Drs. Reed and Schulman have served as consultants for Novartis.
Article first published online: 18 OCT 2004
Copyright © 2004 American Cancer Society
Volume 101, Issue 11, pages 2584–2592, 1 December 2004
How to Cite
Anstrom, K. J., Reed, S. D., Allen, A. S., Glendenning, G. A. and Schulman, K. A. (2004), Long-term survival estimates for imatinib versus interferon-α plus low-dose cytarabine for patients with newly diagnosed chronic-phase chronic myeloid leukemia. Cancer, 101: 2584–2592. doi: 10.1002/cncr.20674
Portions of this work were presented at the 39th Annual Meeting of the American Society of Clinical Oncology, May 31–June 3, 2003, Chicago, Illinois.
See accompanying article on pages 2574–83, this issue.
- Issue published online: 16 NOV 2004
- Article first published online: 18 OCT 2004
- Manuscript Accepted: 20 AUG 2004
- Manuscript Revised: 18 AUG 2004
- Manuscript Received: 15 JUN 2004
- Duke University Medical Center
- Novartis Pharmaceuticals Corporation, East Hanover, NJ
- clinical trials;
- cost-benefit analysis;
- chronic myeloid leukemia;
- survival analysis
The authors estimated survival among patients with chronic myeloid leukemia for a cost-effectiveness analysis of imatinib versus interferon-α plus low-dose cytarabine (IFN+LDAC).
Two-year survival and cytogenetic response were determined using data from 553 patients who received first-line imatinib in the International Randomized Interferon versus ST571 Study (IRIS). Long-term survival was modeled on complete cytogenetic response (CCyR) after 2 years. Long-term survival for patients with a CCyR was modeled using data from a cohort study of 317 patients with CCyRs. Long-term survival for patients without a CCyR was modeled using data from a trial of 275 patients who were treated with IFN+LDAC. Computation of lifetime survival estimates for imatinib assumed a proportional hazards relation between survival for an age-matched and gender-matched cohort and survival for patients with and without a CCyR.
For IRIS patients receiving imatinib, the estimated survival was 95.8% and the CCyR rate was 73.8%. The average residual life expectancy was estimated to be 16.71 years for CCyR patients and 5.78 years for non-CCyR patients. The estimated life expectancy after treatment with imatinib was 15.30 years, compared with 9.07 years for patients who were treated with IFN+LDAC in previous studies.
Assuming the relation between CCyR and survival with interferon-α holds for imatinib, higher CCyR rates with imatinib therapy will result in an estimated 6.23 life-years gained compared with treatment with IFN+LDAC. Cancer 2004. © 2004 American Cancer Society.
In the 1990s, interferon-α (IFN-α) was the treatment of choice for patients with chronic myeloid leukemia (CML) who were not candidates for allogeneic stem cell transplantation.1–3 More recently, imatinib mesylate (Gleevec™; Novartis, Basel, Switzerland) has shown promising results in clinical trials across all phases of CML.4–6 Currently, the International Randomized Interferon versus STI571 Study (IRIS) is comparing imatinib with IFN-α plus low-dose cytarabine (IFN+LDAC).7 With a median follow-up of 19 months, imatinib has been associated with higher rates of hematologic and cytogenetic response and lower rates of disease progression. However, to our knowledge, the economic impact of adopting imatinib as first-line therapy for CML is unknown. Moreover, estimates of life-years gained, which are necessary to evaluate cost-effectiveness, have not to our knowledge been calculated.8
Several clinical trials involving CML have shown that cytogenetic response is associated with survival.9–12 To our knowledge, complete cytogenetic response (CCyR) has been observed in relatively few patients with CML. In previous studies, observed rates of CCyR were typically < 20% for patients receiving IFN-α,9–12 and CCyR was seldom observed after treatment with busulfan or hydroxyurea.11, 12 In the IRIS study, rates of CCyR among patients randomized to receive imatinib (73.8%) were more than twice as high as rates for patients randomized to treatment with IFN+LDAC (30.2%) in an intent-to-treat analysis of best cytogenetic response (conducted through July 2002).7
For our cost-effectiveness analysis of imatinib versus IFN+LDAC (see also the accompanying article by Reed et al.13), we sought to estimate life-years gained using previously published data regarding the relation between survival and cytogenetic response. For our survival model, we conducted a modified landmark analysis, using observed rates of CCyR from the IRIS study, to predict long-term survival for a cohort of patients with newly diagnosed chronic-phase CML who were receiving imatinib as first-line therapy.
