Unrelated Donor Bone Marrow Transplantation for Chronic Myelogenous Leukemia: A Decision Analysis

Methods

Markov Model

A Markov model is an analytic structure that tracks the clinical events occurring in a hypothetical cohort of patients in various scenarios. We constructed a model to analyze the decision faced by a patient with a new diagnosis of chronic-phase CML who is considering having unrelated donor bone marrow transplantation. Five strategies were compared: no transplantation; transplantation within the first year; transplantation 1 to 2 years after diagnosis; transplantation delayed until 2 to 3 years after diagnosis; and transplantation delayed until more than 3 years after diagnosis.

At any time point, the model considered a patient to be in one of the following clinical states: alive with chronic-phase CML; alive without chronic graft-versus-host disease after transplantation; alive with chronic graft-versus-host disease after transplantation; or dead from progressive CML, complications of transplantation, or other causes (Figure 1). Time spent in each state was adjusted for the quality of life experienced while in that state, and a discount factor was applied. Using the model, we calculated discounted, quality-adjusted life expectancy for each strategy, considering competing risks for illness and death from CML and transplantation. A cycle length of 6 months was chosen to match the available clinical data. All analyses were done with DATA (TreeAge Software, Inc., Williamstown, Massachusetts), a decision analysis program.

Data Sources

Data from the medical literature, transplant registries, and physician assessments were used in the model.

Prognosis of Patients with Chronic Myelogenous Leukemia Who Do Not Receive Transplants

The life expectancy of patients with CML who do not undergo transplantation was calculated by using the survival curves from six published studies, including interferon trials and prognostic staging studies [4-615, 19, 26, 27]. After published information ended, the survival curves were extrapolated by using a function fitted to the clinical data until the entire cohort had died. Life expectancy estimates for 35-year-old patients with CML who do not undergo transplantation are shown in Table 1. Similar analyses were performed for 25- and 45-year-old patients (data not shown). For the 15 survival curves analyzed, the mean undiscounted expected survival was 5.15 years (range, 3.66 to 7.58 years). A survival curve derived from patients with intermediate-prognosis CML (Table 1, rank 6) who had an expected survival of 5.31 years was chosen as the baseline curve [15]. Sensitivity analysis was performed by using data from the best (Table 1, rank 1) [15] and worst (Table 1, rank 15) [19] survival curves to illustrate results for patients with the best and worst prognoses. The life expectancy of the Sokal low-risk group (Table 1, rank 1) is superior to that in the groups randomly assigned to receive interferon in the major studies [4-6, 27] and thus was chosen to represent the patients with the best prognosis.

Outcomes of Unrelated Donor Bone Marrow Transplantation

Data from 778 unrelated donor transplantations performed between 1987 and 1994 for chronic-phase CML were provided by the IBMTR and the NMDP. To eliminate overlap from patients reported to both registries, we combined patients reported to the NMDP (transplantations done in the United States; n = 465) with those reported to the IBMTR (transplantations done outside of the United States; n = 313). The final population included both 5/6 and 6/6 serologic matches.

Each Kaplan-Meier curve is based on data from time of transplantation to death, and a separate Kaplan-Meier curve was calculated for each stratum. Kaplan-Meier survival curves through 5 years after transplantation were stratified by patient age (15 to 29 years of age, 30 to 39 years of age, ≥ 40 years of age) and time from diagnosis to transplantation (<1 year, 1 to 2 years, 2 to 3 years, >3 years) to generate 12 strata containing 40 to 125 patients each. Finer stratification was not possible because of the size of the entire population. Fifty-eight percent of patients were male, median age at the time of transplantation was 36 years, and median time from diagnosis to transplantation was 20 months. No modeling assumptions were made for time periods for which clinical data were available. Data were not stratified by CML prognostic groups because these groups are not predictive once a patient proceeds to transplantation ([28]; IBMTR. Unpublished data). When registry data ended, we estimated a 2% annual excess mortality rate for patients who had transplantation and did not have chronic graft-versus-host disease [10] and a 3% annual excess mortality rate for patients who had chronic graft-versus-host disease. These figures were extrapolated from analyses of large related-donor cohorts and are due to late relapses and treatment-related death [10, 29, 30].

The incidence of chronic graft-versus-host disease in this cohort was 55% to 100%. On the basis of published literature [25], we modeled a 59% cumulative incidence of this complication and then performed sensitivity analyses. Patients with chronic graft-versus-host disease tend to have worse quality of life and decreased survival [28, 29]. Although chronic graft-versus-host disease often resolves in practice, our model considers patients with this complication to have ongoing compromised quality of life and increased mortality. This assumption causes underestimation of the value of transplantation.

