Prediction of Survival with Lower Intensity Therapy Among Older Patients with Acute Myeloid Leukemia

Abstract

Background

The aim of this study was to develop a prognostic model for survival in older/unfit patients with newly diagnosed acute myeloid leukemia (AML) who were treated with lower-intensity chemotherapy regimens.

Methods

The authors reviewed all older/unfit patients with newly diagnosed AML who received lower-intensity chemotherapy from 2000 until 2020 at their institution. A total of 1462 patients were included. They were divided (3:1 basis) into a training (n = 1088) and a validation group (n = 374).

Results

In the training cohort of 1088 patients (median age, 72 years), the multivariate analysis identified 11 consistent independent adverse factors associated with survival: older age, therapy-related myeloid neoplasm, existence of previous myelodysplastic syndrome or myeloproliferative neoplasms, poor performance status, pulmonary comorbidity, anemia, thrombocytopenia, elevated lactate dehydrogenase, cytogenetic abnormalities, and the presence of infection at diagnosis, and therapy not containing venetoclax. The 3-year survival rates were 52%, 24%, 10%, and 1% in favorable, intermediate, poor, and very poor risk, respectively. This survival model was validated in an independent cohort. In a subset of patients in whom molecular mutation profiles were performed in more recent times, adding the mutation profiles after accounting for the effects of previous factors identified IDH2 (favorable), NPM1 (favorable), and TP53 (unfavorable) mutations as molecular prognostic factors.

Conclusion

The proposed survival model with lower-intensity chemotherapy in older/unfit patients with newly diagnosed AML may help to advise patients on their expected outcome, to propose different strategies in first complete remission, and to compare the results of different existing or future investigational therapies.

Plain Language Summary

• Lower intensity therapy can be considered for older patients to avoid severe toxicities and adverse events.

• However, survival prediction in AML was commonly developed in patients who received intensive chemotherapy.

• In this study, we have proposed a survival model to guide therapeutic approach in older patients who received lower-intensity therapy.

Introduction

Intensive chemotherapy with cytarabine and anthracyclines is an established curative therapeutic modality in acute myelogenous leukemia (AML).^1–4 Intensive chemotherapy regimens reported 5-year survival rates of 20%–30% in younger patients and 15% or less in older patients (>60 years old) who were judged able to tolerate intensive chemotherapy.^6–11 The recent introduction of novel targeted therapies with venetoclax and with FLT3 and isocitrate dehydrogenase (IDH) inhibitors may improve the results further.^{1–4, 12–18}

Among patients with AML who were older (older than 65–70 years of age) and those judged to have poor tolerance to chemotherapy (older/unfit), the results using intensive chemotherapy were poor with a high early (4-week) mortality of 15%–40%, low complete response rates of 20%–40%, and poor survival rates (median survivals, 6–9 months; 3-year survival rates, <20%).^{1–4, 6–11, 19, 20}

In the early 1990s, understanding the pathophysiology of methylation in cancer resulted in exploring decitabine as a hypomethylating agent (HMA) rather than as a classical cytotoxic chemotherapy drug.^22–28 This research resulted in the Food and Drug Administration approval of azacitidine and decitabine, in 2004 and 2006, respectively, for the treatment of myelodysplastic syndrome (MDS).^26–28 Similar research with these two HMAs resulted in their later approval in Europe for the treatment of older/unfit AML.^{29, 30} Therapy of older/unfit AML with HMAs gradually became the de facto standard of care therapy in older/unfit AML in the United States and throughout the world. More recently, single-arm and later randomized trials established the combination of HMAs and venetoclax, a BCL2 inhibitor, and a superior regimen in older/unfit AML.^31–33 Several studies have retrospectively analyzed the benefits of intensive versus lower intensity therapy in older AML with variable results, some showing lower intensity therapy to be at least equivalent, if not better,^{20, 34–37} others reporting the opposite results.³⁸

