recipients of liver transplants. Previous research in recipients of organs other than livers has shown that many factors can influence the cost of transplantation, ...
Predictors of the Cost of Liver Transplantation Robert S. Brown, Jr,* John R. Lake,* Nancy L. Ascher,† Jean C. Emond,† and John P. Roberts† Background. Orthotopic liver transplantation (OLT) is a highly effective but costly therapy for end-stage liver disease. However, there are limited data on the demographic and clinical variables that affect cost. We undertook a preliminary study using multiple regression techniques to analyze factors that influence the cost of OLT. Methods. Patient and demographic data, including laboratory values and charges for all liver transplantations performed between June 1992 and June 1993 were analyzed (n 5 111). Linear regression with standard and log-transformed values was performed by using STATA software (Stata Corporation College Station, TX). Independent variables included in the analyses were age, sex, United Network for Organ Sharing (UNOS) status, primary versus retransplantation, liverkidney transplantation, and laboratory parameters of both liver (aspartate aminotransferase, AST; alkaline phosphatase; bilirubin; albumin; and prothrombin time) and kidney (blood urea nitrogen, BUN; creatinine) function. An F-toremove strategy was employed with a signifi-
O
rthotopic liver transplantation (OLT) has advanced markedly as a field over the past decade. Survival rates that were 30% at 1 year in 1980 are now 80% to 90% in most good programs.1 Improvements in immunosuppression, antiviral prophylaxis, and the management of other common post-transplant complications continue to improve patient and graft survival, as well as post-transplant quality of life.2 With this improvement has come a logarithmic increase in the number of patients considered for OLT. With increased waiting time and increased number of patients on the waiting list, the severity of illness in From the Department of Medicine* and the Liver Transplant Program, University of North Carolina at Chapel Hill, NC; and the Department of Surgery and the Liver Transplant Program, University of California, San Francisco, CA. Supported in part by Glaxo Institute for Digestive Health Healthcare Advancement Award (RSB). Address reprint requests to Robert S. Brown, Jr, MD, MPH, Medical Director, Liver Transplant Program, Division of Digestive Diseases, CB 7080, University of North Carolina, Chapel Hill, NC 27599-7080. Copyright r 1998 by the American Association for the Study of Liver Diseases 1074-3022/98/0402-0009$3.00/0
170
cance level set at P 5 .05. Results. The full model with 12 variables explained 37% of the total variation in charges. When one excludes variables that did not have a significant impact on cost, the remaining significant variables were BUN and UNOS status 1. The final model was Charges (US$) 5 3,407 3 BUN 1 74,474 3 status 1 1 102,662 This model accounted for 29% of the total variability with BUN accounting for the vast majority (26%). Conclusions. Renal function is the most important predictor of cost of OLT (P F .001). UNOS status 1 further increases cost, but other hospitalized patients have similar costs when one controls for other clinical variables. The degree of liver impairment is less important in predicting cost. Copyright r 1998 by the American Association for the Study of Liver Diseases
many patients at the time of transplantation has increased. The impact of this shift on cost of transplantation is unknown. Although there is considerable pressure through managed care contracting and global fee setting to make transplantation more cost effective, there are very little data on the factors that affect the cost of transplantation because most research efforts to date have dealt with decreasing morbidity and mortality among recipients of liver transplants. Previous research in recipients of organs other than livers has shown that many factors can influence the cost of transplantation, including disease origins, comorbidities, duration of disease, pretransplant treatments, and type of immunosuppressive therapy.3,4 There have been very few studies that assess the factors that drive the costs of OLT. Limited earlier studies investigated the impact of diagnosis or immunosuppressive therapies used on the overall cost of OLT,5,6 but did not control for other factors, such as severity of disease. A comprehensive study was performed by Evans et al in 1988 involving about 70% of the programs active in that year. He reported median hospital charges for OLT was $145,776 (all amounts in US
Liver Transplantation and Surgery, Vol 4, No 2 (March), 1998: pp 170-176
Predictors of the Cost of Liver Transplantation
