Bone marrow transplantation versus periodic prophylactic blood transfusion in sickle cell patients at high risk of ischemic stroke: a decision analysis

Sickle cell disease (SCD) is associated with the high risk of both ischemic and hemorrhagic stroke. Patients with SCD who experience an ischemic stroke have an even higher risk of recurrent ischemic stroke.1 Several studies have demonstrated that patients with SCD at high risk for stroke may be identified by transcranial Doppler (TCD) measurements of blood velocity in the internal carotid and middle cerebral arteries.2,3 The STOP Study, a stroke prevention trial for patients with SCD at high risk of stroke, demonstrated that periodic blood transfusion (PBT) significantly lowers the risk of stroke in patients with cerebral blood velocity (CBV) greater than 200 cm/s.4 

Currently, patients with SCD who have had an ischemic stroke and are thus at risk for recurrence are typically transfused for their lifetimes. This treatment strategy stems primarily from 2 studies both of a small series of patients in whom the stroke recurrence rate was greater than 50% within a year after transfusions were halted.5,6 However, follow-up studies of similar sizes have demonstrated much more favorable results in patients whose transfusion regimens were stopped or diminished after 3 to 9 years.7-9Nevertheless, most practitioners still recommend that patients with SCD who have a stroke be transfused indefinitely, as the consequences of recurrent stroke can be devastating. The duration of transfusion required to decrease the risk of primary stroke in patients with abnormal TCD is unknown.

Long-term transfusion, although highly effective in preventing ischemic stroke recurrence and other complications of SCD,4 is associated with increased risk of iron overload, infection, and alloimmunization. Iron overload and toxicity lead to organ dysfunction, including early cardiac death. The consequences of iron toxicity can only be prevented by chelation therapy, which is both inconvenient, leading to poor compliance, and expensive. For these reasons, patients who have had an ischemic stroke are considered candidates for bone marrow transplantation (BMT) if they have a matched sibling donor.10 Available data suggest that, despite significant mortality and morbidity, BMT prevents stroke recurrence and obviates the need for long-term transfusions.10-13 

Because of the high efficacy of transfusions in preventing primary strokes, it is currently recommended that all patients with SCD and with high CBV be transfused for an indefinite period to keep their sickle-cell hemoglobin (HbS) levels below 30% of total hemoglobin. In a logical extension from secondary stroke prevention, it has been proposed that patients with abnormal CBV also be considered candidates for BMT.

The decision to offer BMT to patients with SCD is complicated by a number of issues, including the uncertainty of outcomes.10,13 To date, no study examined whether patients with SCD and with a high risk of stroke, identified by abnormal TCD, would have better outcomes with PBT or BMT. Although PBT may be favored because of the relatively high mortality risk (5%-10%) associated with BMT, some clinicians note that lifetime transfusions and iron chelation significantly diminishes quality of life, and that BMT offers patients a chance to be “cured” of SCD and the associated pain crises and morbidity. Thus, the factors that are thought to be of greatest importance in this decision include the risk of death, quality of life, and need for lifelong transfusions for patients receiving PBT. Donor availability and compliance with iron chelation may also play an important role in determining which strategy is preferred.

The factors involved in choosing between BMT and periodic transfusions to manage patients at high risk for strokes are very similar to those faced in recommending unrelated BMT for patients with chronic myelogenous leukemia (CML). For patients with CML, the decision to undergo unrelated BMT, complicated by the unpredictability of outcomes and risk of slow disease progression, must be carefully weighed against the risks of morbidity and mortality from matched unrelated donor BMT. These issues were recently successfully addressed in a decision analysis that found transplantation within the first year of diagnosis of CML provided the greatest quality-adjusted expected survival, when compared with delayed transplantation or no transplantation.14 

We report here a decision analysis study using Markov models to compare BMT with PBT for the management of a population of patients with SCD who have high CBV. The analysis is applicable to the population of patients with SCD who have no indications for BMT, who would be transplanted only because of an abnormally high CBV. Because these SCD patients are assumed not to have multiple recurrent pain crises, episodes of acute chest syndrome, or other abnormalities that would make them candidates for BMT, they are likely to be healthier, in general, than the average SCD patients.