MATERIALS AND METHODS
Our survival analysis relied on data from 1) the IRIS study, which compared imatinib with IFN+LDAC7; 2) two clinical trials comparing IFN-α with IFN+ LDAC14, 15; and 3) two cohort studies that included patients who achieved a CCyR while receiving IFN+LDAC.16, 17 For the IRIS study, we had access to patient-level data. For the other studies, we relied on published reports.
The IRIS is a randomized trial comparing imatinib with IFN+LDAC, with 553 patients randomized to each arm.7 Patients with newly diagnosed chronic-phase CML were eligible to participate. With the exception of hydroxyurea and anagrelide, patients had to have been previously untreated for CML. A CCyR was defined as a finding of 0% Philadelphia-chromosome positive (Ph+) cells in metaphase in a sample of bone marrow. Analysis used IRIS data collected through July 2002.
The Italian Cooperative Study Group on CML (ICSG) conducted a clinical trial comparing IFN-α with IFN+LDAC.15 Patients were eligible to participate if they had Ph+ CML in the chronic phase. With the exception of busulfan and hydroxyurea, patients were ineligible to participate if they had received previous treatment for CML. A total of 275 patients were randomized to receive IFN+LDAC.
The French Chronic Myeloid Leukemia Study Group (FSG) also conducted a clinical trial comparing IFN-α with IFN+LDAC.14 Three hundred sixty patients were randomized to receive IFN+LDAC. Patients were eligible to participate if they were diagnosed with Ph+ CML in the chronic phase. Patients could have received hydroxyurea previously.
The two cohort studies16, 17 examined long-term survival among patients with CML after achieving a CCyR. The first study, conducted by the European Study Group on Interferon in Chronic Myeloid Leukemia (ESG),16 included 317 patients who achieved a CCyR after receiving IFN-α regimens. The second cohort study, conducted at the M. D. Anderson Cancer Center (MDACC), included 140 patients with CML who achieved a CCyR.17 The MDACC patients originally were treated with IFN-α–based regimens.
We compared the baseline characteristics of patients in IRIS who were randomized to receive imatinib with the characteristics of patients in the ICSG and FSG studies who were randomized to receive IFN+LDAC. We performed treatment comparisons on categoric variables using chi-square tests, and we conducted tests of significance for age using Student t tests, comparing IRIS data with randomly generated data with means and standard deviations (SDs) corresponding to the published reports. Similarly, we compared baseline characteristics of IRIS patients randomized to receive imatinib who achieved a CCyR with the characteristics of patients in the ESG and MDACC cohort studies.
Model Structure and Assumptions
Similar to a landmark analysis,18 we estimated long-term survival by conditioning on an intermediate response at a landmark time point. Because of the presumed relation between cytogenetic response and survival, we selected CCyR as the intermediate response. We used a landmark time point of 2 years because most patients receiving IFN+LDAC who achieve a CCyR do so within 2 years.16, 17 According to the principles of landmark analysis, we divided our population into 3 disjoint groups: those who died before 2 years, those who were alive and had achieved a CCyR at 2 years, and those who were alive but had not achieved a CCyR at 2 years.
The objective of the analysis was to estimate long-term survival among patients who were randomized to receive imatinib but could switch to treatment with IFN+LDAC on disease progression. This first-line imatinib strategy corresponds to the regimen received by IRIS patients who were randomized to treatment with imatinib. The comparator, which we have termed the historical IFN+LDAC strategy, was initial treatment with IFN+LDAC without the possibility of switching to imatinib. This regimen is representative of therapy in the preimatinib era.
