Secondary Logo

Journal Logo

Original Articles: Hepatology and Nutrition

Withdrawal of Immunosuppression Following Pediatric Liver Transplantation

A Markov Analysis

Mohammad, Saeed*; Li, Zhe; Englesbe, Michael; Skaro, Anton§; Alonso, Estella*

Author Information
Journal of Pediatric Gastroenterology and Nutrition: August 2014 - Volume 59 - Issue 2 - p 182-189
doi: 10.1097/MPG.0000000000000413


The success of cyclosporine as an immunosuppressive agent and advances in surgical techniques in the 1980s propelled liver transplantation (LT) from a medical curiosity to the standard of care for end-stage liver disease and paved the way for worldwide acceptance (1–3). Five-year patient survival following pediatric LT now exceeds 85% (4). A review of late outcomes following LT in the Studies of Pediatric Liver Transplant database reveals 46% of late mortality is related to infections, whereas rejection and graft loss contribute to only 4.7% of deaths (5). These statistics suggest that although present immunosuppressive strategies are successful in preventing graft loss, patients may be overimmunosuppressed, resulting in life-threatening infectious complications. Likewise, chronic immunosuppression (IS) exposure may cause end-organ damage, resulting in hypertension, diabetes, and chronic kidney disease, decrease patient life span, and adversely affect quality of life.

IS has been successfully withdrawn when life-threatening comorbidities have developed (6,7). Its elective withdrawal is novel, and has only been tested at a few transplant centers and in a pilot safety project funded by the Immune Tolerance Network (ITN) (7–9). Early results are promising, with few episodes of graft rejection and cost savings on medications, but it is unclear whether the risks are justified in the asymptomatic patient.

Decision analysis is a useful technique when there is uncertainty in clinical outcomes. Using this approach, models are built representing specific decision problems and are then analyzed to either determine the best action or discover what information is required to make a more informed decision. Decision models may be designed to compare different treatment options and assess the likelihood of each outcome and costs and benefits. Health outcomes are measured using health utilities, which are markers for patients’ preference for their present health state, typically valued between 0 (anchored at death) and 1 (anchored at perfect health). Health states between these 2 anchor points have a utility value between 0 and 1, and states considered worse than death have a negative value. Utility scores may be measured directly through techniques such as the standard gamble and time tradeoff or indirectly with a health state classification system that assigns values to discrete health states such as the Health Utility Index (HUI). One year in perfect health is equal to 1 quality-adjusted life-year (QALY), therefore combining both survival and quality of life. Utilities are based on the value that individuals place on their life in a given state of health and form the basis for cost-effectiveness analyses. Monetary outcomes are measured in dollars, with $50,000 per QALY a commonly accepted marker for a cost-effective intervention. Our objective was to develop a decision model that would compare the effect of withdrawing immunosuppressive medications on survival, cost, and QALYs, thereby assisting families and health care providers in deciding whether withdrawing IS is efficacious and cost-effective.


Biliary atresia is the most common reason for LT in children, with the majority receiving transplantation by their second birthday (10). Therefore, we used a hypothetical cohort of 8-year-old LT recipients, with an original diagnosis of biliary atresia as our base case. Thus, at age 8, they would be 6 years posttransplant, and would represent our “typical patient” and meet the inclusion criteria proposed for IS withdrawal studies at present (9). In our model, we assumed patients were at least 3 years posttransplant, were receiving monotherapy, were taking daily tacrolimus, and did not have secondary comorbidities. We also assumed they had good graft function, defined by transaminases less than twice the upper limit of normal, and had not had an episode of acute or chronic rejection in the last year.

We questioned whether these children should attempt IS withdrawal or continue their present course? The answer depends on the patients’, parents’, and physicians’ perceptions of risk and benefit, which must be informed by a patient's long-term prognosis and the family's preference for risk of potential graft rejection/loss versus the benefit of reduced secondary comorbidities.