Adjustments for Quality of Life

Estimates of the quality of life in different health states (utilities) were derived from standard gamble questions [31] posed to 12 physicians who were familiar with transplantation outcomes. This technique assigns a utility between 0.0 (death) and 1.0 (perfect health) to quality of life by identifying the maximum gamble between perfect health and death that a person is willing to accept to avoid a compromised health state. For example, a utility of 0.9 means that a person gives equal value to remaining in a compromised health state and accepting a gamble with a 90% chance of perfect health and a 10% chance of immediate death. The mean utility for life without chronic graft-versus-host disease after transplantation was 0.979 (range, 0.95 to 1.0), and the mean utility for life with chronic graft-versus-host disease after transplantation was 0.9 (range, 0.75 to 1.0). These estimates were tested by using sensitivity analyses.

Time Discount Rate: Time Preference and Risk Aversion

The value of future years of life relative to the present were discounted on the basis of two assumptions. Time preference for life-years assumes that persons value present time more than they do distant time. Risk aversion, a related concept, assumes that persons would choose a more certain survival of intermediate duration over a gamble offering equal life expectancy but including the chance of premature death. We modeled these concepts by a declining exponential function of life years. For example, at a 10% annual time discount rate, 1 year of life now is equivalent to 0.90 years of life next year, which is equivalent to 0.81 years of life 2 years from now. Evidence is mixed about the actual degree of time discounting and risk aversion that patients show: Rates from near zero to more than 200% have been found in the literature [32-34].

Baseline Case

The baseline case was a 35-year-old patient with intermediate-prognosis CML. The utility of being alive with CML was assumed to equal 1.0, the utility of being alive after transplantation without chronic graft-versus-host disease was set at 0.979, and the utility of being alive after transplantation with chronic graft-versus-host disease was set at 0.9. The annual time discount rate was set at 3%. The risk for death of a patient with chronic graft-versus-host disease was assumed to be 50% higher than that for a patient without graft-versus-host disease. Time spent in the hospital for transplantation was given a utility of 0.0.

Sensitivity Analysis

We performed sensitivity analyses to test the robustness of our conclusions when the following baseline measures were varied: patient age, CML prognostic group, excess risk for death with chronic graft-versus-host disease, survival or quality of life after transplantation, survival with therapy for CML other than transplantation, and time discount rate. We also tested the effect of varying our assumptions about the incidence of chronic graft-versus-host disease and excess risk for death more than 5 years after transplantation.

Results

Overall Survival Curves

Overall survival curves for the various strategies are shown in Figure 2, which graphically demonstrates the adverse effect of delaying bone marrow transplantation. Because of the high acute toxicity of transplantation and the low risk for progression of CML in the first 2 years after diagnosis, survival is initially lower with transplantation than with other therapy. However, most surviving bone marrow recipients are cured, whereas almost all patients who do not receive transplants die of CML. For patients with intermediate-prognosis CML, the survival curves for transplantation within the first year and no transplantation cross at approximately 4 years from the time of diagnosis.

Optimal Decision for a Patient with Intermediate-Prognosis Chronic Myelogenous Leukemia

Table 2 shows the predicted life expectancies provided by the various strategies for patients with intermediate-prognosis CML. Adjustments for quality of life and time discounting decrease the absolute value of transplantation strategies but do not change the relative rankings. Transplantation within the first year after diagnosis maximizes quality-adjusted, discounted life expectancy, although net benefit decreases with increasing patient age.

Another interpretation of the data is shown in Figure 3. From a utilitarian perspective, we may ask, “If we compare a population of 35-year-old patients with CML who have transplantation within a year of diagnosis with a similar population who decide not to undergo transplantation, when does the cumulative quality-adjusted, discounted survival of the cohort with transplantation surpass that of the group without transplantation?” As Figure 3 shows, this occurs at 5 years for high-risk patients, 8 years for intermediate-risk patients, and 12.5 years for low-risk patients. This time span may be interpreted as a clinical “payback” period for the initial risk of the transplant procedure.

Interaction of Age and Prognostic Group

For older patients with best-prognosis CML, the combination of better survival without transplantation and worse transplantation outcomes significantly decreases the benefit of early transplantation. For a 45-year-old patient with best-prognosis CML, discounted, quality-adjusted life expectancy is 6.28 years without transplantation, 6.57 years with immediate transplantation, 6.23 years with transplantation delayed 1 to 2 years, 6.26 with transplantation delayed 2 to 3 years, and 5.35 with transplantation delayed more than 3 years (data not shown). Thus, the model predicts almost equivalent outcomes for this population regardless of the strategy chosen.

