75th percentile versus one at the 25th percentile (49% increase, RR = 1.49) was at least as large as the difference in the risk of graft failure between a recipient with a MELD score of 28 and another with a MELD score of 16 (RR = 1.47). This center effect on graft outcomes, larger than we expected at the outset of the analysis, was independent of any other variables that may affect graft survival.
To our knowledge, this is the first analysis of the degree to which the geographic disparities in organ availability (and thus quality) partially account for center-specific outcomes. It is likely that a substantial part of the effect attributed to geography represents unmeasured confounding; for example, DSA may quantify unmeasured donor factors, and center may quantify a combination of unmeasured transplant, donor, and recipient characteristics.[18, 19] Indeed, the concordance statistic (the c statistic) was 0.625 for a model without center or DSA, 0.651 for a model including DSA, and 0.657 for a model including center. We note that after we accounted for relevant donor-, recipient-, and transplant-related factors, a measurable proportion (approximately a third) of the variability in LT outcomes at each center was attributable to the geographic location of the center (most directly DSA). This has important implications. First, if we lessen geographic disparities in organ distribution, the performance of some transplant centers may improve. Second, if center-specific outcome data are used by CMS to identify programs that are not performing well and hence may be subject to substantial corrective action, the geographic unit of distribution must also be taken into account. However, it is also important to remember that even after we accounted for geography, the center effect remained substantial (30% increase between the 25th and 75th percentile centers, RR = 1.30). It is clearly important that optimal LT results be achieved at all centers. We acknowledge that center-specific reports from the Scientific Registry of Transplant Recipients recognize that inherent variability exists among centers; however, our work is the first to quantify the magnitude of the center effect and place it in the context of known determinants of graft failure after LT. Our data suggest that an initial step toward that goal is to accurately assess center-specific variability and then to understand its source. Because the center effect is independent of major recipient and donor variables, we postulate that at least some part of the variability results from differences in the quality of care of LT recipients. Because of the large number of hospitals performing LT, it is likely that there is a substantial degree of variability in the structures of LT programs and in the processes of care for individual recipients. For example, a number of surgical factors, including the individual surgeon's experience and the fatigue of the surgical team on a given day, have been associated with graft survival.[20, 21] Furthermore, centers vary in their medical expertise in the perioperative management of complications (eg, infection or critical care expertise) and in their adherence to follow-up care over the long run.[22, 23] The identification of best practices at low-risk centers that can be incorporated at high-risk centers may allow for improved outcomes across centers.
Post-LT outcomes also vary with wait-list management and organ acceptance practices at individual centers. For example, systematic differences exist between centers with respect to their acceptance of high-risk donor organs for patients at different MELD score levels.[15, 17, 24] However, our data show that intercenter differences in outcomes persisted after adjustments for the DRI in the donor and the MELD score in the recipient. We saw a weak trend showing that the outcome variability was larger for high-DRI organs, and this may indicate that the outcomes of marginal organs depend on the center's experience and expertise in managing problems associated with those organs. Finally, our analysis challenges policy proposals to take into account post-LT outcomes in organ allocation decisions (eg, the net benefit model). If a significant portion of the graft outcome is determined by where the transplant occurs, it follows that such a policy must include the center in the allocation decision. However, it would be practically untenable to preferentially allocate organs to candidates at centers with a low risk of graft failure.
We believe that our analysis has accurately estimated the effect of the LT center on graft outcomes. The analysis used national data including all centers from the recent past. Careful statistical methods were employed to calculate intercenter and intracenter variability and to account for known potential confounders such as region, DSA, and transplant volume.[5, 16, 25] However, there may be limitations to the study. First, we measured the center effect, but we were not able to completely elucidate the source of the center effect. We postulate that in addition to surgical, medical, and nursing expertise, there are systematic factors that may influence the quality of care at a transplant center. They may include the effectiveness of multidisciplinary collaboration, systematic efforts at quality improvement, utilization of standardized protocols, and adequate physical and information technology infrastructure. The OPTN data lack detailed information such as this, and a dedicated prospective investigation would be necessary to delineate the critical, center-specific determinants of graft outcomes. Second, although we incorporated well-documented donor and recipient factors that affect graft survival, it is possible that unmeasured donor or recipient characteristics may be in part responsible for the estimated center effect. For example, donor steatosis is a well-known predictor of graft survival, but it is not readily available in most databases. Furthermore, the DRI as a surrogate may not capture all important donor effects. We were unable to adjust for certain recipient characteristics (eg, the presence of hepatopulmonary syndrome and treated hepatitis C versus untreated hepatitis C at the time of LT) that may affect posttransplant outcomes. We do suspect that even if there are some unrecorded confounding variables in the analysis, it is unlikely that they would negate the center effect.
In summary, the LT center is a significant determinant of graft failure and provides a partial explanation for the disparities in outcomes after LT. The center effect remains significant and practically meaningful even after adjustments for region, DSA, and relevant donor and recipient variables. Although resolving the geographic disparities in access to quality organs may reduce a part of this variability, studies to explicitly address what drives the variability in center-specific outcomes are needed to optimize the outcomes of LT in the United States as a whole.