Decision Model

We developed a Markov model simulating individual patients’ health states following a decision to withdraw or continue IS for 10 years (Fig. 1). Our model calculated cumulative outcomes for these children according to whether they had their IS withdrawn or continued present therapy. There were 15 health states, with the probability of transitioning to other states determined by the published literature (Table 1). Patients continuing therapy could remain in a state of good health, develop complications related to IS, develop chronic rejection, or die. Patients who develop chronic rejection may also require a second transplant through which they may survive or die. Patients withdrawing IS would start in a state of good health and may remain in that state, or develop acute rejection requiring restarting of IS, in which case the health outcomes mirror those of the group on IS. Patients who required a second transplant were not candidates for a third transplant and could not have their IS withdrawn. Patients with chronic rejection were also not candidates for IS withdrawal.

Markov model of withdrawing immunosuppression. The same strategy was used for patients who continue on immunosuppression with the exception that they did not have the “good medical outcome off IS” health state. The dashed lines represent 1-time transitions, whereas solid lines represent annual transition. Circular arrows signify a patient remains in that particular health state. IS = immunosuppression.
List of transition probabilities used in the Markov model

The model was based on the patient's/transplant center's perspective and was designed to identify the optimal choice for an individual patient. The primary outcome measure was patient survival, and secondary outcomes included cost and QALYs. The cycle length was 1 year, and a 10-year time horizon was chosen because of limitations of available data.

Model Inputs

Health State Probabilities

Probabilities of developing poor outcomes were based on a review of published observational studies of long-term outcomes in pediatric LT recipients (Table 1). Many of these probabilities were taken from single-center reports, and some data are representative of an older era in transplantation when patients received higher levels of IS than those at present.

Estimating the probability of rejection in patients withdrawing IS was problematic because there was only 1 pediatric multicenter report and several small single-center reports with heterogeneous patient populations. To address this, we divided the withdrawal group into an initial 2-year period during which IS was withdrawn and probability of rejection was higher, and subsequently a “good outcome off IS” state if they maintained tolerance. In the initial year, the rate of rejection is highest and medical costs are greater because of increased laboratory monitoring and visits. Patients with an episode of rejection were restarted on IS and joined the patients receiving IS group. Patients who were tolerant after 2 years had the same probability of rejection as patients who were stable on IS and were monitored similarly.


All costs were based on direct medical care costs. Inpatient and outpatient costs accrued were included, but costs related to lost wages or other societal costs were not. Estimated costs for hospitalizations, procedures, medications, and follow-up were drawn from data at our institution, the University of Michigan, and the Red Book(19) (supplementary Table 2, We adapted the IS withdrawal protocol used by the recently completed ITN pilot withdrawal study to estimate the costs for withdrawing IS. It included the cost of 1 biopsy, 4 office visits, and laboratory values every 2 weeks for 1 year followed by monthly laboratory values and biannual clinic visits for the second year (9). Costs were discounted at a rate of 3% annually and adjusted to 2010 dollars using the consumer price index for medical care/services from the Bureau of Labor Statistics.


We used the HUI to assess utilities because it is the most commonly used measure of health utilities in children. Table 3 lists the utility values for health states represented in the model. The HUI is a widely accepted generic, preference-scored, comprehensive system measuring health-related quality of life and producing health utility scores ( We used their references for healthy children to estimate health utilities in our population. Feeny et al compared healthy children (mean health utility of 0.95) with those who were born extremely-low-birth weight (mean health utility of 0.89) (20). Differences in utility scores of 0.03 are considered clinically important, and therefore we used this value or multiples thereof to differentiate our various health states (21). For example, the health state with the highest utility in our population was good medical outcome 2 years after IS withdrawal and was set at 0.92, which is 0.03 less than what we would expect from a healthy population. We considered children with a good medical outcome on IS to be similar to extremely-low-birth weight patients who reported some problems with vision and cognition and awarded a utility score of 0.86, which is 0.06 less than our best group, mirroring the findings by Feeny et al. The majority of our values are higher than averages reported by children with chronic musculoskeletal disorders (health utility 0.80) (22). Unfortunately, most pediatric health utility data are of poor quality or based on parent or physician preference (23,24). All utilities are therefore estimates based on published literature, expert opinion, and an attempt to extrapolate posttransplant health states to common pediatric illnesses (25).