dollars), with a hospital stay of 33 days.7 The costs were higher for younger patients, nonwhites, second transplants, transplants ending in death, and transplants performed on intensive care unit (ICU) patients, in nonteaching hospitals, or in new or large transplant centers. This study was limited by the lack of a clinical database; thus, most of the analyses were confined to demographic factors. Wiesner et al reported that in patients with primary biliary cirrhosis (PBC) or primary sclerosing cholangitis (PSC), the cost of OLT at the Mayo Clinic increased with increasing disease severity,8 as determined by the Mayo prognostic scoring system.9 The most recent economic analysis of liver transplant recipients to date was performed on a subset of patients participating in the US MultiCenter FK506 Study.10 The authors found that the hospital charges accrued by patients randomized to FK506-based immunosuppression were almost $20,000 less than that accrued by patients randomized to cyclosporine-based immunosuppression ($141,569 to $122,279) in two well-matched groups. The differences in cost could be largely accounted for by differences in the incidence and severity of allograft rejection between the two groups of patients. This study showed that relatively modest clinical benefits (i.e., a decrease in the incidence of acute rejection from 68% to 67% and steroid-refractory rejection from 33% to 22%) can lead to striking economic benefits. These studies represent the first steps toward identifying the drivers of the cost of liver transplantation, together with the magnitude of their effect; however, each study has deficiencies. For example, neither the Evans nor the Wiesner study contained pretransplant clinical information. The US FK506 Study Group included only relatively low risk patients in its study, which limited the ability to comprehensively examine the effects of pretransplant status on the cost of OLT. To date, no study has carefully examined the interaction between pretransplant variables, such as renal insufficiency, on the resources used for transplantation. Moreover, there has not yet been an attempt to derive and validate a model that can accurately predict the cost of OLT from the pretransplant clinical variables.
Materials and Methods Patient Population This study represents a retrospective analysis of all patients receiving a liver transplant or a combined liver
171
and kidney transplantation who were discharged from the University of California, San Francisco (UCSF), Medical Center between July 1, 1992, and June 30, 1993. One hundred fourteen patients who underwent 119 transplants were included. Patients receiving living related liver transplants (n 5 4) were excluded because they represent a unique operation and donor source and used different cost centers than patients undergoing cadaveric transplantation. Patients who received a second liver transplantation during the same hospitalization (n 5 2) were excluded because of difficulty determining the cost of each transplant operation. One patient who received a living related transplant who received retransplantation during the same hospitalization was eliminated because of having a living related transplant, and the number of transplants was then decremented to account for the retransplant. One patient who left the operating room without a transplant and subsequently died was also excluded. The final study population was 109 patients who underwent 111 transplants. Patient and demographic data were extracted from the UCSF Liver Transplant Database. Laboratory data obtained within 24 hours preceding surgery were obtained from the hospital laboratory database. Charges were obtained from the billing department and were broken down by cost centers.
Outcome Variables The primary outcome variable is cost. For this study, the charges billed by the institution were used as a proxy for the outcome variable, cost. The use of charges is a widely used method of economic analysis because of difficulties in assessing true costs, i.e., the degree of resources used by the institution to provide the service.3,4 Because costs are usually estimated by the institution, charges are not prone to bias because of the method of assigning institutional cost. Finally, although they are not generalizable to other centers for absolute numbers, we were comparing relative charges, which would likely parallel relative costs. Hospital charges were obtained from the hospital accounting system for the index hospitalization. For patients who were admitted to the hospital at the time of transplantation, all charges associated with their hospital stay were included. For patients hospitalized at the transplant center before transplantation, charges from the day before the transplant date until discharge were included because this included all the charges for preoperative assessment while minimizing potential bias toward higher costs among inpatients. The charges include donor organ acquisition expenses but exclude physician professional fees. Length of stay (LOS) was used as a secondary outcome variable.