The decision analysis model was designed to answer the following question: “For sickle cell patients identified as high risk of stroke by transcranial Doppler with no other indications for BMT, should the recommended treatment be bone marrow transplantation or periodic prophylactic blood transfusion?” The overall decision tree diagram is shown in Figure 1.

Markov models were constructed to represent the clinical course after BMT, as well as the course of SCD during PBT. Medical literature and expert opinion provided estimates of transition probabilities from one clinical state to another, including the risks of stroke and death for each of several disease states. This decision analysis approach allows the synthesis of information on the clinical course of SCD, risk of stroke, outcomes of BMT and PBT, and the clinical judgment of physicians experienced in the care of patients with SCD to estimate quality-adjusted life years (QALYs), the outcome of interest, associated with each management strategy.

Because not all patients with SCD have suitable donors, and because some patients with SCD refuse BMT and/or PBT, 2 separate analyses were performed on hypothetical cohorts of 100 000 patients, followed for 20 years. First, an intention-to-treat analysis was used to determine the QALYs of each strategy on an entire population of patients with SCD, which included patients who may have refused treatment. Second, QALYs of those patients who actually received BMT were compared with the QALYs of patients who actually received PBT. Specific attention was given to examine the benefits incurred by those patients who were compliant with PBT and iron chelation.

We used Markov models to represent the clinical course over the lifetime of patients undergoing BMT and PBT. These models consisted of several mutually exclusive disease states, and estimates were made of the yearly transition probabilities from one state to another over time, using data from published medical literature and expert opinion.

After undergoing BMT, the patient may have a successful outcome without complications. However, the patient may experience chronic graft versus host disease (cGVHD), BMT rejection, hemorrhagic stroke, ischemic stroke, or death. The model incorporating these clinical states is illustrated in Figure 2. Several published studies provided data on the risk of death, risk of cGVHD, and risk of transplant rejection among patients receiving BMT. Data are reported by Walters et al10 (n = 32), Abboud et al11(n = 9), Vermylen et al12 (n = 50), and Bernaudin et al13 (n = 26). Table 1summarizes the data obtained from these studies. Of the 117 total BMT patients, 7 (6.0%) died within 1 year, 19 (16.2%) experienced cGVHD, and 14 (12.0%) rejected the transplant. Among the patients receiving BMT, risk of dying is assumed to be twice that of a person without SCD (still much lower than patients with SCD). In addition, there is a high (6%) risk of dying during the first year after transplantation, and this risk diminishes over the next 2 years. The risk of cGVHD developing is assumed to be 10% within the first year and to diminish to 0% by 5 years after BMT. The risk of an ischemic stroke after BMT (Table 2) is assumed to be the same as the risk of a patient with SCD who receives PBT (90% lower than it would have been without BMT or PBT, according to results from the STOP study). The risk of a hemorrhagic stroke in a patient with SCD is presumed to be unchanged by BMT. The risk of BMT rejection is approximately 9% in the first year and rapidly decreases to 0% by 5 years.

Among patients who receive BMT and in whom cGVHD develops, the risk of death is increased by 15% compared with patients with successful BMT. The risk of stroke and risk of BMT rejection are similar to the risks associated with successful transplantation, described above. Among patients with SCD who experience BMT rejection, the risk of death is twice as high as it is in other patients with SCD who did not receive BMT, and this risk lessens with age. Among patients with SCD who experience BMT rejection, the risk of hemorrhagic stroke is assumed to be equal to that of a typical patient with SCD, but the risk of ischemic stroke is assumed to be equal to that of a high-risk stroke patient who is not transfused.

For BMT, the availability of a human leukocyte antigen (HLA)–matched donor sibling was obtained from a study by Walters et al,15along with refusal and exclusion rates. Of 315 patients with sickle cell, 44 (14%) had an HLA-compatible sibling, with 25 (57%) of the 44 patients accepting BMT.