Because to our knowledge there are no long-term data regarding treatment with imatinib, the validity of any estimate of long-term survival depends on the validity of several assumptions. The key assumption in our base–case cost-effectiveness model was as follows. Among patients alive at 2 years who were receiving either therapy, the survival of patients with a CCyR was found to be similar to the survival of patients in the ESG study with a CCyR, and the survival of patients without a CCyR was found to be similar to the survival of patients in the ICSG trial who were without a CCyR.
For each treatment strategy, we calculated the overall survival curve and derived the area under the curve (AUC), a measure of mean survival. For the first-line imatinib strategy, we used the Kaplan–Meier all-cause survival estimate from the IRIS trial to calculate the 2-year AUC. For the historic IFN+LDAC strategy, we obtained 2-year and 5-year AUC estimates by applying digitization software to published survival curves for patients receiving IFN+LDAC in the ICSG and FSG studies (UnGraph, version 4.0; Biosoft, Ferguson, MO).
For patients alive at 2 years, we posited separate hazard functions depending on CCyR status. For the CCyR cohort, we assumed a proportional hazards model with the hazard function
in which λ0(t) is the baseline hazard corresponding to an age-matched and gender-matched cohort from the general U.S. population, and θCCyR is the hazard ratio comparing the survival of CCyR patients with the age-matched and gender-matched cohort.19 Similarly, for the non-CCyR cohort we assumed a proportional hazards model with the hazard function
in which θnon-CCyR is the hazard ratio comparing the survival of non-CCyR patients with the age-matched and gender-matched cohort. By weighting life-table estimates for the general U.S. population according to the baseline age and gender distributions of the IRIS population, we obtained the baseline hazard (or survival) function corresponding to λ0(t).20 To limit the right-hand tail of the survival distribution, we assumed that no one in the matched population lived beyond age 100 years.
To estimate the hazard ratio parameters, θCCyR and θnon-CCyR, we followed methods proposed by Parmar et al.21 For both treatment strategies, we calculated the expected survival by combining the observed survival for the initial 2 years with the projected survival for the CCyR and non-CCyR cohorts.
In the survival component of the cost-effectiveness model, we conducted several sensitivity analyses. To determine the generalizability of our findings, we modeled the survival of the CCyR cohort using data from the MDACC study and compared the results with the base–case estimates from the ESG study. Similarly, we modeled the survival of the non-CCyR cohort using data from the FSG study. Across all levels of baseline risk according to Sokal et al.22 and Hasford et al.,23, 24 first-line imatinib appears to induce a much higher rate of CCyR than historic IFN+LDAC. As a result, patients who achieve a CCyR after receiving imatinib most likely will have a worse baseline risk compared with patients who achieve a CCyR after receiving IFN+LDAC. We examined two methods to quantify the potential bias caused by this confounding. In the first analysis, we conducted simulation studies in which patients receiving imatinib had 50% higher mortality hazard rates compared with patients receiving IFN+LDAC in both the CCyR and non-CCyR cohorts. In the second analysis, we assumed that only patients who achieved a CCyR after receiving imatinib and who had low risk profiles as determined by Sokal et al.22 and Hasford et al.23, 24 experienced a survival benefit compared with patients who did not achieve a CCyR. In the current study, patients at intermediate or high risk and who were treated with imatinib were assumed to have a survival rate that was similar to that of patients in the non-CCyR cohorts, even if they achieved a CCyR within 2 years.
We have assumed that survival is independent of treatment, conditional on CCyR status. Using IRIS data, we sought to verify this assumption. To attempt to identify a treatment effect among patients not achieving a CCyR, we applied a log-rank test comparing the treatments in which patients were censored at the time of CCyR. To examine the treatment effect among CCyR patients, we used a log-rank test to compare patient survival from time of the first CCyR. Both tests compared groups based on the randomization therapy.