Predictor Variables Independent variables were selected based on whether they were likely to influence cost or outcome of transplantation. Only variables that were objective and had been
172
Brown et al
measured prospectively were included in the models. The predictor variables analyzed age, sex, UNOS status, primary versus retransplantation, single organ versus liver-kidney transplantation, and laboratory parameters of both liver (aspartate aminotransferase, AST; alkaline phosphatase; bilirubin, albumin, and prothrombin time) and kidney (blood urea nitrogen, BUN; creatinine) function. All the objective measures in the Child-Pugh score were included individually, and thus the ChildPugh score was not included separately. Sex, multiple organ transplantation, and retransplantation were classified as dichotomous variables. For stratification of illness, UNOS codes for listing for liver transplantation were used. (The order of the status codes was changed by UNOS in April 1995. The analysis here uses the new status code order but does not reflect the changes made in January 1997.)11 These codes relate to the functional status of the patients as follows: Patients who are at home and requiring only routine medical care are listed at a status 4; patients who are at home but require frequent medical care are listed as status 3; patients who are continuously hospitalized are listed as a status 2; and patients who are in the ICU and are expected to live less than 7 days are listed at status 1. The patients’ status at the time of transplantation was used in the analyses and was modeled as a dummy variable. All other variables were treated as continuous variables.
Table 1. Selected Demographic Features of the Study Population Age (mean) 44 Sex (M/F) 57/54 Diagnoses (N) Alcoholic liver disease 21 Chronic active hepatitis C 16 Primary biliary cirrhosis 13 Cryptogenic cirrhosis 12 Chronic active hepatitis B 8 Fulminant hepatic failure 8 Autoimmune hepatitis 7 Rejection 6 Primary sclerosing cholangitis 5 Extrahepatic biliary atresia 4 a1-Antitrypsin deficiency 3 Subacute hepatic failure 3 Miscellaneous 5 UNOS Status at Time of Transplantation (N, %) Status 1 24 (21) Status 2 32 (29) Status 3 55 (50) Status 4 0 (0)
Predictive Models for Charges Statistical Analyses Linear regression was performed with standard and log-transformed values and the STATA software package. An F-to-remove strategy was employed for all regression analyses. Survival was calculated by the Kaplan-Meier method. The significance level was defined as a 5 .05.
Results One hundred nine patients receiving 111 transplants were studied. Two patients were retransplanted during different hospitalizations within the study period and were included, as were 8 patients who had received a transplant before the study period and were undergoing retransplantation during the study period. Nine patients underwent combined liver-kidney transplantation. The indications for transplantation and UNOS status codes at the time of transplantation are seen in Table 1. The mean age at transplantation was 44 years (6SD). Eleven patients were 16 years old or less at the time of transplantation. The 1-year survival for all patients was 92%. The median charge for all patients was $153,772, with a range of $88,981 to $1,324,030. The median length of stay was 19 days, with a range of 7 to 226.
Predictive models were derived by multiple linear regression. The full model with all 12 variables explained 37% of the total variation in charges. The b-coefficients and the P values for each predictor variable are shown in Table 2. BUN and creatinine had significant interaction and covariation (correlation coefficient, .64) and, hence, were not useful when employed together in the model. Because BUN explained a much larger proportion of the variation in charges (25% v 3.6%), BUN was used in all further models. UNOS status 1 accounted for 10% of the variation in charges when it was in the model alone but only 3% when BUN was controlled for. Compared with patients who were UNOS status 3, patients who were UNOS status 1 had charges that were $132,000 higher when BUN was not controlled for and $74,000 higher when the impact of BUN was controlled for. Patients who were UNOS status 2 had average charges $31,000 higher than UNOS status 3, but this difference was not statistically significant (P 5 .38). None of the measures of liver function significantly predicted cost, although there was a trend toward increased cost at higher levels of albumin (P 5 .12). Age, gender, liver-kidney transplantation, and retransplantation did not affect cost when UNOS status
173
Predictors of the Cost of Liver Transplantation
Table 2. Impact of Clinical and Demographic Factors on Cost of Transplantation: Full Multivariate Model Variable
b-Coefficient ($)
95% Confidence Interval
P Value
Female sex Age (per year) Retransplant Liver/kidney Tx Blood urea nitrogen Aspartate aminotransferase Alkaline phosphatase Bilirubin Prothrombin time Albumin UNOS status 1 UNOS status 2 Creatinine
11,097 21,120 233,725 52,672 3,950 247 264 631 3,001 47,364 38,388 24,111 254,340
(249891, 72084) (23158, 917) (2141105, 73,655) (275043, 180387) (2029, 5870) (2138, 44) (2169, 240) (21539, 2800) (22830, 8832) (26733, 101460) (253891, 130667) (250275, 98496) (2113054, 4374)
.719 .278 .535 .415 ,.001 .309 .225 .565 .310 .085 .411 .522 .069
and BUN were controlled for in the analysis, although the number of patients retransplanted or receiving liver-kidney transplantation was small (n 5 10 and 9, respectively). When one excludes variables that did not have a significant impact on cost, the remaining significant variables were BUN and UNOS status 1. The final model was
ues and length of stay as the outcome variable were not significantly different (data not shown). The relationship between expected and actual costs is shown in Figure 1. Examples of cost prediction with this model are shown in Table 3.