In the PBT model (Figure 3), there are 5 states: on PBT (healthy), alloimmunization, hemorrhagic stroke, ischemic stroke, and death. Patients with abnormal TCD receiving PBT have a reduced risk of death when compared with patients with SCD who do not have a high risk of stroke, as transfusions reduce the rate of complications associated with SCD.7-9 The risk of hemorrhagic and ischemic stroke (Table 2) is the same as that for patients who receive BMT. The risk of alloimmunization, defined as development of 1 or, more commonly, multiple antibodies that make it impossible to transfuse a patient safely, is assumed to be 0.2% per year. Patients who receive PBT and who develop alloimmunization are assumed to be no longer eligible for PBT, and thus their risk of ischemic stroke is the same as the risk for a high-risk stroke patient. Patients compliant with transfusions and any iron-chelating therapy that may be necessary (ie, desferoxamine) are assumed not to suffer consequences associated with iron overload. Patients who were noncompliant with iron chelation have an increased risk of ischemic stroke, and the risk of death increases as iron overload develops and then decreases slowly over time as interventions to reduce iron are initiated (eg, intravenous desferoxamine, stopping transfusions). Patients who were noncompliant with the transfusion regimen were assumed to receive half as many transfusions (ie, once every 2 months), resulting in a 5-fold increased risk of stroke, relative to the transfused patient. Patients who refuse transfusion therapy are assumed to have twice the risk of death and 10 times the ischemic stroke risk of those patients who are transfused on a regular basis.

The probability of a patient accepting the PBT treatment regimen was taken from STOP study data. In this study, 162 (79%) of 206 patients agreed to undergo PBT to determine whether transfusion lowered the risk of stroke. Given the definite benefit from transfusion in reducing stroke, the number of patients with SCD who would accept PBT might actually be higher than what has been projected. The probability that a patient would be compliant with the once-per-month transfusion regimen was assumed to be 0.80, whereas the probability that a patient would be compliant with iron-chelating therapy was assumed to be 0.35.

The risk of hemorrhagic stroke was derived from work published by Ohene-Frempong et al.16 In this study, the risk of hemorrhagic stroke varied by age, with the highest rates among those aged 6 to 9 years old (0.25 strokes per 100 patient-years) and those aged 20 to 29 years old (0.44 strokes per 100 patient-years). These rates were applied to both the BMT patients and the PBT patients. The risk of death immediately after hemorrhagic stroke was assumed to be 33%, as data from 2 studies of stroke were combined.16,17 

The risk of ischemic stroke among high-risk stroke patients has not been fully studied. In the STOP study,4 the risk of stroke was estimated at 10% per year for high-risk patients, and the risk among patients who received PBT was shown to be reduced by 90%. Ohene-Frempong et al16 described the natural history of ischemic stroke in patients with sickle cell, with ischemic stroke occurrence peaking in the preteenage years and again after age 30. Using this pattern of stroke risk described by Ohene-Frempong et al,16 combined with the rates observed in the STOP study, we determined probabilities of ischemic stroke as is shown in Table 2.

Mortality rates among the average sickle cell patients were obtained from a study by Platt et al,18 whereas mortality rates for the average healthy black population were obtained using life tables derived from United States vital statistics data.19 The average SCD patient has a life expectancy of 42 years, compared with 71 years for the black population of the US. Depending on treatment and outcome, patient category–specific mortality probabilities were obtained by multiplying the Platt et al18 and life table mortality probabilities by relative risks estimated by an expert panel of clinicians. Table 3 lists the disease state classification, the respective reference group, and the relative risk of death applied in the analysis. For example, patients who undergo a successful BMT were assumed to die at a rate equal to twice that of the healthy US black population, an assumption that was based on a recent study in which mortality rates for patients surviving at least 2 years after BMT for conditions, including acute myelogenous leukemia, acute lymphoblastic leukemia, chronic mylogenous leukemia, or aplastic anemia were shown to be substantially (4 to 26 times) higher than age-, sex-, and nationality-matched general populations.20 Patients with SCD who survive BMT were not assumed to have quite the dramatic increases in risk of death as the patients with leukemia; nevertheless, patients with SCD who survive BMT probably have some increased risk of death compared with the general population. We assumed this risk of death to be 2 times that of the general population. Patients who reject a BMT were assumed to die at a rate equal to twice that of patients with SCD. The risk of death was assumed to increase as iron levels increase, and decrease slowly as the physician intervenes to reduce iron load, and the magnitude of this elevated risk of death was obtained from data published by Olivieri et al.21 

