Acute myeloid leukaemia (AML) is a heterogenous disease characterized by a wide variety of morphologic ( Bennett et al, 1976 ; Cheson et al, 1990 ), cytogenetic ( Grimwade et al, 1998 ), immunophenotypic ( Knapp et al, 1994 ) and other parameters ( Wood et al, 1994 ; Hart et al, 1994 ). It is well known that certain features confer a good or poor prognosis ( Wheatley et al, 1996 ; Goldstone et al, 1996 ). Categorization of patients into risk groups is clinically important if it can be used to assist in making therapeutic decisions. In particular, reliable identification of a group of patients at very low risk of relapse would enable selective use of bone marrow transplant (BMT) which would not be indicated for the low-risk group because the high toxicity and cost of the procedure would outweigh the likely benefits. At the other extreme, identification of a very poor risk group for whom conventional treatment is ineffective could allow them to be offered alternative, perhaps novel, therapies. For a prognostic index to be widely used to direct therapy, it needs to be simple so that it can be applied easily by the many clinicians who will be making these decisions. We present here the prognostic index that has been developed based on the results of the AML 10 trial and which is currently being used in the AML 12 study. It is based on just three parameters — the response status after course 1, the cytogenetic result and French–American–British (FAB) type ( Bennett et al, 1976 ; Cheson et al, 1990 ) — so is easy to use in clinical practice.
Data on 1711 patients, aged up to 55 years, in the MRC AML 10 trial were used to create a prognostic index for use in risk-directed therapy decision making for younger patients with acute myeloid leukaemia (AML). Two parameters, response after course 1 and cytogenetics, were strongly predictive of outcome. For patients with complete remission, partial remission and resistant disease, 5-year survival from the start of course 2 was 53%, 44% and 22% and relapse rates were 46%, 48% and 69% respectively, and for patients with favourable, intermediate and adverse karyotypic abnormalities, survival was 72%, 43% and 17% and relapse rates were 34%, 51% and 75% respectively (all P < 0.0001). Patients with FAB type M3 but no cytogenetic t(15;17) also had a low relapse rate (29%). These three factors were combined to give three risk groups: good (favourable karyotype or M3, irrespective of response status or presence of additional abnormalities), standard (neither good nor poor), poor (adverse karyotype or resistant disease, and no good-risk features). Survival for these three groups was 70%, 48% and 15% respectively and relapse rates were 33%. 50% and 78% (both P < 0.0001). The index is simple (based on just three parameters), robust (derived from 1711 patients), highly discriminatory (55% survival difference between good and poor risk) and validated, so can be applied in the clinical setting to assist with therapeutic decisions as in the current AML 12 trial.
PATIENTS AND METHODS
The AML 10 protocol has been described in detail elsewhere ( Hann et al, 1997 ) and the treatment schedules are summarized in Fig 1. Ablative therapy for both allogeneic BMT (allo-BMT) and autologous BMT (A-BMT) in first remission consisted of cyclophosphamide and total body irradiation (TBI). There were minor variations in the protocol for children (aged 0–14 years): triple intrathecal therapy with methotrexate, cytosine arabinoside and hydrocortisone was administered after each course to children without CNS disease at presentation; for children aged <1 year all chemotherapy doses were reduced by 25%; for children aged <2 years, ablative therapy for BMT was busulphan and cyclophosphamide.
AML 10 was a trial which was designed for patients up to the age of 55. Patients with any form of de novo or secondary AML (s-AML, defined as having a preceding myelodysplastic phase, a previous haematological disorder or following prior chemotherapy or radiotherapy for non-haematological disorders) were eligible, as long as they had not received any prior cytotoxic therapy for leukaemia and had given informed consent to take part in the trial. Patients with CNS disease at presentation were eligible. Patients in blast crisis of chronic myeloid leukaemia were not eligible.
Complete remission (CR) was defined by a normocellular bone marrow containing <5% blast cells and showing evidence of normal maturation of other marrow elements. Partial remission (PR) was defined as 5–15% blasts. Patients with <5% blasts but with a hypocellular marrow precluding CR were also classified as being in PR. Resistant disease (RD) was defined as >15% blasts in the bone marrow. The bone marrow for the assessment of response status after course 1 was scheduled for 2 weeks after the completion of the course, with a repeat marrow being performed after a further week if the initial one was not evaluable.