Table 1 shows key prognostic variables for IRIS patients randomized to receive imatinib compared with the baseline characteristics of patients in the ICSG and FSG studies who were randomized to receive IFN+LDAC. Primarily for administrative reasons, > 30% of IRIS patients did not have baseline risk scores available. Compared with IFN+LDAC patients from the ICSG study, IRIS patients were older and had a higher baseline risk according to the model of Hasford et al.23, 24 Compared with IFN+LDAC patients in the FSG study, IRIS patients were more likely to be men and to have higher risk scores according to the model of Sokal et al.22
|Characteristic||IRIS patients randomized to receive imatinib (n = 553)||ICSG patients randomized to receive IFN + LDAC (n = 275)||P Valueb||FSG patients randomized to receive IFN + LDAC (n = 360)||P valueb|
|Mean age (yrs) (SD)||48 (13)||45 (13)||<0.001||—||—|
|Median age (yrs) (range)||50 (18–70)||—||—||50 (7–71)||>0.99|
|Male gender||342 (62)||162 (59)||0.42||195 (54)||0.02|
|Sokal et al. risk group22|
|Low||201 (53)||139 (50)||168 (47)|
|Intermediate||111 (29)||82 (30)||142 (39)|
|High||71 (19)||54 (20)||0.88||50 (14)||0.008|
|Hasford et al. risk group23, 24|
|Low||171 (46)||157 (57)||—|
|Intermediate||166 (44)||92 (33)||—|
|High||38 (10)||26 (10)||0.01||—||—|
According to IRIS data, of the 553 patients randomized to treatment with imatinib, 408 (73.8%) achieved a CCyR. Table 2 shows the characteristics of these patients and of those in the previous CCyR cohorts. IRIS patients were similar to ESG patients, except with regard to the risk score established by Hasford et al.23, 24 More IRIS patients had intermediate or high risk scores. Summary statistics for age, gender, and Hasford et al.23, 24 risk score were not available for the MDACC cohort. Sokal et al.22 risk scores were available, and IRIS patients who attained a CCyR were judged to be sicker than MDACC patients at baseline.
|IRIS patients randomized to receive imatinib (n = 408)||ESG patients (n = 317)||P Valueb||MDACC patients (n = 140)||P valueb|
|Mean age (yrs) (SD)||48.6 (12.8)||47.4 (13.0)||0.21||—||—|
|Median age (yrs) (range)||51 (18–70)||49 (9–73)||—||—||—|
|Male gender||248 (61)||180 (57)||0.28||—||—|
|Sokal et al. risk group22|
|Low||165 (58)||179 (62)||105 (75)|
|Intermediate||81 (28)||76 (26)||24 (17)|
|High||40 (14)||35 (12)||0.60||11 (8)||0.002|
|Hasford et al. risk group23, 24|
|Low||129 (46)||164 (56)||—|
|Intermediate||127 (45)||101 (35)||—|
|High||24 (9)||26 (9)||0.01||—||—|
We compared published survival data from previous CCyR cohorts with survival data from the IFN+LDAC studies, as shown in Figure 1. The time scale (horizontal axis) for the CCyR cohorts corresponds to the time from first CCyR, whereas the time scale for the IFN+LDAC studies represents the time from randomization. Across 10 years of follow-up, CCyR survival rates rarely differed by > 5%, and both 10-year survival probabilities were > 70% (72.8% in the ESG study and 74.7% in the MDACC study). For the ICSG and FSG studies, the observed IFN+LDAC survival curves crossed several times in the 5 years of follow-up; estimated 5-year survival probabilities were 68.7% and 73.2%, respectively. The 10-year survival probabilities for the CCyR cohorts were equal to or better than the 5-year survival probabilities for the IFN+LDAC patients. Figure 1 is not a direct survival comparison of CCyR and non-CCyR patients because some IFN+LDAC patients achieved a CCyR during the course of treatment. In particular, at least 14.2% of patients in the ICSG and 14.8% of patients in the FSG achieved a CCyR in the 2 years after randomization.