Charges (US$) 5 3407 3 BUN
Discussion
1 74,474 3 status 1 1 102,662 This model accounted for 29% of the total variability, with BUN accounting for the vast majority (26%). Repeat analyses with log-transformed val-
Figure 1. Actual versus predicted charges for transplantation (in $1000s). Line shows the best fit regression line (R 2 5 .29, P F .0001).
These analyses show the important impact of renal function and UNOS status on the cost of transplantation. It also shows the significant interaction between these two variables. Although UNOS Status 1 explained 10% of the variation in cost in a
174
Brown et al
Table 3. Predicted Average Cost of Transplantation by UNOS Status and BUN (US$) UNOS Status
BUN (mg/dL)
Status 1
Status 3
10 50 100
$211,205 $347,480 $517,823
$136,731 $273,006 $443,349
univariate model, this decreased to only 3% when BUN was independently controlled for in the analysis, and the absolute dollar impact was decreased by approximately 44%. Because charges were used, the absolute increase in cost cannot be determined from these analyses, although the relative costs are likely to parallel the relative charges. The impact of severity of illness on cost of transplantation has been previously demonstrated. Wiesner et al showed that for patients with PBC or PSC the cost of transplantation rose progressively with increasing Mayo risk score.8 The correlation coefficient was .62 (R2 5 .36, P , 0.0001). Although they did not look specifically at UNOS status, one could predict that patients with more advanced disease, and thus higher Mayo risk scores, would be more likely to be UNOS status 1. However, this study did not control for other clinical data, such as renal function, and was limited to patients with PBC or PSC. It does not state the reason patients were in the ICU or proportion of patients with renal failure. Evans et al also demonstrated the impact of UNOS status on cost.7 Their study analyzed a random sample of up to 25 patients from 72% of the 74 active transplant programs in 1988. These centers performed 47% of the 1680 liver transplant procedures that year, but the sample available for analysis was 416 patients. In their data the cost of ICU-bound patients was $166,013, twice as high as patients at home ($75,529, P , .05). Length of stay was also longer, 50 versus 25 days. The increase in cost seen for patients who are UNOS status 1 was similar to that seen in our data. Evans data, however, similar to the Wiesner data, did not include a clinical database, and thus the impact of renal function could not be analyzed. In addition, multivariate analyses were not performed in this study. The study of Lake et al could not analyze the impact of renal dysfunction or UNOS status as patients who were UNOS status 1 or who had renal
failure were excluded from the US Multicenter FK 506 Trial. The impact of renal function on the cost of transplantation has not been analyzed previously. We have shown, however, that in the large NIDDK Liver Transplantation database, renal dysfunction is associated with increased resource utilization as measured by ICU and hospital LOS.12 In this study, patients with fulminant hepatic failure and renal dysfunction (defined as creatinine . 1.6 mg/dL or dialysis) had an ICU and hospital LOS 5 days longer than those with normal renal function did. For patients with chronic liver disease, LOS was 20, 26, and 15 days for patients with renal insufficiency not on dialysis, those on dialysis, and those without renal insufficiency, respectively. Thus, in this study the differences in resource utilization were larger with more profound renal failure. Whether BUN would be a better predictor of resource utilization was not analyzed in this study. Unfortunately, cost data were not available at the time of the study to analyze dollar figures; however, it is likely that dollar figures parallel LOS, which was seen in our analyses and has been demonstrated in other studies.7,13 The reason for increased cost for patients with renal dysfunction are likely multifactorial. Resource utilization for patients who require dialysis are likely due to both longer LOS as well as the resource utilization associated with dialysis itself. In addition, the average cost in patients with renal insufficiency would be increased by