Many risk models have been proposed to assess survival and risk of early mortality with intensive chemotherapy in both younger and older patients with AML receiving intensive chemotherapy.^{19, 20, 39–47} No such models with lower intensity chemotherapy have been proposed in a large number of older patients with AML.^48–50 In a recent report, a three-tiered survival model was proposed with three prognostic factors, including the absence of complete response, the presence of ASXL1 mutations, and adverse karyotypes in patients who received HMA and venetoclax.⁵¹

Here, we analyzed all older patients with AML treated at our institution since 2000 with lower intensity chemotherapy, investigated the known prognostic variable associated with survival, and developed and validated a predictive model for survival. Mutational studies, broadly available since 2010, were analyzed in that subset of patients for their additional predictive values. Particular lower intensity chemotherapies were also added as variables to evaluate the possible differential effect of such therapies on survival.

Materials and Methods

Study Group

This analysis included all adults 60 years and older with newly diagnosed AML referenced to our institution from 2000 until 2020 and treated with lower intensity chemotherapy. Patients with acute promyelocytic leukemia (APL) were excluded because they are treated with highly curative nonchemotherapy regimens. Patients with core-binding factor (CBF) AML were also excluded because they are treated with modified intensive regimens (fludarabine, high dose cytarabine, and gemtuzumab ozogamicin) with a potential cure rate above 40%–50%.^{52, 53} Lower intensity therapy was defined as the cumulative dose of cytarabine less than 500 mg/m² during induction therapy.

Patients were treated on AML protocols available during the particular time periods. These are detailed in previous publications.^{29, 31, 34, 48, 49, 54–72} Informed consents were obtained according to institutional guidelines and in accordance with the declaration of Helsinki.

Evaluation of outcomes and statistical considerations

Survival was analyzed from the date of diagnosis until death from any cause. We analyzed all patient and leukemia associated factors known to be associated in the literature with differences in survival. Mutational molecular abnormalities in AML became broadly available since 2010 and were analyzed in the subsets of patients where they were available, after developing the predictive survival models, to evaluate their additional predictive values with lower intensity chemotherapy. Particular therapies were also coded and analyzed to analyze the potential differential effect or benefit of such therapy (e.g., HMA vs. non-HMA; addition of venetoclax) on survival.

Patients were randomly divided (3:1 ratio) into a training group (n = 1088) and a validation group (n = 374). Backward multivariate Cox regression was performed for the prediction of survival.^{73, 74} Multiple imputation was performed for missing variables to reduce bias.⁷⁵ We performed the analyses including and excluding treatment era (2000–2010 vs. 2011–2021). The significant variables by univariate analyses were used to develop a predictive model for survival with lower intensity chemotherapy in older AML. The prognostic factors identified consistently in the two models were then used to propose predictive model in the training group.

We assigned a risk score based on hazard ratios: a hazard ratio (HR) of 1.0–1.4 was assigned a risk score of +0.5; a HR of 1.5–1.9 was assigned a risk score of +1.0; a HR of 2.0–2.4 was assigned a risk score of +1.5; and a HR of 0.50–0.67 was assigned a risk score of −1. There was no HR exceeding 2.5. The second decimal of hazard ratios were rounded to the first decimal place for the development of the prognostic model. The training group was then divided into four risk groups; favorable risk with a total score of –1 to 0 (zero or less); intermediate risk with a total score of 0.5–1.0; poor risk with a total score of 1.5–2.0; and very poor risk with a total score of 2.5 or above. Given the overall poor prognosis in older patients with AML, we divided the whole cohort into four prognostic groups by the results of 1-year survival for practical clinical use. The survival probability was calculated by the Kaplan–Meier method and compared by the log-rank test.^{76, 77}

The predictive model from the training group was then applied in the independent validation group and the additional predictive value of mutations in patients who had mutation profiles (since 2010) and was explored to assess the value of these mutations after accounting for the other pretreatment patient and leukemia characteristics. The accuracy and complexity of the combined models were assessed with concordance index (C-index) and conditional Akaike information criterion (AICc), respectively.⁷⁸ A p value of .05 or less was considered statistically significant. To validate the variable selections, we performed the least absolute shrinkage and selection operator (LASSO) regression to identify optimal number of variables and to validate risk scoring. The accuracy of prediction was assessed with time-dependent receiver operating characteristic at 6 years in the training and validation groups.^{79, 80} Statistical analyses were performed using SPSS software (version 24.0; IBM, Armonk, New York) and R Statistical Software (version 4.0.4; R Foundation for Statistical Computing, Vienna, Austria).