Long-Term Survival by Cytogenetic Response Status
Table 3 summarizes the calculation of hazard ratios describing the survival difference between the age-matched and gender-matched U.S. cohort and the CCyR and non-CCyR cohorts. The overall hazard ratio comparing the ESG cohort with the age-matched and gender-matched group was 3.46, indicating that patients who achieved a CCyR had a 246% increased risk of death compared with patients without CML. The hazard ratio was found to be reasonably constant across time intervals, with a minimum of 2.97 and a maximum of 3.99. In calculations not shown, the hazard ratio comparing the CCyR cohort in the MDACC study with the age-matched and gender-matched group was 3.71.
|Cohort||Years from first CCyR|
|ESG cohort with CCyR|
|Effective no. at riska||317.0||158.8||54.8|
|Effective no. of deaths||20.0||16.1||7.4|
|Cohort matched by age and genderb|
|No. at riska||100,000||98,053||95,559|
|No. of deaths||1947||2494||4364|
|Overall HR (95% CI)c||3.46 (2.62–4.56)|
|Years from randomization|
|ICSG cohort without CCyR|
|Effective no. at riska||186.2||155.3||79.8|
|Effective no. of deaths||23.9||18.4||9.4|
|Cohort matched by age and genderb|
|No. at riska||100,000||99,405||98,758|
|No. of deaths||595||648||704|
|Overall HR (95% CI)d||19.4 (14.8–25.1)|
The overall hazard ratio comparing the non-CCyR cohort in the ICSG with the age-matched and gender-matched group was 19.4. Therefore, non-CCyR patients appear to have an increased (instantaneous) risk of death of approximately 1840% compared with the general population. Survival for the non-CCyR cohort in the ICSG had been adjusted for the survival effect of the 14.2% of patients who achieved a CCyR in the initial 2 years. The hazard ratios were reasonably constant within the 3 time intervals (range, 16.6–21.6). Compared with the CCyR cohort, the non-CCyR cohort had a mortality hazard rate of 5.62 (19.4/3.46). Compared with patients in the ICSG, the non-CCyR cohort in the FSG appeared to have a much lower mortality hazard ratio. The hazard ratio for the non-CCyR cohort in the FSG compared with the age-matched and gender-matched group was 13.9. The survival projections for the CCyR and non-CCyR cohorts, based on being alive at 2 years, are depicted in Figure 2. As in Figure 1, there is clear separation in the survival curves. The separation is slightly larger in Figure 2 because the estimated survival effect of the CCyR patients was removed from the ICSG cohort.
Depending on the modeling assumptions, the mean survival projection after CCyR at 2 years ranged from 16.05–16.71 years, compared with estimates ranging from 5.78–7.27 years for non-CCyR cohorts.
Long-Term Survival by Treatment Strategy
We incorporated our survival estimates into our overall cost-effectiveness model (see accompanying article by Reed et al.13). For each population-level simulation run, we generated survival curves for CCyR and non-CCyR cohorts using normal random variables for the log-hazard ratio. Table 3 shows calculations of the mean and variance of the log-hazard ratio. One thousand patients were simulated for each treatment strategy in each run. For each patient, we generated random uniform numbers to simulate 2-year survival and CCyR rates. Survival was calculated in 3-month intervals. In the initial 2 years, survival and CCyR were assumed to be independent. For patients alive at 2 years, we generated an additional random number and transformed it into a survival estimate according to either the CCyR and non-CCyR survival curve. Results in Table 4 correspond to the mean survival estimate across population-level runs.
|Treatment group||Estimate of mean survival (yrs)|
|Using ICSG to model non-CCyR survival||Using FSG to model non-CCyR survival|
|Using ESG to model CCyR survival||15.30a||15.73a|
|Using MDACC to model CCyR survival||14.69a||15.15a|
|Using ESG to model CCyR survival||9.07b||10.14c|
|Using MDACC to model CCyR survival||8.92b||9.88c|
As shown in Figure 3, the survival curves for first-line imatinib therapy and historical IFN+LDAC were compared with the age-matched and gender-matched cohort. For the first-line imatinib strategy, the 2-year CCyR rate was 73.8%, which corresponds to the best observed CCyR rate of 73.8% reported by O'Brien et al.7 The 2-year CCyR rate for historic IFN+LDAC was 14.2%.15
Summary statistics for the long-term survival projections are shown in Tables 4, 5, and 6. The base–case analysis assumed that CCyR survival was modeled using the ESG data and that non-CCyR survival was modeled using the ICSG data. Under the base–case assumptions, the life expectancy estimates for first-line imatinib and historic IFN+LDAC were 15.30 years and 9.07 years, respectively. The difference of 6.23 years (SD of 0.98 years) corresponds to the area between the survival curves for first-line imatinib and IFN+LDAC as shown in Figure 3. In Table 4, life expectancy estimates for first-line imatinib were reported to vary from 14.69–15.73 years, depending on the populations used to model the CCyR and non-CCyR survival distributions. Similarly, for the historic IFN+LDAC population, life expectancy estimates were reported to vary from 8.92–10.14 years.