the higher organ acquisition costs for those who received liver-kidney transplantation. Second, impaired renal function is likely a marker for more severe acute or chronic comorbidity. Other forms of organ failure are generally viewed as contraindications to transplantation, and thus patients with renal failure may be the most severely ill of patients undergoing transplantation. In addition, renal dysfunction is common in patients with sepsis, severe gastrointestinal bleeding, or marked liver dysfunction, all of which are likely associated with increased cost of transplantation. Lastly, renal dysfunction per se presents risks for increased cost. Patients with uremia are more likely to have bleeding complications after transplantation, and due to their relative immunosuppression, they may be more prone to posttransplant infection. In addition, lower rates of rejection, which has been seen among patients with renal insufficiency, would likely lower costs in patients with renal insufficiency, making the rela-
Predictors of the Cost of Liver Transplantation
tionship between cost and renal function stronger, because any confounding by rejection rates would likely bias the results toward the null. Although we had hypothesized that renal function would be a major determinant of cost, we were surprised to find that BUN was a better predictor of cost than creatinine. There are two possible reasons for this result. First, it is possible that BUN is measuring more than renal function and that these other factor(s) may also influence cost. Gastrointestinal bleeding, dehydration, corticosteroid use, and muscle breakdown can all increase BUN values. These factors may also influence cost. However, it appears unlikely that the influence of these factors is more profound than that of renal function. The impact of renal function on cost is also supported by the fact that elevated creatinine values were also associated with increased cost when BUN was not in the model, although this relationship was less strong. It is also possible that, contrary to what is seen in healthy individuals, BUN may be a better measure of renal function than creatinine in patients with end-stage liver disease. This may be true due to (1) decreased muscle mass in patients with end-stage liver disease and (2) interference by high levels of bilirubin with the assay used to measure creatinine. This last phenomenon is likely quite important because renal dysfunction is most common in patients with the most advanced liver disease and thus the highest bilirubin levels. Although currently there are assays that mitigate this interference, they were not used at the time of the study. Ideally, direct measurement of glomerular filtration rate would be the best way to assess the impact of renal function on cost, but it is not routinely measured before transplantation and was not available for the patients in this study. Other limitations of our study include the fact that measurements were made immediately before transplantation. Laboratory values can change rapidly, and thus the values may not reflect the steady state for the patient, particularly for hospitalized patients. However, this approach was necessary to avoid bias, because the patients who were transplanted as UNOS status 3 (i.e., from home) would often have only one value immediately prior to OLT. Thus, we standardized the approach for all patients to avoid bias in the hospitalized patients. Any misclassification of renal function that results is likely to be nondifferential and thus would tend to bias toward the null, strengthening our conclusions. Second, our database did not include all
175