The following endpoints were defined: survival was the time from the start of the second course of chemotherapy to death; for patients who achieved CR, disease-free survival (DFS) was the time from CR to first event (whether relapse or death in CR) and relapse rate was the actuarial risk of relapse from CR (censoring at death in CR). All percentage values given in the text for these endpoints are at 5 years.
Contingency tables and chi-squared tests were used to assess the relationships between parameters. For analysis of the survival, DFS and relapse rate endpoints, Kaplan-Meier life-tables were constructed and were compared by means of the log-rank test ( Peto et al, 1977 ). Cox regression was used for multivariate analysis. All P-values are two-tailed. Surviving patients were censored at the date of last complete follow-up which was 30 June 1998, apart from 29 patients who were lost to follow-up and who were censored at the date they were last known to be alive.
In total, 1966 patients were entered into AML 10 between May 1988 and April 1995, of whom 255 have been excluded from the current analyses for the following reasons: 26 were subsequently rediagnosed as not having AML; 172 died before, during or just after course 1 and therefore did not receive course 2; 48 were alive but did not receive further therapy after course 1; and 9 for whom neither data on status after course 1 nor a diagnostic cytogenetic result are available. Thus, the analyses presented here are based on 1711 patients. 1 Table I gives details of the presentation features of the patient population. Median follow-up of surviving patients is 6.0 years (range 0.4–10.1 years).
Survival from the start of course 2 was 46% in the entire study population. Among the 1567 patients who achieved CR at any stage, DFS was 43%, the relapse rate was 49%, and 13% of patients died in CR (7% during consolidation chemotherapy, 5% following BMT, and <1% some time after completing chemotherapy).
Response status after course 1
Patients were placed in three groups based on their response status after course 1 (status was not known for 54 patients). Of the patients with known response status, the majority (n = 1110, 67%) were in CR, with 17% (n = 277) in PR and 16% (n = 270) with RD. The division between PR and RD was based on the standard definition of PR as 5–15% blasts. The data confirmed that this was an appropriate cut-off: patients with 10–15% blasts did much better than those with 15–20% blasts whose outcome was as poor as that of patients with more than 20% blasts (survival 42%, 23% and 22%, and relapse rates of 46%, 68% and 70% respectively). Patients in PR did a bit worse than those in CR with respect to survival but, if they subsequently achieved CR, there was no difference in relapse rate. Those with RD had an extremely poor prognosis. Survival was 53%, 44% and 22% (P < 0.0001) and relapse rates were 46%, 48% and 69% (P < 0.0001) for the CR, PR and RD groups respectively (Fig 2). 89% of those in PR after course 1 went on to achieve CR, compared to only 65% of those with RD (P < 0.0001). Even if they did achieve CR, the latter group had a very high early relapse rate of 49% in the first year. The risk of dying in CR was independent of response status, with 13%, 12% and 13% of patients with CR, PR or RD respectively dying in CR (P = 0.5), so DFS in the three groups (46%, 44% and 25% respectively) reflected the relapse rate.
A diagnostic cytogenetic result was available for 85% of patients. These patients were classified into three groups. Favourable karyotypes (n = 337, 23%) were those with the abnormalities t(8;21) (n = 117), t(15;17) (n = 170) and inv(16) (n = 50). Patients with these abnormalities were classified as favourable even if other cytogenetic abnormalities were present ( Grimwade et al, 1998 ). Adverse karyotypes (n =137, 9%) were those with monosomy of chromosomes 5 (n = 22) or 7 (n = 48), deletion of the long arm of chromosome 5 [del(5q)] (n = 22), abnormalities of the long arm of chromosome 3 [abn(3q)] (n =33) or a complex karyotype (defined as more than four abnormalities) (n = 69). The total number of individual adverse abnormalities was greater than the total number of patients in this group because some patients had more than one adverse factor. Patients with normal karyotypes (n = 628) or with abnormalities other than those defined as favourable or adverse (n = 355) were placed in the intermediate cytogenetic group. There were substantial differences in survival and relapse rate between the three groups (Fig 3). For the favourable, intermediate and adverse groups respectively, survival was 72%, 43% and 17% (P < 0.0001), and relapse rates were 34%, 51% and 75% (P < 0.0001). The risk of dying in CR was not related to karyotypic group, with 11%, 13% and 9% of patients in the favourable, intermediate and adverse groups respectively dying in CR (P = 0.3), so DFS showed the same pattern as relapse rate at 58%, 41% and 21% respectively.
