American Journal of Transplantation 2015; 15: 170–179 Wiley Periodicals Inc.
Copyright 2014 The American Society of Transplantation and the American Society of Transplant Surgeons doi: 10.1111/ajt.12970
Variation in Biliary Complication Rates Following Liver Transplantation: Implications for Cost and Outcome D. A. Axelrod1,*, N. Dzebisashvili1, K. L. Lentine2,3, H. Xiao2,3, M. Schnitzler2,3, J. E. Tuttle-Newhall3 and D. L. Segev4
MELD, Model for End-Stage Liver Disease; NAS, nonanastomotic stricture; OPTN, Organ Procurement and Transplantation Network
Received 26 March 2014, revised 09 June 2014 and accepted for publication 19 July 2014
Section of Transplant Surgery, Dartmouth-Hitchcock Medical Center, Lebanon, NH 2 Center for Outcome Research, Saint Louis University, St. Louis, MO 3 Department of Surgery, Saint Louis University, St. Louis, MO 4 Department of Surgery, Johns Hopkins School of Medicine, Baltimore, MD Corresponding author: David A. Axelrod, [email protected]
Although biliary complications (BCs) have a significant impact on the outcome of liver transplantation (LT), variation in BC rates among transplant centers has not been previously analyzed. BC rate, LT outcome and spending were assessed using linked Scientific Registry of Transplant Recipients and Medicare claims (n ¼ 16 286 LTs). Transplant centers were assigned to BC quartiles based upon risk-adjusted observed to expected (O:E) ratio of BC separately for donation after brain death (DBD) and donation after cardiac death (DCD) donors. The median incidence of BC was 300% greater in the highest versus lowest DBD quartiles (19.0% vs. 5.9%) and varied 250% between DCD quartiles (20.3%–8.4%). Donor and recipient characteristics suggest that high BC centers actually used lower donor risk index organs, fewer split livers and fewer imports (p < 0.001 for all). Transplant at a center in the highest O:E quartile was associated with increased posttransplant mortality (adjusted hazard ratio [aHR] 2.53, p ¼ 0.007) in DCD transplant and increased graft loss (aHR 1.21, p ¼ 0.02) in DBD transplant. Medicare spending was $22 895 (p < 0.0001) higher at centers in highest versus lowest BC quartile. In summary, BC rates vary widely among transplant centers and higher rates are a marker for an increased risk of death, graft failure and health-care spending. Abbreviations: CPT, Common Procedure Terminology; DBD, donation after brain death; DCD, donation after cardiac death; ERCP, endoscopic retrograde cholangiopancreatography; ICD-9, International Statistical Classification of Diseases and Related Health Problems-9; KM, Kaplan–Meyer; LT, liver transplantation; 170
Introduction Wide variation in the rate of procedural complications following complex surgery has been identified among centers performing cardiac, oncologic and general surgical procedures (1–7). Factors associated with higher rates of complications include center volume, surgeon and hospital specialization, and employment of specific technical procedures (e.g. intraoperative cholangiography during laparoscopic cholecystectomy) (8–10). Recent reports have focused on the surgeon’s technical performance using video taped analyses, demonstrating that technical proficiency contributes to postoperative complications (8). These investigations demonstrate a clear association of perioperative complications with higher costs and inferior long-term outcomes for patients, leading to initiatives to reduce or eliminate avoidable complications. Liver transplantation (LT) is, similarly, a technically complex procedure with attendant risks of perioperative complications. Among these events, the development of biliary complications (BCs; leak, stricture and cholangitis) has been strongly associated with inferior clinical outcomes and increased health care spending in single center series, systematic reviews and database analyses (11–21). However, the degree of variation in the incidence of BCs among transplant centers has never been rigorously studied, as BC data are not collected through clinical registries. Large database studies are needed to assess the degree of center-level variation after controlling for previously identified donor and recipient factors that contribute to the development of these complications including the use of donation after cardiac death (DCD) donors (21). Although retrospective chart review with trained data abstractors remains the gold standard for assessing the incidence of clinical complications, such reviews are not feasible on a national basis given existing data sources.