Results

Patient and Leukemia characteristics

A total of 1462 patients were included, 1088 patients in the training set and 374 patients in the validation set (Table 1). Their characteristics are detailed in Tables 1 and S1. Their median age was 72 years (range, 60–94 years).

Table 1.

Patient characteristics

No. (%) / median [IQR]	Overall N= 1462	Training N= 1088	Validation N= 374
Era
2000–2009	527 (36)	396 (36)	131 (35)
2010–2020	935 (64)	692 (64)	243 (65)
Age group
60–64 years	161 (11)	127 (12)	34 (9)
65–74 years	728 (50)	531 (49)	197 (53)
75- years	573 (39)	430 (40)	143 (38)
Age, (year)	72 [68–77]	72 [68–77]	72 [68–77]
Female	862 (59)	647 (60)	215 (58)
Caucasian	1188 (83)	886 (83)	302 (82)
Therapy-related	318 (22)	237 (22)	81 (22)
Antecedent history of dysplasia	271 (19)	192 (18)	79 (21)
ECOG PS
0–1	1006 (70)	764 (71)	242 (66)
2	387 (27)	273 (26)	114 (31)
3–4	45 (3)	34 (3)	11 (3)
Comorbidity
Any	1073 (77)	794 (76)	279 (79)
Renal	116 (8)	79 (8)	37 (11)
Hepatic	31 (2)	27 (3)	4 (1)
Cardiac	910 (65)	668 (64)	242 (69)
Pulmonary	177 (13)	133 (13)	44 (13)
Neurologic	107 (8)	83 (8)	24 (7)
Metabolic	411 (29)	310 (30)	101 (29)
Infection
None	1314 (90)	983 (90)	331 (89)
Fever of unknown origin	33 (2)	24 (2)	9 (2)
Others	26 (2)	18 (2)	8 (2)
Pneumonia	89 (6)	63 (6)	26 (7)

The details of therapies are shown in Table S2. The complete remission (CR) rate was 38% and the overall response rate (CR + marrow CR) was 54%, similar in the training and validation sets. The 30-day mortality was 11% in older patients with lower intensity therapy.

Multivariate Analysis of Variables Associated with Survival

Factors associated with survival (including era of therapy) by multivariate analysis were (Table S3; highlighted in red): older age; therapy-related AML; prior MDS/myeloproliferative neoplasm (MPN) poorer performance status; pulmonary comorbidities; anemia with hemoglobin <8 g/dL; thrombocytopenia with platelets <50 × 10⁹/L; increased lactic dehydrogenase >2 upper limit of normal (ULN); cytogenetics abnormalities; and the presence of infection at diagnosis. Therapy with venetoclax-based regimens was independently favorable.

Excluding era of therapy, the same variables were identified by multivariate analysis, except for metabolic comorbidity and hypoalbuminemia at diagnosis. These are shown in Table 2 (multivariate selected variables highlighted in red).

Table 2.

Univariate and multivariate Cox regression for survival without era: Training (N=1088)