|Treatment group||Mean survival (yrs) (Survival estimate, %)||Median lifetime survival (yrs)a|
|Lifetimeab||Years 0–2b||Years 0–5c||Years 0–10ae|
|First-line imatinib (IRIS)||15.30||1.97 (95.8)||4.65 (83.2)a||8.30 (63.1)||13.6|
|Historic IFN + LDAC (ICSG)||9.07||1.96 (95.3)||4.42 (68.7)||6.91 (33.5)||7.4|
|Historic IFN + LDAC (FSG)||10.14||1.95 (92.8)||4.43 (73.2)||7.31 (42.6)||8.7|
|Cohort||Mean survival (yrs) (survival estimate, %)||Median lifetime survival (yrs)|
|Lifetimeb||Years 0–2c||Years 0–5d||Years 0–10e|
|ESG cohort||1.99 (97.6)||4.77 (86.7)||8.70 (72.8)|
|Model based on the ESG cohort||16.71f||1.96 (95.8)f||4.72 (88.3)f||8.77 (72.8)f||16.1|
|MDACC cohort||2.00 (99.2)||4.82 (87.4)||8.88 (74.7)|
|Model based on the MDACC cohort||16.05f||1.96 (95.5)f||4.70 (87.5)f||8.68 (71.0)f||15.5|
|Survival model based on the ICSG cohort||1.78 (78.5)f||3.70 (49.8)f||5.28 (16.8)f|
|Survival model based on the FSG cohort||1.84 (84.0)f||4.01 (60.6)f||6.17 (27.8)f|
Tables 5 and 6 present long-term survival projections along with survival estimates at 2 years, 5 years, and 10 years. The 2-year survival estimate for first-line imatinib was 95.8%, which compared favorably with 2-year estimates from the historic IFN+LDAC populations in ICSG (95.3%) and FSG (92.8%). Under the base–case assumptions, the projected median survival for first-line imatinib was 13.6 years, compared with 7.4 years for historic IFN+LDAC. The difference is due to the large difference noted between the projected survival for the CCyR and non-CCyR cohorts.
Increasing the mortality hazard by 50% for imatinib patients who were alive at 2 years, we estimated the mean life expectancy to be 12.63 years. Estimated life-years gained between first-line imatinib and historic IFN+LDAC remained greater than 3.5 years. If only low-risk patients (based on the score by Sokal et al.22) who achieved a CCyR had a survival benefit, the estimated life-years gained for first-line imatinib was 1.57 years. Tests to determine whether survival depended on treatment given CCyR status were not found to be statistically significant. The log-rank test comparing treatment-specific survival from first CCyR was not found to be statistically significant (P = 0.85). Similarly, the log-rank test censoring individuals at the time of CCyR was not statistically significant (P = 0.74).
We found that patients achieving a CCyR have superior survival compared with typical groups of patients with CML. Using two independent data sources for CCyR and non-CCyR survival, we obtained multiple estimates of life-years gained with highly consistent results. For CCyR patients who were alive at 2 years, we obtained estimates of remaining life expectancy ranging from 16.05–16.71 years. For non-CCyR patients who were alive at 2 years, the estimated remaining life expectancy ranged from 5.78–7.27 years. Combining this information with the large difference noted with regard to rates of CCyR, we predicted an average of 6.23 life-years gained for the first-line imatinib strategy compared with the historic IFN+LDAC strategy.