Sensitivity Analysis

We evaluated the extent to which our conclusions might be affected by different assumptions. The following variables were tested individually: excess mortality more than 5 years after transplantation, improvement in transplantation or CML outcomes, time discount rate, quality of life, increased mortality with chronic graft-versus-host disease, and cumulative incidence of chronic graft-versus-host disease (Table 3). Thresholds at which assumptions affected conclusions were identified for time discount rate and excess transplantation-related mortality; when the annual discount rate is greater than 21% or the excess long-term annual mortality in transplant recipients is greater than 20%, the model predicted that quality-adjusted survival is maximized by no transplantation. Conclusions were not altered by the other assumptions tested or by limitation of the analysis to patients with HLA-identical donors.

The model also allows simultaneous testing of multiple assumptions. Figure 4 shows the effects on preferred therapy of simultaneously changing assumptions about quality of life and incidence of chronic graft-versus-host disease for a 35-year-old patient with intermediate-prognosis CML. If the incidence of chronic graft-versus-host disease is actually 90% and the anticipated quality of life after transplantation is less than 0.5, then not performing transplantation maximizes the expected outcome. However, a utility of 0.5 is low and would mean that a patient equates the certainty of living with chronic graft-versus-host disease and a gamble offering a 50% chance of immediate death and a 50% chance of perfect health.

Strategy of Annual Reevaluation

We assumed that patients who did not elect to have early transplantation would reevaluate their situation each year. Similarly, a patient who does not have an unrelated donor at diagnosis but acquires one at a later date would face the decision about transplantation at a time removed from the initial diagnosis. Although the median time from the initiation of a donor search to transplantation for CML is currently 6 months (NMDP. Unpublished data), many patients are unable to identify a donor quickly or must delay transplantation for other reasons.

For a patient with intermediate-prognosis CML, quality-adjusted expected survival is still maximized by transplantation despite delay. The only exception occurs in older patients who delay transplantation beyond year 3. If a 45-year-old patient survives in chronic phase until year 3, the model predicts that expected survival is equivalent with and without transplantation (Table 2).

Interferon and Cytogenetic Monitoring

Although interferon does not seem to cure CML, it has conferred a survival advantage in several randomized trials [4-6, 27]. Patients who attain complete or major cytogenetic responses do better than those who do not, although it is not clear whether this is due to a change in the natural history of CML or to the identification of a subpopulation with more favorable features. One proposed strategy is to give interferon to all patients with CML and then perform transplantation for those who do not achieve major cytogenetic responses. We reconsidered data from Kantarjian and colleagues' retrospective study of 274 patients treated with interferon [3]. The median time to response in this series was 12 months, and 20% of patients had major (complete or partial) cytogenetic responses by this time point. With longer follow-up, 38% eventually had major responses but 53% never responded.

Although this strategy may spare some patients from transplantation, the 53% who did not respond had a median survival of 5 years. By undergoing a trial of interferon, these patients lose the possibility of transplantation within the first year and decrease their discounted, quality-adjusted expected survival by approximately 2.0 years (Table 2), even if all of those who do not have complete remission proceed to transplantation immediately during the second year. Thus, the average life expectancy of the patients who eventually respond to interferon therapy must be more than 12 discounted, quality-adjusted life-years to make this strategy equivalent to immediate transplantation for all patients. We assumed that patients with CML or patients receiving interferon had a perfect quality of life, but this may not be appropriate because 9% to 25% of patients cannot tolerate interferon because of side effects [4, 6, 27, 35-39].

Interferon and Lowest-Risk Chronic Myelogenous Leukemia

The results of the randomized trial of interferon reported by Hehlmann and colleagues [27] did not show a survival benefit of interferon over hydroxyurea for the entire population but did show an interaction of Sokal score with therapy for CML [40]. Patients with the lowest Sokal score who were randomly assigned to receive interferon were reported to have a median survival of more than 8 years. This is still inferior to the undiscounted life expectancy with early transplantation predicted by our model.

Discussion

Half of patients with a new diagnosis of CML are younger than 50 years of age and are therefore candidates for bone marrow transplantation. Approximately 33% have family member donors [41] and should undergo transplantation within the first year after diagnosis [9, 10, 42]. For the other two thirds, unrelated donor transplantation is the only curative treatment. About 70% of these patients will locate a suitable unrelated donor (NMDP. Unpublished data). Although outcomes are improving, unrelated donor transplantation still carries greater risks for acute illness, death, and long-term side effects than does related donor transplantation.

The decision to undergo or not to undergo transplantation, which is often faced by patients who feel entirely well, is a difficult one [43, 44]. Our results suggest that transplantation within the first year after diagnosis maximizes discounted, quality-adjusted life expectancy. These findings were robust even when we varied all baseline assumptions. However, the model also illustrates some specific clinical situations in which transplantation may not be the best option. These include situations in which utilities for long-term complications of transplantation are extremely low or the time discount rate is high. Patients in these situations maximize their expected discounted, quality-adjusted survival by electing to have treatment other than transplantation. For older patients with best-prognosis CML, all strategies, including no transplantation, resulted in similar discounted, quality-adjusted life expectancies.