By combining the response status and cytogenetic groups, nine groups were obtained ( 2 Table II). There was a strong correlation between the two parameters; patients with adverse karyotypes were much more likely to have resistant disease than those with favourable karyotypes. The survival of patients in each of the nine groups is also given in 2 Table II. Patients in the CR + intermediate and PR + intermediate groups had an average prognosis, with survivals of 51% and 41% respectively. Patients with favourable karyotypes had a much better outlook (survival of 61% or more), irrespective of their marrow status, though the number of patients in the RD + favourable group is very small and the survival estimate for this group has wide confidence limits. Patients with either RD (and no favourable karyotype) or adverse cytogenetics did very poorly, with survival of 26% or less.
Of the 257 patients with acute promyelocytic leukaemia (APL, FAB type M3), 66% were known to have the t(15;17) translocation from cytogenetic analysis (and therefore were in the good-risk group), whereas 22% with a successful result showed no cytogenetic evidence of t(15;17) and in 12% a cytogenetic result was not available. APL patients without cytogenetic evidence of t(15;17) had similar DFS (62% v 58%, P = 0.7) and relapse rates (29% v 34%, P = 0.5) to those with favourable cytogenetics.
Three risk groups have therefore been defined ( Table III), with about half (51%) of the patients falling into the standard-risk group, with 28% and 20% in the good and poor risk groups respectively. The survival and relapse rates for these groups differed enormously (Fig 4), with survival of 70%, 48% and 15% and relapse rates of 33%, 50% and 78% respectively. Risk group was unrelated to the likelihood of dying in CR, with 11%, 13% and 11% of good, standard and poor risk patients respectively dying in CR (P = 0.1), so the differences in DFS between the three groups (59%, 43% and 19% respectively) were similar to those for relapse rate.
Dividing the patients into three age groups (children aged 0–14, younger adults aged 15–34 and older adults aged 35+), the differences between risk groups were found at all ages within AML 10 ( 4 Table IV). Since age is an important independent prognostic factor, children had better survival than younger adults in the corresponding risk group, who in turn had better outcome than older adults.
The data presented so far relate to all patients, whether they received conventional chemotherapy alone or whether they received, in addition, BMT (either allogeneic or autologous): 201 and 199 patients underwent allo-BMT and A-BMT in first CR respectively. When patients who were transplanted in first CR were censored at the time of BMT, there were still large differences in relapse rates and survival (both P < 0.0001) by risk group for chemotherapy only patients. The same pattern was seen after BMT, with differences following both allo-BMT (relapse: P = 0.006; survival: P = 0.07) and A-BMT (relapse: P = 0.006; survival: P = 0.0002) ( V. Outcome at 5 Table V).
Univariate analysis of other prognostic factors suggested that no other parameter was comparable in importance with response status or karyotypic group as a determinant of outcome ( 6 Table VI). Multivariate analysis on the 1401 patients for whom data on all six parameters were available indicated that they were all of independent prognostic significance ( Table VI), though in some cases of greatly reduced importance (e.g. WBC and type of AML). Although the P-value for the multivariate analysis may suggest that karyotypic group was of less importance than response status, this result was related to the competitive effect of FAB type which was correlated with karyotype. If patients with FAB type M3 and no cytogenetic t(15;17) were placed in the favourable cytogenetic group on the basis that they have the t(15;17) translocation at the molecular level, karyotypic group and response status were of similar significance (both P < 1 × 10−30). Thus, multivariate analysis confirmed that response status and karyotypic group were the two most important, and independent, prognostic factors.
The very small number (n = 18) of good-risk patients with a WBC > 100 × 109/l had survival of 67% but the standard error was large (S.E. 11). Standard-risk patients with a high WBC had survival of 40%, still much better than that of poor-risk patients as defined here.
Validation of the index
Preliminary data from the MRC AML 12 trial have been used to test the reproducibility of the index. Data on response status were available for 1108 patients, of whom 821 were in CR, 161 in PR and 126 had resistant disease. Cytogenetic results were available for 999 patients: 255 favourable, 636 intermediate and 108 adverse. An additional 33 patients had M3 morphology in the absence of a cytogenetic result. A risk group could be assigned to 986 patients: 288 good risk, 519 standard risk and 179 poor risk. Survival at 2 years for these three parameters (with the corresponding figures for AML 10 in parentheses) were response status: CR 68% (62%), PR 55% (53%), RD 34% (29%); cytogenetic group: favourable 77% (80%), intermediate 62% (54%), adverse 25% (23%); risk group: good 79% (77%), standard 65% (58%), poor 29% (22%). In AML 12 the differences in survival between the three groups within each of the three parameters were all significant at P < 0.0001 and were comparable to those observed in AML 10. Although based on relatively short follow-up, the majority of events occur within the first 2 years, so it is very unlikely that the highly significant differences already observed in AML 12 will be lost with longer follow-up.