Based on previous studies of IFN-α, we observed that achieving a CCyR is not sufficient evidence of cure. The estimated life-expectancy for an age-matched and gender-matched population was approximately 31 years, compared with < 19 years for patients who achieve a CCyR. Using CCyR as the intermediate outcome for survival, the life expectancy estimate of 15.30 years for patients treated with the first-line imatinib strategy predicts that CML reduces a typical patient's life expectancy by half. Therefore, although CCyR may be an intermediate endpoint for increased survival, it may not be an intermediate endpoint for evidence of cure, and one could hypothesize alternative intermediate endpoints. One alternative is molecular response, which is considered to be more significant than CCyR for prognosis; however, to our knowledge the long-term benefits of molecular response have yet to be established. Hughes et al.25 found that significantly more imatinib patients enrolled in the IRIS achieved a molecular response compared with patients in the IFN+LDAC group. Among those patients who achieved a CCyR, 57% of patients in the imatinib group achieved a major molecular response within 12 months of treatment compared with 24% of patients in the IFN+LDAC group (P = 0.0003).
The survival model represents our best estimate for average long-term survival in a cohort of patients with newly diagnosed chronic-phase CML who were receiving first-line imatinib treatment. Because of the modeling assumptions and the lack of long-term data regarding imatinib, the model is an approximation based on historical data. Consequently, the model lacks the flexibility and specificity to predict the survival distribution for any particular patient adequately.
Because of the chronic nature of CML and the lack of long-term follow-up in patients treated with imatinib and IFN+LDAC, we do not have empiric evidence describing the length and shape of the tail of the survival distribution. By constructing the age-matched and gender-matched cohort, we were able to describe the survival distribution for a general population. We have assumed that the general population and the CML patients have similar survival distributions, with the exception that the hazard for death should be proportionally higher for patients with CML.
Our survival model has several limitations. The model does not account for the possibility that imatinib-induced mutations could accelerate mortality rates in later years. We have likely introduced a small amount of error by using digitization software. When possible, we have verified our estimates using published data. The reported rate of CCyR in the FSG study is a 1-year estimate rather than a 2-year estimate.14 This underestimation of the 2-year rate may have resulted in a lower hazard rate for the FSG cohort compared with the ICSG non-CCyR cohort. To our knowledge, none of the clinical trial reports listed in Table 1 published the percentage of patients alive with a CCyR at 2 years. Some patients who achieved a CCyR within 2 years of randomization may have lost their response or died before the 2-year mark. For this reason, in the cost-effectiveness model, we assumed that survival and CCyR were independent for the first 2 years.
In a definitive meta-analysis conducted by the Chronic Myeloid Leukemia Trialists' Collaborative Group,2 IFN-α was found to have a statistically significant and substantial survival benefit compared with hydroxyurea or busulfan. IFN-α was predicted to have a 15% higher 5-year survival estimate compared with standard chemotherapy (57% vs. 42%). Similarly, our model predicts a 14.5% higher 5-year survival estimate for imatinib compared with historic IFN-α (83.2% vs. 68.7%). In the short term, readers can accept our life expectancy model as reasonable if they are willing to accept that the benefit of IFN-α compared with standard chemotherapy in CML is similar to the benefit of imatinib compared with historic IFN+LDAC.26 In the long term, data will be available to assess quantitatively the validity of these survival predictions. It is interesting to note that, because of the model assumption that patients continue to receive imatinib until disease progression, the results of the cost-effectiveness analysis were relatively insensitive to changes in the estimated number of life-years gained.13
Assuming that the relation between cytogenetic response and survival holds for patients treated with imatinib, the higher rate of CCyR noted among patients treated with imatinib results in 6.23 life-years gained compared with IFN+LDAC.
The authors thank Damon Seils for editorial assistance and article preparation.
- 8GoldMR, SiegelJE, RussellLB, WeinsteinMC, editors. Cost-effectiveness in health and medicine. New York: Oxford University Press, 1996.
- 10UK Medical Research Council randomised, multicentre trial of interferon-alpha n1 for chronic myeloid leukaemia: improved survival irrespective of cytogenetic response. The UK Medical Research Council's Working Parties for Therapeutic Trials in Adult Leukaemia. Lancet. 1995; 345: 1392–1397., , .
- 19Regression models and life tables. J R Stat Soc B. 1972; 34: 187–220..