In this study, the major outcome of interest was QALYs. In the absence of direct reports of quality of life from patients with SCD, estimates of quality of life were based on information provided by 2 physicians (M.R.A., S.M.J.) experienced in the care of patients with SCD. Perfect health was assigned a quality of life of 1, whereas death was assigned a quality of life of 0. Patients who underwent a successful BMT were assigned the highest quality of life (0.95), followed by BMT patients with cGVHD (0.85), compliant PBT patients (0.85), patients noncompliant with the transfusion regimen (0.85), transfused patients who have become alloimmunized (0.80), patients who refuse BMT and PBT (0.80), PBT patients noncompliant with iron chelation (0.75), and BMT patients who reject the transplant (0.70). In a study of stroke patients aged 18 to 57,22 quality of life values were estimated at 0.45 for major stroke, which we have adopted in our model. These quality of life values are well within the ranges of the Health and Activity Limitation Index (HALex), developed using data from the National Health Interview Survey and reported for the more common conditions, such as diabetes (median HALex = 0.63) and chronic sinusitis (median HALex = 0.92).23 Because we have accounted for the quality of life in our models, the main outcome of comparison between the 2 strategies is the number of expected QALYs.

The value of future years of life was adjusted, based on the assumption that people value the present time more than they value future time, in a manner similar to the way a person would rather receive a dollar today than 10 years from now. The most common method of accounting for such a preference is by valuing future years of life with a declining exponential curve. If the rate of decline is 5%, then 1 year of life now is worth 0.95 years of life next year, and 0.9025 years of life in 2 years. Although discount rates vary throughout the literature, recent recommendations for cost-effectiveness analysis made by The Panel on Cost-Effectiveness in Health and Medicine include a base case discount rate of 3%,24 which has been adopted in this study.

For the purposes of examining the hypothetical cohorts within this study, a baseline case scenario was developed that included the following assumptions. The patient with SCD has been identified as high risk for stroke by means of TCD, and no other indications for BMT have been identified. No assumption of gender is made, but we have assumed that the patient is 5 years old at the time the treatment decision needs to be made. Patients with SCD who receive PBT are assumed to need transfusions indefinitely (ie, for the remainder of their lives). The baseline mortality rate of these patients is equal to 80% of that of the typical sickle cell patient, with transfusions providing additional protection from death. We have assumed that 80% of patients who receive PBT will be compliant with the PBT regimen and that, of these patients, 35% will be compliant with iron-chelating therapy. Life years are assumed to be discounted at 3% per year.

Because of the uncertainty in the assumptions made in our model, sensitivity analyses were performed on variables used in the intention-to-treat analysis. This method helps identify variables that have the greatest impact on the QALYs associated with each treatment strategy. The sensitivity analyses varied the discount rate, age at transplantation, the duration of PBT, relative risk of death among those transfused, probability of noncompliance (both with the transfusion regimen and with iron chelation), refusal of PBT and BMT, probability of donor availability, 1-year mortality risk associated with BMT, and quality of life values. In addition, Monte Carlo simulations were performed (10 000 trials for each strategy) to estimate the variance around the expected QALYs of the PBT and BMT strategies.

For patients with SCD at high risk of stroke, a decision to manage patients with an “intention-to-transplant” strategy would yield, on average, 15.2 discounted QALYs per patient, compared with 14.9 discounted QALYs per patients under an “intention-to-transfuse” strategy. Table 4 shows expected outcomes after 20 years in hypothetical cohorts of 100 000 patients with SCD under the 2 different intention-to-treat strategies. Although there are slightly more patients living stroke-free under the BMT strategy than under the PBT strategy, the magnitude of this difference is small (0.8%). In addition, the difference in the total number of patients alive after 20 years under each strategy is even smaller in magnitude (0.1%). Thus, a recommendation of BMT for 100 000 such patients would result in 810 more patients living stroke-free after 20 years than a recommendation of PBT; however, there would also be 111 more deaths in the group for whom BMT was recommended. Further analysis indicates that under the BMT strategy, 7.5% of patients die within 5 years, compared with 7.4% under the PBT strategy. A key reason why these 2 strategies have similar results is that only 8% of patients intended for BMT actually receive BMT, due to lack of donor availability and refusals.

