Revised Acute Myeloid Leukemia Composite Model Using the 2017 European LeukemiaNet Risk Classification

Abstract

In this cohort modeling study, the acute myeloid leukemia composite model is updated using the European LeukemiaNet 2017 classification for cytogenetic and/or molecular risks.

We previously developed and validated the acute myeloid leukemia composite model (AML-CM).¹ Our work preceded the development of the European LeukemiaNet (ELN) classification (version 2017) for cytogenetic and/or molecular risks.² Herein, we revise the design and validation of the AML-CM by using the ELN classification (version 2017). The 2017 ELN classification differs from the 2010 ELN classification used in our original article,¹ principally by (1) defining only 3 risk groups (“favorable,” “intermediate,” and “adverse”) rather than 4 groups (with intermediate divided into “intermediate 1” and “intermediate 2”) and (2) including several mutations, such as RUNX1, ASXL1, and TP53, while classifying FLT3 internal tandem duplications according to the proportion of involved alleles (ie, allelic ratio).

Methods

To update the model, we followed the Enhancing the Quality and Transparency of Health Research (EQUATOR) reporting guidelines using the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) criteria.^3,4 In general, we used the same methods in this cohort modeling study as those used in the original classification reported in our earlier article¹; for example, we used the same training set (n = 733) to generate new weights for the AML-CM 2017 (Table 1). Median follow-up for the training set was 39 months, and median follow-up for the validation set was 33 months.

Table 1. Revised Multivariate Analysis of Associations Between Individual Comorbidities and Other Covariates With Post-Initial Therapy Mortality (288 Deaths Over 1 Year After Initial Therapy).

Variables	HR (95% CI)	Assigned Score for AML-CM 2017	P Value
Cardiac	1.6 (1.1-2.2)	1	.01
Diabetes	1.1 (1.7-1.6)	0	.64
Hepatic	1.4 (1.0-1.8)	1	.04
Infection	1.3 (1.0-1.8)	1	.10
Peptic ulcer	1.4 (0.8-2.5)	1	.20
Renal
Mild	1.1 (0.8-1.7)	0	.51
Moderate/severe	1.0 (0.7-1.5)	0	.99
Prior malignancy	1.2 (0.9-1.6)	0	.25
Heart valve disease	1.5 (0.8-2.7)	1	.16
Hyperlipidemia	0.9 (0.7-1.3)	0	.68
Hypertension	1.1 (0.8-1.5)	0	.51
Albumin, g/dL
<4-3.5	1.2 (0.9-1.7)	0	.28
<3.5-3	1.3 (0.9-1.9)	1	.11
<3.0	1.6 (1.1-2.5)		.03
Platelet count, ×103/µL
<100-50	1.0 (0.7-1.4)	0	.88
<50-20	1.0 (0.7-1.4)	0	.94
<20	1.2 (0.8-1.8)	0	.35
LDH level, U/L
>200-500	1.7 (1.2-2.5)	1	.01
>500-1000	1.8 (1.2-2.7)	1	.01
>1000	2.4 (1.5-3.9)	2	<.001
Sex
Male	1.0 (0.8-1.3)	0	.73
Female	1.0 [Reference]	0	NA
Age, y
0-49	1.0 [Reference]	0	NA
50-59	1.7 (1.1-2.6)	1	.01
60-69	2.0 (1.3-3.0)	2	.002
≥70	2.5 (1.5-4.0)	2	<.001
Cytogenetic/molecular risks per ELN 2017
Favorable	1.0 [Reference]	0	NA
Intermediate	1.8 (1.2-2.8)	1	.01
Adverse	3.7 (2.4-5.7)	3	<.001
Initial regimen intensity
Low	1.6 (1.1-2.3)	NA	.02
Intermediate	1.0 [Reference]	NA	NA
High	1.3 (0.9-1.9)	NA	.14

Abbreviations: AML-CM, acute myeloid leukemia composite model; ELN, European LeukemiaNet; HR, hazard ratio; LDH, lactate dehydrogenase; NA, not applicable.

SI conversion factors: To convert albumin to grams to per liter, multiply by 10; to convert LDH to microkatals per literal, multiply by 0.0167.

Results

In this updated analysis, platelet counts of less than 20 × 10³/µL did not acquire a weight because the hazard ratio (HR) dropped to 1.2 (95% CI, 0.8-1.8) compared with an HR of 1.3 (95% CI, 0.9-2.0) in the original analysis. Additionally, an adverse ELN 2017 risk classification acquired a weight of 3 given a new HR of 3.7 (95% CI, 2.4-5.7) compared with the original HR of 2.8 (95% CI, 1.9-4.3) using the ELN 2010 in the original model. We also used the same validation set (n = 367) to validate the AML-CM 2017 and compared it with the original AML-CM (Table 2). Overall, the 2 models performed similarly, with slight improvement in discriminative capacity per C-statistic estimate and area under the curve (AUC) for a 1-year mortality rate in favor of the AML-CM 2017.

Table 2. Comparisons of the Performance of the Original AML-CM vs the AML-CM 2017.

Performance Metric	Original AML-CM		Revised AML-CM
	No. of Patients	Estimate	No. of Patients	Estimate
C statistic for 1-y mortality (SD)	292	0.719 (0.022)	292	0.728 (0.021)
True AUC for 1-y mortality (SD)	277	0.758 (0.030)	277	0.773 (0.026)
True AUC for 2-mo mortality (SD)	283	0.776 (0.035)	283	0.774 (0.035)
Overall survival at 1 y, % (95% CI)
AML-CM scores 1-4	78	84 (76-92)	81	85 (77-93)
AML-CM scores 5-6	71	65 (53-76)	62	67 (55-79)
AML-CM scores 7-9	90	52 (41-62)	92	51 (41-62)
AML-CM scores ≥10	53	21 (9-33)	57	21 (10-32)

Abbreviations: AML-CM, acute myeloid leukemia composite model; AUC, area under the curve.

Discussion

The original AML-CM was designed to account for both patient overall health, as measured by age and comorbidity burden, as well as AML aggressiveness, as measured by cytogenetic/molecular risks. The AML-CM 2017 does not statistically or substantively exceed the performance of the original AML-CM; however, this revision is important because it uses the more recent ELN classification. This revision matches the general direction of continuing refinement of classification of the aggressiveness of AML per different mutations.⁵ The revised model ensures that the AML-CM 2017 is appropriate to use in the current era. We have updated the calculator to reflect the addition of the ELN 2017 version (http://www.AMLCompositeModel.org).