We have identified two parameters which, in the context of the MRC AML 10 and AML 12 trials, have very highly significant prognostic value in AML, namely the patient's response status after the first course of therapy and the diagnostic karyotype. Although the curves for cytogenetic groups (Fig 3) and risk groups (Fig 4) appear similar, the inclusion of response status in the index allows an additional 11% of the patients (i.e. those with RD but without either a favourable or an adverse cytogenetic result) to be classified as poor risk, thereby more than doubling the size of the poor-risk group from 9% of patients to 20%. A third parameter, FAB type M3, may be considered a surrogate for the t(15;17) translocation, since all patients with APL have the been shown to have the molecular rearrangement associated with the t(15;17) translocation ( Fenaux et al, 1997 ). From these three parameters we have constructed a simple prognostic index. Despite its simplicity, this index is very strongly predictive of outcome with 5-year survival of the good-risk group being a very good 70%, compared to the very disappointing outcome in the poor-risk group of 15% alive at the same time point, a survival differential between the two groups of 55%. The index is also robust, as it is derived from a large population of 1711 patients. The chi-squared values for the two individual factors and the risk group itself are all >100, giving P values of <1.0 × 10−30, so there can be no doubt that these results are real. Indeed, they have been validated in the AML 12 trial, though there is still a need for more data to confirm reliably the risk group categorization of the groups with small numbers of patients; for example, is the combination of favourable karyotype with resistant disease after course 1, or high WBC, really good risk? This model needs to be flexible and, if evidence accumulates to suggest that some patients have been wrongly assigned, it can be modified at a future date (as, indeed, has already happened [ Wheatley et al, 1995 ]). Furthermore, analyses of prognostic factors in AML identify outcome probabilities as a continuum, so inevitably the cut-off points between risk groups will be, to some extent, arbitrary.
It is likely that the outcome of patients with APL in this study, although good, is nonetheless suboptimal, since the majority were entered before treatment with all-trans retinoic acid (ATRA) became standard (95 patients in the latter part of AML 10 were entered into the MRC's ATRA trial). ATRA substantially improved the outcome of patients with APL ( Fenaux et al, 1993 ; Tallman et al, 1997 ), so current and future patients who are treated with ATRA would be expected to have even better outcome than the overall APL population considered here.
Analysis of other factors indicated that no other parameter was of the same order of importance as response status and cytogenetics. Although WBC and type of AML (i.e. de novo or s-AML) were all univariately significant at P < 0.0001, there was no good evidence to suggest that they should be included in the model since, in multivariate analysis, their relevance is substantially reduced.
As the main purpose of developing the index was to use it to determine therapy in the next MRC trial, we chose a simple model which uses just three parameters and combines these in a manner which is easy to understand. Recruitment to AML 10 came from almost 300 clinicians in 163 hospitals, so the index needs to be simple and, therefore, readily applicable by the large number of doctors who have to make decisions on the choice of treatment for individual patients. Although the model could have been made somewhat more predictive by the inclusion of other factors, the relatively little additional discrimination achieved by doing so would have been at the expense of much greater complexity and, thus, less practical relevance.
This prognostic index cannot be applied at diagnosis, since response status is not known until after the first course of therapy, although there are presentation features which predict resistant disease ( Goldstone et al, 1996 ). Similarly, the time needed to culture samples for cytogenetic analysis means that these data will also not be available at diagnosis, although molecular techniques ( Grimwade et al, 1996 ) could enable the presence of the favourable t(8;21), t(15;17) and inv(16) abnormalities to be ascertained within 24–48 h after presentation. However, given the current therapeutic options, a policy of giving one course of standard induction therapy before assessing the patient's risk group with a view to adapting subsequent therapy based on this evaluation may be as good as any.
As would be expected, the proportion of patients in the three risks groups varies with age: favourable cytogenetic abnormalities are more frequent in younger patients, whereas the incidence of resistant disease and adverse cytogenetic abnormalities increases with age ( Mertens et al, 1993 ). Nevertheless, large differences in outcome by risk group were still found at all ages. Since AML 10 was a trial for younger patients, it provided no information on whether these risk groups would be valid in patients above the age of 55. A full investigation of the relevance of these risk groups in the elderly will be carried out on data from the parallel MRC trial for older patients (AML 11).