	N	Survival % 1-y / 3-y	Univariate			Backward multivariate*
			P	HR	95% CI	P	HR	95% CI
Age (year)
60 – 64	127	52 / 26	ref	ref	ref	ref	ref	Ref
65 – 74	531	42 / 15	0.004	1.366	1.102–1.693	0.007	1.299	1.075–1.570
75 -	430	36 / 7	<0.001	1.832	1.471–2.281	<0.001	1.693	1.393–2.058
Gender: Female	441	41 / 13	ref	ref	ref
Gender: Male	647	40 / 13	0.560	1.039	0.914–1.180
Ethnicity: Non-white	177	39 / 14	ref	ref	ref
Ethnicity: White	886	41 / 13	0.951	1.005	0.849–1.190
Non-therapy-related	851	45 / 14	ref	ref	ref	Ref	ref	Ref
Therapy-related	237	26 / 8	<0.001	1.451	1.248–1.688	<0.001	1.365	1.194–1.559
No prior MDS/MPN	896	42 / 14	ref	ref	ref	ref	ref	Ref
Prior MDS/MPN	192	32 / 6	0.002	1.285	1.093–1.510	0.006	1.212	1.056–1.391
ECOG performance status
0 – 1	764	45 / 15	ref	ref	ref	ref	ref	Ref
2	273	30 / 9	<0.001	1.408	1.219–1.625	<0.001	1.287	1.133–1.462
3 – 4	34	20 / NA	<0.001	2.384	1.669–3.405	<0.001	1.935	1.408–2.659
Comorbidity
None	254	44 / 17	ref	ref	ref	ref	ref	Ref
Any	794	40 / 12	0.014	1.203	1.038–1.395
Renal**	79	36 / 7	0.001	1.384	1.134–1.688
Hepatic	27	42 / 14	0.020	1.509	1.066–2.136
Cardiac	668	39 / 11	<0.001	1.289	1.130–1.470
Pulmonary	133	38 / 11	0.012	1.247	1.050–1.480	0.012	1.217	1.045–1.417
Neurological	83	43 / 10	0.076	1.197	0.982–1.460
Metabolic	310	39 / 11	0.069	1.132	0.990–1.294
White blood cell count (×106/L)
< 30,000	1001	42 / 13	ref	ref	ref	ref	ref	ref
30,000 –	78	23 / 10	0.017	1.333	1.053–1.687
Absolute neutrophil count (×106/L)**
> 1.0	424	32 / 10	ref	ref	ref
=< 1.0	651	46 / 15	<0.001	0.738	0.650–0.839
Hemoglobin (g/dL)
>10.0	212	52 / 19	ref	ref	ref	ref	ref	Ref
8.0 – 10.0	676	38 / 12	<0.001	1.393	1.181–1.642	0.250	1.088	0.942–1.256
< 8.0	191	37 / 9	<0.001	1.598	1.299–1.965	<0.001	1.428	1.193–1.709
Platelet count (×106/L)
> 100,000	215	52 / 18	ref	ref	ref	ref	ref	Ref
50,000 – 100,000	275	44 / 15	0.021	1.247	1.034–1.504	0.118	1.088	0.942–1.256
20,000 – 50,000	372	40 / 13	0.001	1.364	1.142–1.630	0.001	1.302	1.114–1.522
< 20,000	216	25 / 6	<0.001	1.921	1.575–2.344	<0.001	1.595	1.330–1.911
Percentage of blasts in peripheral blood, (%)
=< 50	890	40 / 12	ref	ref	ref
> 50	155	34 / 13	0.212	1.121	0.937–1.341
Percentage of blasts in bone marrow, (%)
< 20	89	22 / 9	ref	ref	ref	ref	ref	Ref
20 – 50	507	47 / 13	0.099	0.831	0.667–1.036
> 50	408	37 / 15	0.340	0.896	0.716–1.122
Albumin (mg/dL)
>=3.0	830	44 / 14	ref	ref	ref	ref	ref	ref
<3.0	223	27 / 8	<0.001	1.491	1.279–1.739	0.051	1.156	0.999–1.338
Total bilirubin (mg/dL)
=< 1.3	1001	41 / 14	ref	ref	ref
> 1.3	68	28 / 7	0.002	1.476	1.149–1.895
Creatinine (mg/dL)
=< 1.3	942	42 / 14	ref	ref	ref	ref	ref	ref
> 1.3	135	28 / 6	<0.001	1.434	1.193–1.724
Lactate dehydrogenase (U/L)
=< 2ULN	877	44 / 14	ref	ref	ref	ref	ref	ref
> 2ULN	194	22 / 6	<0.001	1.574	1.343–1.845	0.004	1.228	1.066–1.415
Phosphate (mg/dL)
=< 5.0	1026	41 / 13	ref	ref	ref	ref	ref	Ref
> 5.0	46	20 / 8	0.015	1.452	1.076–1.960
Cytogenetic group
Diploid/-Y	438	52 / 19	ref	ref	ref	ref	ref	Ref
Others	300	41 / 15	0.016	1.207	1.036–1.407	<0.001	1.270	1.111–1.453
Complex	296	23 / 3	<0.001	2.083	1.788–2.426	<0.001	2.079	1.815–2.380
Infection group
No infection	983	42 / 13	ref	ref	ref	ref	ref	Ref
FUO	24	31 / 5	0.017	1.676	1.097–2.561	0.004	1.735	1.196–2.515
Others	18	28 / 0	0.020	1.768	1.093–2.860	0.001	1.977	1.320–2.961
Pneumonia	63	32 / 15	0.007	1.455	1.110–1.907	<0.001	1.700	1.340–2.156
Venetoclax-based therapy
No Venetoclax	923	39 / 11	ref	ref	ref	ref	ref	Ref
Venetoclax	165	51 / 29	<0.001	0.603	0.495–0.733	<0.001	0.538	0.449–0.644