Those patients actually receiving BMT could expect 18.6 discounted QALYs, whereas the average PBT patient could expect 15.7 discounted QALYs. Patients compliant with PBT and chelation had 19.2 expected discounted QALYs. Patients who are noncompliant with iron-chelating therapy could expect 14.1 discounted QALYs, whereas those noncompliant with transfusion could expect 14.8 discounted QALYs. Patients who refuse PBT could expect 12.3 discounted QALYs. Further analysis indicates that among those receiving BMT, 11.6% die within 5 years, compared with 2.6% among compliant PBT patients, 6.5% among those noncompliant with the transfusion regimen, 10.4% among those noncompliant with iron chelation, and 8.6% among those who refuse PBT and BMT. Thus, patients compliant with transfusion would do better, on average, than patients given BMT. However, the noncompliant patients would do worse, on average, than the BMT patients.

Table 5 shows patient outcomes after 20 years for several hypothetical cohorts of 100 000 patients: (a) those patients who actually received BMTs, (b) those patients who actually received PBT, and (c) those patients who actually received and were compliant with BMT and PBT. All patients who received BMT were presumed to be compliant with treatment, including posttransplant follow-up and immunosuppression. The results show that those receiving BMT are slightly more likely to be alive after 20 years than those patients who received PBT. Those patients who were compliant with PBT, however, had the best results, with a substantially greater proportion of patients alive after 20 years (78%) compared with BMT patients (65%). If donors were available and 100 000 patients received BMT, there would be 13 050 fewer patients alive and stroke-free after 20 years and 8814 more deaths among those receiving BMT than if the 100 000 patients were treated and compliant with PBT. Thus, patients who comply with PBT and with iron chelation have, by far, the best outcomes.

Results of sensitivity analyses are summarized in Tables6 and7. The tables list the variables for which the sensitivity analyses were performed, the baseline and ranges of values chosen, and the QALYs resulting from the intention to transplant and the intention-to-transfuse management strategies for patients with SCD.

At a discount of 0%, the decision to consider transplantation offered only a marginal benefit over considering PBT (27.8 QALYs vs 26.8 QALYs). As the discount rate was increased to 10%, the benefit of BMT over PBT diminished to zero (6.8 QALYs for both BMT and PBT). Discounting had a greater impact on life years after BMT, because the BMT cohort experienced more life years in the distant future that are subsequently affected by discounting.

As the age at transplantation varied from 2 to 10 years, the difference between the 2 strategies remained constant (0.3 QALYs). In addition, the number of QALYs associated with each strategy was not sensitive to changes in the age at transplantation, given that for BMT, the number of expected QALYs was 15.3 for a 2-year-old and 15.0 for a 10-year-old, and for PBT, the number of expected QALYs was 15.0 for a 2-year-old and 14.7 for a 10-year-old.

As the number of years of transfusion to decrease the stroke risk was varied from a lifetime of periodic transfusions down to 10, 5, and 3 years, the degree to which BMT was favored diminished significantly. If only 5 years of transfusion are required to reduce the risk of stroke in these patients, and if the percentage of patients who refuse PBT treatment can be kept to a minimum (less than 5%), then BMT no longer offered any benefit over PBT.

As the risk of death associated with transfusions changed, there was little change in the difference between the QALYs associated with BMT and those associated with PBT. With transfused patients having half the typical SCD patient's risk of death (ie, RR = 0.5), the difference between BMT and PBT was 0.2 QALYs, as was the case when a relative risk of 0.3 was chosen.

As the percentage of patients who are noncompliant with their regular transfusions varied, there was little change in the difference between the QALYs associated with considering BMT and those associated with considering PBT. If 50% of patients receiving PBT were noncompliant with the transfusion regimen, the decision to consider BMT would offer 15.4 QALYs versus 15.1 QALYs for considering PBT. This difference diminished only slightly as the proportion of PBT patients who were noncompliant decreased.

As compliance with iron-chelating therapy increased, PBT resulted in greater QALYs. For example, when the proportion of patients receiving PBT who are noncompliant with desferoxamine was reduced to 25%, the difference in expected QALYs between the decision to consider BMT and the decision to consider PBT was less than 0.01 QALYs. In addition, improvement in compliance with iron chelation in combination with improvement in compliance with the transfusions made PBT the preferred choice.