Our results must be viewed with the understanding that the reported values represent expected, or average, population values. On an individual basis, some patients will not do well with transplantation and clearly will decrease their survival by electing to have early transplantation. For a population of 35-year-old patients with intermediate-prognosis CML, a cross-sectional evaluation at 3 years would show that more persons had died with a strategy of early transplantation than would have been expected if none had had transplantation. However, a cross-sectional evaluation at 5 years would show the opposite conclusion. Thus, the preferred strategy may differ if the patient hopes to see a son enter high school in 3 years or to see a daughter graduate in 5 years.

This analysis also shows that patients who are cured must survive for a long time to compensate for those who die prematurely of transplantation. However, we suspect that patients who are contemplating transplantation highly value this chance for cure and (just like this decision analysis) are strongly influenced by the potential for long-term survival with transplantation.

Two possible sources of bias in our results must be noted. First, we compare transplantation results with survival curves for CML without transplantation taken from the literature. Although the intermediate- and best-prognosis curves for CML used in this analysis were derived from a series of patients younger than 45 years of age [15], the population that has transplantation may still be fundamentally different from the historical cohort that did not have transplantation. Second, time from diagnosis to transplantation may vary systematically in the IBMTR and NMDP databases. For example, difficulty in locating a donor may reflect the presence of a rarer genotype and, consequently, a worse chance of matching molecularly for class II and minor histocompatibility antigens, leading to worse outcomes. Although we only included transplantation data from patients with chronic-phase CML, patients and clinicians may proceed to transplantation when subtle signs of disease advancement that are not classified in the registries as accelerated disease intervene. However, delaying transplantation may reflect the ability to remain in chronic phase, success with interferon therapy, or clinical assessment of good prognosis, suggesting that patients who delay would have a better outcome. Without a randomized trial, these issues can not be addressed.

Both related [7, 42, 45-47] and unrelated donor transplantation [25] are most successful when performed within the first year after diagnosis of CML. In general, transplantation delayed more than 1 year after diagnosis results in very similar life expectancies; what is gained in survival by delaying a risky procedure is lost as a result of worsened transplantation outcomes and greater risk for CML progression. The seemingly paradoxical advantage for 35-year-olds versus 25-year-olds reflects a true observation in the data that may be due to small numbers (each cohort of 25-year-old patients was smaller than the corresponding groups of 35- and 45-year-old patients) or as-yet unidentified biological characteristics. Because we made no modeling assumptions when actual data were available, any inconsistencies in the clinical data are reflected in the model results.

We did not model the effects of several investigational strategies because we did not think that sufficient long-term data existed to allow us to model outcomes accurately. These strategies include autologous transplantation; empirical T-cell infusions done after transplantation to prevent disease recurrence; adoptive immunotherapy for relapse after transplantation; and up-front therapy for CML with interferon plus cytarabine, which has recently shown promising results [48-56]. The primary reason for exclusion of these data was our uncertainty about the long-term durability of the results. For example, if the results of Guilhot and associates for interferon plus cytarabine [56] were extrapolated, the predicted 10-year survival rate would be 52% and the predicted 20-year survival rate would be 26%, with a resultant life expectancy 75% better than that seen in any previous interferon study. However, as transplantation and therapy for CML other than transplantation evolve, this model may be updated to reflect new outcomes data and clinical considerations.

Dr. Kuntz: Department of Medicine, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115.

Dr. Horowitz and Ms. Sobocinski: International Bone Marrow Transplant Registry, Statistical Center, Medical College of Wisconsin, 9701 Watertown Plank Road, PO Box 26509, Milwaukee, WI 53226.

Dr. McGlave: Department of Medicine, University of Minnesota, Box 480, Mayo Building, Harvard Street at East River Road, Minneapolis, MN 55455-0311.

Dr. Goldman: Haematology Department, Royal Postgraduate Medical School, Du Cane Road, London W12 0NN, United Kingdom.

Ms. Hegland and Dr. Kollman: National Marrow Donor Program, 3433 Broadway Street NE, Suite 500, Minneapolis, MN 55413.

Dr. Parsons: Division of Pediatric Oncology, Dana-Farber Cancer Institute, 44 Binney Street, Boston, MA 02115.

Dr. Weinstein: Department of Health Policy and Management, Harvard School of Public Health, 718 Huntington Avenue, Boston, MA 02115.

Dr. Antin: Division of Hematology/Oncology, Department of Medicine, Brigham and Women's Hospital, 75 Francis Street, Boston, MA 02115.