Health utility of health states in the model

Sensitivity Analysis

One-way sensitivity analysis was used to identify threshold values for key variables using a willingness-to-pay level of $50,000. We also conducted probabilistic sensitivity analysis using Monte Carlo second-order simulation based on distributions for all model indicators (probabilities, costs, and utilities). Beta distributions were used to address uncertainty associated with probabilities of transiting to different health states and health utilities based on mean and standard deviation data from the published literature. We used normal distributions for costs with ranges symmetric around the mean value. Monte Carlo simulation was used to generate confidence intervals using the uncertainty around the model inputs. A total of 1000 Monte Carlo iterations were performed, and random values for each indicator were generated during each iteration based on the distributions described above. TreeAge Pro (TreeAge Software, Williamstown, MA) was used to generate the model.


Withdrawing IS in our hypothetical group would lead to a cumulative 10-year survival of 95.8% compared with 88.6% for IS continuation (Fig. 2). During 10 years, patients who attempted to withdraw IS experienced an average of 8.61 QALYs at an average cost of $43,227.30 compared with 8.01 QALYs at $54,622.75 for those continuing on IS. Withdrawing IS yields an average of 0.60 additional QALYs at an average cost saving of $18,992.41 per QALY.

Survival curve: 10-year survival is depicted for the 2 strategies.

Sensitivity Analysis

We examined how variables affected QALYs and costs for a range of possible values and demonstrate those that contribute the most to variability in the model in a tornado diagram (Fig. 3A and B). The most influential variables were health utility off IS, probability of successful withdrawal in the first year, probability of remaining off IS after the second year, and the health utility of a good outcome on IS.

Tornado diagram. The effect of individual variables on QALYs gained (A) and incremental cost-effectiveness of immunosuppression withdrawal (B). Bars indicate changes in the dependent variable when indicators are varied in 1-way sensitivity analyses from the lowest to the highest extreme of their plausible range (extreme values are shown adjacent to bars). The median line represents the QALYs and incremental cost-effectiveness calculated for the base case. A, Health utility off IS, probability of successful wean in first year, and health utility on IS are 3 variables that when varied may lead to a change in strategy. B, Probability of successful wean in first year and probability of remaining off IS contribute the most to variability in the incremental cost-effectiveness. IS = immunosuppression; QALYs = quality-adjusted life-years.

Health Utility

We compared health utility on and off IS in a 1-way sensitivity analysis (Fig. 4A and B). Maintaining health utility on IS at 0.86, withdrawing IS was the preferred option unless the health utility off IS was <0.81. Alternatively, with the health utility off IS at 0.92, continuing IS would only yield more QALYs if health utility on IS was >0.94.

One-way sensitivity analysis. Dotted line represents the threshold that will lead to a change in strategy. A, Health utility of good outcome off IS and QALYs; withdrawal strategy is preferred until utility is <0.816. B, Health utility on IS and QALYs; withdrawal strategy is preferred until health utility on IS exceeds 0.941. C, Probability of successful IS withdrawal and QALYs; withdrawal strategy is preferred until probability decreased <19.1%. D, Probability of successful IS withdrawal and cost; withdrawal strategy is more cost-effective until probability decreases <51.5%. IS = immunosuppression; QALYs = quality-adjusted life years.

Successful Withdrawal of IS During the First Year

We varied the probability of a successful outcome of IS withdrawal in the first year against QALYs and cost (Fig. 4C and D). If the probability of success exceeded 19.1%, then withdrawal was the preferred option; however, the probability had to be >51.5% for it to also be cost-effective (incremental cost-effectiveness ratio <$50,000).

Successfully Treating Rejection

We compared the probability of successfully treating rejection with the probability of a good outcome after the first year of IS withdrawal, that is, no rejection. With the probability of rejection set at 50%, withdrawing IS always yields greater QALYs; however, it only becomes cost-effective (ICER <$50,000) when the probability of successfully treating rejection is >70% (supplementary Table 4,

Monte Carlo

Monte Carlo analyses suggested that withdrawing IS was associated with improved outcomes in 76.7% of all iterations (mean 8.59 QALYs, 95% CI 6.85–9.44) compared with continuing IS (mean 8.03 QALYs, 95% CI 6.24–9.02). Withdrawal of IS had a cost of <$50,000 per QALY in 92.5% of the simulations. Figure 5A and B displays Monte Carlo scatterplot results.