Increasing donor availability and the acceptance of BMT for patients with available donors slightly improved outcomes associated with the intent to transplant. However, the relative change in overall QALYs was still relatively small. For example, changing the probability of donor availability from 0.14 to 0.30 changed the expected QALYs from 15.2 to 15.6 (a 2.6% increase). Given the degree of variability associated with the expected QALYs, this is still not significantly greater than outcomes associated with the intent to transfuse.

The 1-year risk of death associated with BMT was varied from 3% to 12%. Higher probability of death from BMT was associated with diminished expected QALYs under the intent-to-transplant arm; however, the expected QALYs were still relatively close between the 2 intention-to-treat arms, regardless of the actual value chosen for the BMT mortality risk.

A series of 1-way sensitivity analyses on the quality-of-life utility values (Table 7) revealed that, although the expected number of QALYs is sensitive to changes in some disease state quality-of-life values, the difference between the expected number of QALYs for each treatment strategy remains relatively small across quality of life values ranging from 0 (death) to 1 (perfect health). For example, as the quality of life for patients receiving PBT who are noncompliant with iron chelation varies from 0 to 1, the expected QALYs for the PBT arm range from 9.7 to 16.7, whereas the expected QALYs for the BMT arm range from 10.4 to 16.8. The reason that the expected outcomes for the BMT arm change when quality-of-life values associated with PBT are varied is again due to the fact that so many patients for whom BMT is intended will ultimately not receive BMT, either because of lack of donor availability or refusal of BMT. In addition, as the quality of life assigned to cGVHD varies from 0 to 1, the expected number of QALYs for the BMT arm varies from 15.0 to 15.3, whereas the expected number of QALYs for the PBT arm remains constant at 14.9.

If all patients accepted PBT and were compliant with both the transfusions and with iron chelation, PBT would offer more QALYs than BMT, with or without the discounting of future years of life. Thus, combinations that improve (1) the proportion of patients who accept PBT, (2) the proportion of patients who are compliant with the transfusion regimen, and (3) the proportion of patients who are compliant with iron chelation would all make PBT the optimal choice. For example, if we assume that 95% of patients would accept PBT, 95% of those who accepted to be compliant with the PBT regimen, 95% of those patients to be compliant with iron-chelating therapy, and between 3 and 5 years of transfusion, PBT would be the optimal choice. The benefit provided by PBT would still be small (less than 1 QALY), but the risk of short-term death would still be small in contrast to BMT.

In a series of Monte Carlo simulations, the standard deviations around these estimates of expected QALYs were determined to be quite large (approximately 6 QALYs). Because a large variation exists in some of the point estimates of risk, the magnitude of the variation of the expected QALYs under each strategy is also large.

This decision analysis provides insight into the decision to recommend transplantation or transfusion for patients with sickle cell whose only potential indication for BMT is the high risk of stroke. This study incorporates results from published medical literature and the most recent clinical knowledge of SCD. Although a decision analysis cannot take the place of a randomized clinical trial (RCT), the models generated allow comparisons in treatment strategies in the absence of direct clinical trials, comparing the management strategies. The model identifies key variables that need to be considered in the design of any future clinical trials of BMT and PBT, including compliance with transfusion regimen and with iron chelation, the duration of time necessary for transfusions to sufficiently reduce the risk of stroke, bone marrow donor availability, and BMT mortality. Such a randomized trial of BMT and PBT might not be feasible because of limited BMT donors, low consent rates, reluctance to enroll patients, or other factors. In lieu of such evidence from an RCT, this study may help inform individual decisions and assist in developing practice policies and recommendations.