We have shown that the three risk groups are independent of the post-remission therapy received and apply whether or not BMT was performed. The analysis presented here can not be interpreted as indicating that BMT is beneficial. Firstly, it is possible that patients receiving a BMT are a selected group with a better intrinsic prognosis. Secondly, relapse rates after BMT will be lower in all risk groups when compared to relapse rates after CR for non-transplanted patients, since the median times from CR to allo-BMT and A-BMT were 170 and 180 d respectively (i.e. nearly 6 months). Thus, patients receiving BMT will already have an improved prognosis by virtue of having remained in CR long enough to reach transplant ( Gray & Wheatley, 1991). This selection factor will apply especially to poor-risk patients who have a very high early relapse rate, so those who reach BMT represent a better risk subset within this group. Assessment of the value, if any, of BMT in AML 10, using appropriate statistical methods, has shown that, although transplant substantially reduces relapse rates, any survival benefit is smaller and takes time to emerge ( Burnett et al, 1996 , 1998).
The design of the current MRC AML 12 trial for adults is based on the prognostic index developed from AML 10. Because of their low relapse risk (and good chance of responding to reinduction therapy if relapse occurs), good risk patients in AML 12 are not recommended for BMT (either allogeneic or autologous), since it is unlikely that any further reduction in relapse risk achieved by the transplant would be worth the acute toxicity and long-term side-effects ( Zittoun et al, 1997 ; Watson et al, 1996 ). Poor-risk patients are now entered after the first course into the current MRC trial for high-risk AML (AML-HR) which is evaluating the roles of alternative therapies (fludarabine, high-dose Ara-C, growth factor and retinoic acid), although it is recognized that these patients may constitute a group for whom none of the currently available therapeutic options are effective. Standard-risk patients, who form the majority, follow a treatment regimen similar to the AML 10 protocol and receive intensive chemotherapy, with or without BMT. Therefore the index developed here has direct relevance to the management of patients with AML.
The following were members of the MRC Adult and Childhood Leukaemia Working Parties during the period of AML 10: N. C. Allan, C. C. Bailey, P. P. H. Barbor, A. Barrett, C. Barton, V. A. Broadbent, A. K. Burnett (Chairman, Adult Leukaemia Working Party), P. J. Carey, J. M. Chessells, J. A. Child, R. E. Clark, R. Collins, M. Cuthbert, P. J. Darbyshire, S. I. Dempsey, N. J. Dodd, I. J. Durrant, O. B. Eden (Chairman, Childhood Leukaemia Working Party), P. Emerson, I. M. Franklin, D. A. G. Galton, B. Gibson, J. M. Goldman, A. H. Goldstone, R. Gray, T. J. Hamblin, I. M. Hann, C. Haworth, M. Hewitt, F. G. Hill, A. V. Hoffbrand, R. M. Hutchinson, F. G. C. Jones, D. J. King, S. E. Kinsey, J. Kohler, I. J. Lewis, J. S. Lilleyman, D. C. Linch, T. J. Littlewood, I. C. MacLennan (Past Chairman, Adult Leukaemia Working Party), M. Madden, J. Mann, J. Martin, W. Maton-Howarth, S. R. McCann, T. J. M. McElwain (Past Chairman, Childhood Leukaemia Working Party), S. T. Meller, D. W. Milligan, C. D. Mitchell, P. Morris-Jones, G. Mufti, B. Murphy, A. C. Newland, A. Oakhill, A. C. Parker, J. Peto, R. Peto, R. L. Powles, A. Prentice, H. G. Prentice, M. Radford, J. K. H. Rees, M. Reid, A. Rejman, S. Richards, S. A. Schey, R. S. Shannon, A. G. Smith, R. F. Stevens, G. P. Summerfield, D. Swirsky, P. J. Tansey, A. Thomas, E. N. Thompson, R. Varcoe, D. Walker, H. Wallace, D. Webb, K. Wheatley, J. A. Whittaker, M. V. Williams, J. A. Yin.
We thank Rachel Clack, Sarah Cullip, Sue Knight and Angela Radley at the MRC Clinical Trial Service Unit, Oxford, for data management, and Cathy Harwood for preparing the manuscript.
Our greatest thanks go to all the clinicians who entered their patients into the AML 10 trial and documented their progress, thereby enabling the analyses presented here and to be undertaken on a very large, and hence unusually reliable, data set.