Biliary Complications After Liver Transplant
Health care claims comprise an alternative data source, which can be linked with detailed clinical data from national registries to provide insight in the incidence, outcomes and economic implications of surgical complications. Using these methods, we identified BCs in 11.7% of Medicare recipients of deceased donor liver transplant, which consistent with the rates of stricture (12%) and biliary leak (7.8%) reported in a meta-analysis of more than 11 000 patients reported by Akamatsu et al (18). Patients with BCs identified using health care claims were more than twice as like to likely to experience graft loss (adjusted hazard ratio [aHR] 2.08 [CI: 1.77–2.44]) and posttransplant mortality (aHR 2.21 CI: 1.99–2.46) after adjustment for donor and recipient characteristics including receipt of a DCD allograft. An accurate assessment of variation in the incidence of BCs at the nation’s transplant centers is vital to determining the potential impact of a national quality improvement initiative to reduce the incidence of these events. While prior single center analyses have suggested that the incidence of BCs differs markedly across centers, potentially due to center level practice variation, these data are limited by differences in donor and recipient characteristics (12,18,22). The aims of the current study are to utilize our robust claims analysis methodology to characterize the risk-adjusted variation in the incidence of BCs, assess the differences in the donor and recipient characteristics between high and low complication centers, and the correlate transplant centers’ risk-adjusted frequency of BC with the overall outcome and cost of transplantation.
Definition of BCs BCs were ascertained from posttransplant Medicare billing claims with corresponding International Classification of Diseases 9th Edition Clinical Modification-9 (ICD-9-CM) and Common Procedure Terminology (CPT) codes within 180 days of transplant. The 180-day window was specified to identify patients whose BCs are related to the transplant procedure rather than recurrent disease. Complications were identified using a combination of a relevant biliary diagnosis (e.g. cholangitis, biliary stricture) and an appropriate endoscopic or radiologic procedure (e.g. endoscopic retrograde cholangiopancreatography. BCs were defined at the level of the patient, thus, patients with multiple procedures were counted only once.
Statistical analyses Baseline characteristics of the study cohort with and without BCs were described. Incidence estimates were computed for the overall population and after stratification for transplants of DBD and DCD organs. The expected incidence of BCs for each center was determined using a logistic regression model to predict the likelihood (predicted probability) of BCs, controlling for the center’s donor and recipient characteristics (Supplemental digital content). A binary outcome was used (BC within 180 days or not). Separate models were developed for recipients of DCD and DBD organs. Patient-level predicted probabilities for all recipients at a center were summed, and the number of patients observed to have a BC was divided by the center-level cumulative expected number of affected patients to yield an observed to expected (O:E) ratio of BCs for each center. Centers were excluded from the analysis if data was available for fewer than 10 DBD transplants or 5 DCD transplants. Centers were then grouped into quartiles based on the O:E ratio of BCs, and the center’s O:E quartile assignment was included in the multivariate models described below. Differences in donor and recipient characteristics between the quartiles were assessed using univariate analyses. Statistical significance of the O:E ratio was assessed using a Poisson distribution with the mean centered at the center’s expected incidence, with a p < 0.05 considered significant. Single center results are not reported in accordance with data use agreements with the OPTN and Centers for Medicare & Medicaid Services.
Materials and Methods Data sources and study sample A novel database linking clinical, demographic and claims information from the Organ Procurement and Transplant Network (OPTN), for adult (age >18) patients who received a deceased donor LT between 2002 and 2008 and had linked Medicare billing claims, was created and analyzed for this study. The OPTN maintains records for all solid organ transplant candidates and recipients in the United States including detailed donor and recipient demographic characteristics, graft failure and patient death. Medicare billing claims include diagnostic and procedure codes as well as total payments for patients with Medicare fee for service primary or secondary insurance. Data linkage was performed as previously described (21,23). The cohort was restricted to patients noted to have Medicare as the primary insurer in the OPTN records at the time of transplantation with an appropriate charge for the transplant procedure.
Recipient and donor clinical characteristics Information on recipient clinical and demographic characteristics were drawn from OPTN Standard Analysis and Research files and included records of recipient age, gender, race, blood type, primary cause of endstage liver disease, biologic Model for End-Stage Liver Disease (MELD) at transplant and medical comorbidities. Donor characteristics included mode of death (brain death [DBD] or DCD), ischemic time, age and comorbidities. Patients enrolled in Medicare Managed Care Plans were excluded, as their data does not appear in the records assembled for primary Medicare beneficiaries.
American Journal of Transplantation 2015; 15: 170–179
Dates of death and graft failure were defined by OPTN reports. Coxproportional hazard modeling was used to assess the adjusted impact of the transplant center’s center O:E quartile for BCs on the overall risk of patient death and graft loss among all liver transplants for up to 3 years following LT. Time of origin was the date of transplant, and patients were censored at last expected OPTN follow-up or end of study (August 2010).