Our analysis of a policy of intending to treat SCD at high risk of stroke with BMT or PBT indicates that neither policy has a substantial advantage. Both strategies would result in patients expecting, on average, about 16 discounted QALYs after the treatment decision is made. However, the series of Monte Carlo simulations demonstrated that the standard deviations around these estimates of expected QALYs were quite large (approximately 6 QALYs). In other words, because of the uncertainty in some probabilities used in the model, no strategy can be singled out as the “best.” In addition, outcomes in hypothetical cohorts of 100 000 patients with SCD, followed for 20 years, were remarkably similar under the 2 strategies, with only 0.1% more patients alive under the intent-to-transplant strategy compared with the intent-to-transfuse strategy. The choice of optimal strategy is not sensitive to the risk of BMT mortality, the duration of PBT, and other study variables, including those listed in Table 6.

In an editorial entitled “The Toss-Up,” Kassirer and Pauker25 address the issue of 2 clinical strategies with quite similar expected outcomes and emphasize the need for decision analysis to highlight such occurrences. The authors note that, when the error in calculated utilities exceeds their difference, one cannot choose any 1 strategy with a high degree of confidence.

In the analysis of patients receiving the specific treatment regimens, patients undergoing BMT could expect 18.5 discounted QALYs, compared with 15.7 QALYs for the patients receiving PBT. Patients who were compliant with PBT and chelation had the best outcomes and could expect 19.2 discounted QALYs, compared with patients who were noncompliant with iron chelation and the transfusion regimen and patients who refuse BMT and PBT, who could expect 14.1, 14.8, and 12.3 discounted QALYs, respectively.

As shown in Table 5, patients who are compliant with PBT will have outcomes that are substantially better than those undergoing BMT. These compliant patients are much less likely to have strokes and/or to die than patients who are either noncompliant with the transfusion regimen or with iron chelation. Thus, future efforts to increase compliance among patients undergoing PBT may have a great impact on patient outcomes. Furthermore, if a shorter duration of transfusion (eg, 3 to 5 years) is required to diminish the risk of stroke, BMT has no advantage over PBT. Because the optimal period of transfusion required to decrease stroke risk is unknown, future efforts should also be concentrated on determining this duration.

Our study should be viewed in the context that the results are based on a population model. There may be an instance in which a given individual is unlikely to be compliant with PBT and thus may have a much better chance for success with BMT. There may also be an instance in which there are specific factors such that an individual SCD patient may not do well with transplantation, in which case PBT would have a better outcome. In addition, preferences of some parents of patients with SCD might reflect a set of quality-of-life values that are quite different from what has been modeled for a population of patients with SCD. In that case, the difference between the expected value of the QALYs for each treatment strategy could be much more pronounced. Researchers have been cautioned in the past about making individual treatment decisions based on group-level utilities.26 A survey in 1991 demonstrated that 37% of the parents of children with SCD were willing to accept a mortality risk associated with BMT as high as 15%, and 54% of the parents were willing to accept some risk.27 This variability in acceptance of BMT highlights the necessity of taking into consideration the parents' preferences into these types of treatment decisions.

These results may also be limited in that values for certain model parameters were based on clinician opinion and not on empirical evidence. For example, physicians' and other health care practitioners' utility estimates of quality of life have been shown to overestimate those estimates made by the parents of extremely low-birth-weight (ELBW) children and estimates made by adolescents who were ELBW themselves.28 Although the sensitivity analyses were performed to cover a range in clinicians' estimates, the results are, nevertheless, limited by the validity of the estimates used in the analysis. The uncertainty present in these estimates precludes making recommendations that either treatment is superior to the other.

This analysis has highlighted several areas of research that may help improve future decisions regarding recommendations for BMT and PBT. For example, if outcomes associated with BMT were significantly improved and were available to a larger subset of patients, the intention to transplant these patients would likely be the proper management strategy. Determining the length of time necessary for these patients to be transfused would also play an important role in assisting in this decision-making process. Other important research areas include improving compliance with the transfusion regimen and with iron-chelation therapy and determining quality-of-life utility values for SCD (from the perspective of both patient and parent).

Our analysis shows that BMT does not have a significant advantage over PBT and thus cannot be considered the optimal strategy for these patients. In an editorial on PBT for SCD patients at high risk of stroke, Cohen points out that 40% of patients identified as having high stroke risk by TCD measurement will likely never have a stroke.29 Transplantation, with high initial mortality, would appear to be even less favorable from this perspective. Although patients with SCD who meet current criteria for BMT may reasonably receive BMT, we recommend against elevated CBV being the only criterion for BMT.