Monte Carlo analyses graphically showing incremental effectiveness (QALYs) and lower costs ($) in the majority of simulations when withdrawing IS (A) and decreased effectiveness and higher costs when continuing IS (B). For each one of the 1000 trials, values for the indicators in the model are selected from their respective distributions and an ICER is calculated. A 95% confidence ellipse is placed around approximately 95% of the points and represents the uncertainty in the ICER estimate. Points falling above the dotted line have an ICER of >$50,000 per QALY; those falling below the line have ICERs of <$50,000 per QALY. The “area above the horizontal” is cost-increasing, and to the “right of the vertical” clinically beneficial. A strategy is said to be dominant when it is both more effective and less costly (right lower quadrant), and it is dominated when it is less effective and more costly (left upper quadrant). Cost-effectiveness of withdrawing IS versus continuing IS. A total of 76% of the 1000 simulation iterations fell below the willingness-to-pay threshold of $50,000, and the majority resulted in gains in QALYs at a lower cost (dominant). ICER = incremental cost-effectiveness ratio; IS = immunosuppression; QALYs = quality-adjusted life-years; WTP = willingness to pay.


The premise of lifelong IS for solid organ transplant recipients is increasingly in question because of concern over the long-term adverse effects of IS and its effect on health-related quality of life. IS withdrawal is an experimental procedure, and clinicians, patients, and families must balance its risks and benefits. The financial and health costs of withdrawing IS to a patient who has not developed tolerance include acute rejection, increased IS medication during the short term, chronic rejection, graft loss, and patient death. Our model demonstrates that patients who withdrew IS had an increase in survival of 0.60 years at a cost of −$11,395.45 during the first 10 years. Every additional QALY gained is attained at a cost of −$18,992.41 and is therefore cost saving.

Sensitivity analysis on factors that contribute to the greatest variability in the model reinforced our overall findings. The successful withdrawal of IS during the first year was an important variable in both QALYs and costs (Fig. 3A and B) and is likely one that is preeminent in the minds of clinicians and patients. Feng et al (9) successfully withdrew IS in 60% in their cohort, whereas data from the Kyoto group reported a 15% success rate in a much larger population (6). We used the higher of these in our base case analysis because the population and methodologies we based our model on more closely resembled the study by Feng et al. To test the robustness of our results, we varied the probabilities more than and less than these values (10%–85%) in both our sensitivity analysis and Monte Carlo analyses. In both cases, withdrawal of IS was the favored approach. The probability of success in the first year had to exceed 19.1% for the approach to yield greater QALYs and would be cost-effective if the probability were >51.5%. We believe that by using the successful protocols developed by the groundbreaking ITN study, these targets are achievable.

The clinical significance of these results must be interpreted in the context of the individual patient. Ideally, LT as a cure for end-stage liver disease should allow a child to have a normal quality of life. Earlier studies have shown that these patients report a quality of life similar to children with chronic diseases (26). Adverse effects of IS and increased susceptibility to infections may contribute to this decline. Renal dysfunction caused by prolonged exposure to calcineurin inhibitors is a potentially modifiable cause of significant comorbidity and death in pediatric and adult LT recipients (12,27).

We expected health utility would improve in patients who are no longer receiving IS compared with those who continued receiving IS (0.92 vs 0.86). Withdrawal remained the approach with greater QALYs even if utilities were to decrease less than our base case estimate of 0.86 (Fig. 4A). This is likely because of the overall favorable outcomes in patients who withdraw IS with respect to improved survival, decreased costs, and fewer comorbidities.