possible predictors of cost. Measures of functional status (e.g., Karnofsky score) and comorbidities such as diabetes, nutritional status, or other demographic or clinical factors may also influence cost. Although this may have explained an additional portion of the variation in cost it is unlikely that these factors would significantly confound the relationship between UNOS status or BUN and cost. First, much of the renal dysfunction is acute and thus unlikely to be related to these other factors. Second, our data are in keeping with other reports in the literature. Finally many of these factors (especially Karnofsky score and nutritional status) are related to the severity of liver impairment, which, at least as measured by albumin, prothrombin time, and bilirubin, did not significantly impact cost. Although it is impossible to exclude some degree of confounding, the strength of the relationship makes it unlikely to be significant. Finally, retransplantation, which has been shown to increase cost,7 did not have such a significant impact in our study. This may be due to a small number of retransplants in our dataset resulting in insufficient power to detect a difference. Alternatively, this may be due to the fact that we did multivariate analysis that controlled for UNOS status, which had not been done previously. Thus, much of the difference in charges seen in prior studies may be due to higher UNOS status among retransplanted patients. It is important to note that retransplantation did not have a significant impact in our study in the univariate analysis. Studies with higher numbers of retransplants will be needed to fully address this question. In summary, this report shows the profound impact of renal dysfunction and, to a lesser degree, UNOS status on the cost of transplantation. It is likely that the stronger influence on cost of UNOS status seen in prior studies reflects in part the increased frequency of renal dysfunction in patients who are UNOS status 1. UNOS status 2 did not have a significant impact on cost once renal function was controlled for. Future research that makes use of a larger dataset is needed to determine the reasons for increased cost in this group of patients and more importantly, whether therapies that improve renal function (e.g., transjugular intrahepatic portosystemic shunts or TIPS for hepatorenal syndrome, liver-kidney transplantation) will lower the cost of OLT in this group of patients in dire need of transplantation.
176
Brown et al
References 1. Starzl TE, Demetris AJ, Van Thiel D. Liver transplantation. N Engl J Med 1989;329:1014. 2. Lake JR. Changing indications for liver transplantation. Gastroenterol Clin North Am 1993;22:213. 3. Showstack J, Katz P, Amend W, Bernstein L, Lipton H, O’Leary M, et al. The effect of cyclosporine on the use of hospital resources for kidney transplantation. N Engl J Med 1989;321:1086. 4. Showstack J, Katz P, Amend W, Salvatierra O. The association of cyclosporine with the 1-year costs of cadaver-donor kidney transplants. JAMA 1990;264: 1818. 5. Staschak S, Wagner S, Block G, Van Thiel Dh, Jain A, Fung J, et al. A cost comparison of liver transplantation with FK 506 or CyA as the primary immunosuppressive agent. Transplant Proc 1990;22:47. 6. Pageaux GP, Souche B, Perney P, Calvet B, Delande G, Fabre JM, et al. Results and cost of orthotopic liver transplantation for alcoholic cirrhosis. Transplant Proc 1993:1135. 7. Evans RW, Manninen DL, Dong FB. An economic analysis of liver transplantation. Costs, insurance coverage, and reimbursement. Gastroenterol Clin North Am 1993;22:451. 8. Wiesner RH, Porayko MK, Dickson ER, Gores GL, La
9.
10.
11.
12.
13.
Russo NF, Hay JE, et al. Selection and timing of liver transplantation in primary biliary cirrhosis and primary sclerosing cholangitis. Hepatology 1992;16:1290. Dickson ER, Grambsch PM, Fleming TR, Fisher LD, Langworthy A. Prognosis in primary biliary cirrhosis: Model for decision making. Hepatology 1989;10:1. Lake JR, Gorman KJ, Esquivel CO, Wiesner RH, Klintman GB, Miller CM, et al. The impact of immunosuppressive regimens on the cost of liver transplantation: Results from the U.S. FK506 multicenter trial. Transplantation 1995;60:1089. Annual Report of the Scientific Registry of Transplant Recipients and the Organ Procurement and Transplantation Network—Transplant Data: 1988-1993. UNOS, Richmond, VA, and the Division of Organ Transplantation, Bureau of Health Resources Development, Health Resources and Services Administration, US Department of Health and Human Services, Rockville, MD, 1994. Brown RS, Jr, Lombardero M, Lake JR. Outcome of patients with renal insufficiency undergoing liver or liver-kidney transplantation. Transplantation 1996;62: 1788. Brown RS, Jr, Lake JR, Emond JC, Bachetti P, Randall HB, Ascher NL, Roberts JP. Impact of surgical complications following liver transplantation on resource utilization. Arch Surg 1997; 132:1098.