Economic analysis To provide a robust economic assessment of the impact of center O:E quartile for BCs on costs of LT, Medicare expenditures for all LT services recorded in Medicare Part B, inpatient, outpatient, home health and hospice claims files were analyzed using multivariable linear regression models (n ¼ 10 829) limited to patients with continuous Medicare coverage. Medicare payments are minimally adjusted for wage and price differences by region, but generally conform to a standardized fee schedule based on diagnostic related groups and evaluation and management codes; costs were not adjusted for changes in the Consumer Price Index, as Medicare payment schedules have changed minimally over the study period. Total accumulated costs per patient were estimated from 3 days before transplant to 1-year posttransplant. MELD spline terms are included to adjust for the nonlinear relationship between MELD score and cost (24). Reported costs include the cost of subsequent retransplant procedures, which occurred within the period of analysis. The incremental cost associated with LT (combining DCD and DBD) at a center with a high O:E for BCs was estimated from this model including an indicator for O:E quartile for DBD transplant as well as a binary variable indicating mode of donor death (e.g. DCD). Data
Axelrod et al management and analyses were performed using SAS 9.3 (SAS Institute Inc., Cary, NC).
Table 1: Characteristics of liver transplant recipients with and without biliary complications No biliary complication (14 475)
Approval The study was approved by the Dartmouth College and Saint Louis University Institutional Review Boards, the data oversight committee of the OPTN, and the Health Resources and Services Administration.
Results Incidence of early BCs Linked registry and Medicare claims data were available for 16 286 LT events, which occurred between 2002 and 2008; 15 724 (93.3%) were performed with a DBD and 562 (3.3%) with a DCD liver allograft. Among DCD liver transplant recipients, 16% developed a BC compared with 11% of DBD liver transplant recipients (p ¼ 0.002). Compared to the liver transplant cohort without BC, the cohort with BC included higher proportion of patient over the age 56, hepatitis C and hepatocellular carcinoma as primary diagnoses, more patients with diabetes, fewer employed patients and fewer patients with a college degree (Table 1). Recipient factors associated with an increased likelihood of BC, included older recipient age, female gender, employment status, malignant neoplasms as type of liver diagnosis and MELD at transplant (p < 0.05 for all) (Supplemental digital content). Donor factors predictive of BC included donor age, partial/split grafts and liver/kidney grafts. Variation in the incidence of BCs by center The risk-adjusted O:E ratio of BC varied dramatically across transplant centers for both DCD and DBD LTs. Among DCD transplant centers (n ¼ 43), O:E ratios for BCs varied from 60% less than expected (O:E 0.40) to 800% higher than expected (O:E 9.03). Rates were statistically significantly higher (p < 0.05) in 10 centers (24%). DBD O:E ratios ranged from 0.28 to 11.10 among 104 centers, 23 (22%) of which were statistically higher than expected. Centers were divided into quartiles on the basis of O:E ratio for BCs (ranked separately for DCD and DBD transplants), with an equal number of centers in each quartile (Table 2). Donor and recipient characteristics were compared among center quartiles for each donor type. DCD: Centers in the lowest O:E quartile of BC after DCD transplant had a mean BC incidence of 8.4%, while those in the highest quartile had an incidence of 19.2% (p < 0.03). The mean O:E ratio varied from 0.5 in the lowest quartile to 3.67 in the worst quartile (p < 0.0001). Centers with the best performance had a higher volume of Medicare DCD transplants in the database, used more imported organs, and transplanted a greater number of organs with a high donor risk index than centers with higher rates of BC. There was a strong trend suggesting an inverse relationship between the volume of DCD transplants performed for Medicare patients and the rate of BC (p ¼ 0.06) (Figure 1A). 172
Recipient age 19–35 36–40 41–55 56–70 71 Recipient gender Male Female Race/ethnic group White Black Hispanic Other Blood type A B AB O Median MELD at transplant College degree Employed Liver diagnosis Noncholestatic cirrhosis Malignant neoplasms Metabolic diseases Acute hepatic necrosis Biliary atresia Cholestatic liver disease Other Comorbidities Hypertension Diabetes Peripheral vascular disease Cerebrovascular disease Region 1 2 3 4 5 6 7 8 9 10 11 Donor age 0–18 19–35 36–50 >50
Biliary complication (n ¼ 1811) p-Value
4.7% 3.2% 45.5% 44.0% 2.6%
2.8% 2.3% 45.0% 46.3% 3.8%