There is also a significant cost difference between the 2 groups, driven mainly by the annual cost of IS in those continuing IS. Other factors include the larger proportion of patients who develop poor outcomes, which in turn have a higher annual cost. In our simulations, the 95% CIs reflect uncertainty in the model indicators and not individual outcomes. They should be interpreted as the range of outcomes. More relevant for interpretation is the percentage of model simulations in which 1 strategy is preferred over the other. Withdrawal of IS leads to improved QALYs in 76.7% of simulations at a cost per QALY that is <$50,000 in >90% of iterations. The question then arises, why are not more transplant centers withdrawing IS from stable LT recipients?

The foundation of LT has been based on successful IS, and uncertainty regarding outcomes and fear of graft loss have certainly limited the number of physicians and parents willing to embark on IS withdrawal, except in cases of life-threatening comorbidities. As transplant survival rates increase and the long-term adverse effects of immunosuppressive medications affect an increasing number of recipients, it may become a more common practice in the future. It is only with varying health utility on IS above plausible ranges (>0.94) that continuing IS became the preferred option. Our model suggests that it is time to develop randomized controlled trials to remove IS in pediatric LT recipients. Any model is only as good as its inputs; therefore, additional research to clarify the likelihood and the quality of life decrement from long-term complications may help inform patient and physician decision making.

There are several limitations to the present analysis. First, the accuracy of the model output is directly dependent on the availability of “true” probabilities, and utilities. Participants with varying abilities may understand and assess the hypothetical situations proposed in typical standard gamble and time tradeoff scenarios differently (24,28). The measure of such parameters in children and/or involvement of parents as proxies may further obfuscate results. To address these questions related to the potential variability or inaccuracy of model indicators, sensitivity analyses were performed varying the major model indicators simultaneously as described above. These analyses demonstrated that continuing IS is the preferred option only if the health utility values of a patient 1 year after withdrawing IS were lower than those in patients who continued on IS. Although there are no published data, we do not believe that patients off IS will report lower health utility values when compared with those who continue to take daily IS. Our model represents patients with biliary atresia who have received transplantation by age 2, and, although common, this may not be generalizable to all transplant recipients.

By choosing a 10-year time horizon, gains associated with prolonged survival after IS withdrawal and adverse effects of long-term IS were potentially minimized. This was an attempt to limit bias favoring withdrawal of IS. Additionally, 10 years was selected because of concerns about the ability to extrapolate beyond this time period, especially in light of the significant evolution in transplant practices and the transition of many patients to adulthood.

Our model demonstrates the feasibility of withdrawing IS in recipients of LT with good graft function who are 6 years posttransplant and provides a conceptual framework for physicians, patients, and their families to understand the relative risks and benefits of both strategies. In light of our findings, clinical trials investigating biomarkers and other predictors of successful IS withdrawal must be aggressively pursued. These efforts will provide more accurate data for future models and help personalize IS regimens for future transplant recipients.


1. McDiarmid SV. Current status of liver transplantation in children. Pediatr Clin North Am 2003; 50:1335–1374.
2. Otte JB. History of pediatric liver transplantation. Where are we coming from? Where do we stand? Pediatr Transplant 2002; 6:378–387.
3. Starzl T. Memoirs of a Transplant Surgeon. The Puzzle People. Pittsburgh:University of Pittsburgh Press; 1992.
4. Diamond IR, Fecteau A, Millis JM, et al. Impact of graft type on outcome in pediatric liver transplantation: a report from Studies of Pediatric Liver Transplantation (SPLIT). Ann Surg 2007; 246:301–310.
5. Soltys KA, Mazariegos GV, Squires RH, et al. Late graft loss or death in pediatric liver transplantation: an analysis of the SPLIT database. Am J Transplant 2007; 7:2165–2171.
6. Koshiba T, Li Y, Takemura M, et al. Clinical, immunological, and pathological aspects of operational tolerance after pediatric living-donor liver transplantation. Transpl Immunol 2007; 17:94–97.
7. Takatsuki M, Uemoto S, Inomata Y, et al. Weaning of immunosuppression in living donor liver transplant recipients. Transplantation 2001; 72:449–454.
8. Mazariegos GV, Reyes J, Marino IR, et al. Weaning of immunosuppression in liver transplant recipients. Transplantation 1997; 63:243–249.
9. Feng S, Ekong UD, Lobritto SJ, et al. Complete immunosuppression withdrawal and subsequent allograft function among pediatric recipients of parental living donor liver transplants. JAMA 2012; 307:283–293.
10. McDiarmid SV, Anand R, Lindblad AS. Studies of pediatric liver transplantation: 2002 update. An overview of demographics, indications, timing, and immunosuppressive practices in pediatric liver transplantation in the United States and Canada. Pediatr Transplant 2004; 8:284–294.
11. Hurwitz M, Desai DM, Cox KL, et al. Complete immunosuppressive withdrawal as a uniform approach to post-transplant lymphoproliferative disease in pediatric liver transplantation. Pediatr Transplant 2004; 8:267–272.
12. Ng VL, Fecteau A, Shepherd R, et al. Outcomes of 5-year survivors of pediatric liver transplantation: report on 461 children from a North American multicenter registry. Pediatrics 2008; 122:e1128–e1135.
13. Bueno J, Medina A, Ortega J, et al. Liver transplantation in childhood with more than 10 years of follow-up: analysis of a single-center experience. Transplant Proc 2007; 39:2288–2289.
14. Avitzur Y, De Luca E, Cantos M, et al. Health status ten years after pediatric liver transplantation—looking beyond the graft. Transplantation 2004; 78:566–573.
15. Ueda M, Oike F, Ogura Y, et al. Long-term outcomes of 600 living donor liver transplants for pediatric patients at a single center. Liver Transpl 2006; 12:1326–1336.
16. Ng V, Anand R, Martz K, et al. Liver retransplantation in children: a SPLIT database analysis of outcome and predictive factors for survival. Am J Transplant 2008; 8:386–395.
17. Adani GL, Baccarani U, Risaliti A, et al. A single-center experience of late retransplantation of the liver. Transplant Proc 2005; 37:2599–2600.
18. Jain A, Mazariegos G, Kashyap R, et al. Pediatric liver transplantation. A single center experience spanning 20 years. Transplantation 2002; 73:941–947.
19. PDR Network. Red Book 2010: Pharmacy's Fundamental Reference, 114th ed. Montvale, NJ: PDR Network, LLC; 2010.
20. Feeny D, Furlong W, Saigal S, et al. Comparing directly measured standard gamble scores to HUI2 and HUI3 utility scores: group- and individual-level comparisons. Soc Sci Med 2004; 58:799–809.
21. Grootendorst P, Feeny D, Furlong W. Health Utilities Index Mark 3: evidence of construct validity for stroke and arthritis in a population health survey. Med Care 2000; 38:290–299.
22. Brunner HI, Maker D, Grundland B, et al. Preference-based measurement of health-related quality of life (HRQL) in children with chronic musculoskeletal disorders (MSKDs). Med Decis Making 2003; 23:314–322.
23. Ladapo JA, Neumann PJ, Keren R, et al. Valuing children's health: a comparison of cost-utility analyses for adult and paediatric health interventions in the US. Pharmacoeconomics 2007; 25:817–828.
24. Griebsch I, Coast J, Brown J. Quality-adjusted life-years lack quality in pediatric care: a critical review of published cost-utility studies in child health. Pediatrics 2005; 115:e600–e614.
25. Carroll AE, Downs SM. Improving decision analyses: parent preferences (utility values) for pediatric health outcomes. J Pediatr 2009; 155:21–25.25.e1–25.e5.
26. Taylor R, Franck LS, Gibson F, et al. A critical review of the health-related quality of life of children and adolescents after liver transplantation. Liver Transpl 2005; 11:51–60.[discussion 7–9].
27. Cole CR, Bucuvalas JC, Hornung R, et al. Outcome after pediatric liver transplantation impact of living donor transplantation on cost. J Pediatr 2004; 144:729–735.
28. Petrou S. Methodological issues raised by preference-based approaches to measuring the health status of children. Health Econ 2003; 12:697–702.

cost-effectiveness; decision analysis; health utility; quality-adjusted life-years

Supplemental Digital Content

© 2014 by European Society for Pediatric Gastroenterology, Hepatology, and Nutrition and North American Society for Pediatric Gastroenterology,