^*A cutoff of 0.100 was used for variable selection from univariate to multivariate analysis.

^**Given expected close associations between renal comorbidity and elevated creatinine >1.3 mg/dL and between leukocytosis and absolute neutrophil counts, renal comorbidity and absolute neutrophil counts were excluded from multivariate analysis.

Abbreviations: HR, hazard ratio; CI, confidence interval; ref, reference; MDS, myelodysplastic syndrome; MPN, myeloproliferative neoplasm; ECOG, Eastern Cooperative Oncology Group; ULN, upper limit of normal.

The risk scores based on the HR assigned to each of the 11 consistent independent variables (10 pretreatment; one therapy related) are shown in Table 3. Survival curves by total risk score were well-separated in the training group (Figure 1A). The training group was divided into favorable (total score, −1.0 to 0.0; 52 patients; 5%), intermediate (total score, 0.5–1.0; 230 patients; 21%), poor (total score, 1.5–2.0; 337 patients; 31%), very poor (total score, 2.5 or above; 364 patients; 33%) with estimated 3-year survival rates of 52%, 25%, 11%, and 3%, respectively (p < .001) (Table 4 and Figure 1A). The prediction of survival with the risk classification was validated in the separate validation group (Table 4 and Figure 1B).

Table 3.

Proposed prognostic risk classification for survival: Scoring

Category		Point
Demographic data
Age	60–64	0
	65–74	+0.5
	75+	+1
Therapy-related		+0.5
Antecedent history of dysplasia		+0.5
ECOG PS	0–1	0
	2	+0.5
	3–4	+1
Pulmonary comorbidity		+0.5
Laboratory data
Hemoglobin	≥ 8.0 g/dL	0
	< 8.0 g/dL	+0.5
Platelet count	≥ 50,000 /μL	0
	20,000 – 50,000 /μL	+0.5
	< 20,000 /μL	+1
Lactate dehydrogenase	< 2 upper limit of normal	0
	≥ 2 upper limit of normal	+0.5
Chromosomal abnormalities
Diploid (including -Y)		0
Others / Complex		+0.5
Complex		+1.5
Infection at diagnosis
None		0
Fever of unknown origin / others / pneumonia		+1
Therapy
Venetoclax-based regimen		−1

Abbreviations: ECOG, Eastern Cooperative Oncology Group; PS, performance status.

Figure 1.

Survival by risk group in the training (1A) and validation (1B)

Table 4.

Proposed prognostic risk classification for survival: Risk classification and survival

	Training (n = 1088)		Validation (n = 374)
	Total No.	1-Year/3-year survival, %	Total No.	1-Year/3-year survival, %
Total risk score
−1.0 to 0.0	52	79/52	6	100/100
0.5	63	67/32	26	85/54
1.0	167	54/23	50	66/32
1.5	172	51/12	49	45/11
2.0	165	37/10	65	39/20
2.5	137	32/6	44	42/5
3.0	103	24/1	41	29/0
3.5	68	9/0	28	11/0
4.0–6.0	56	11/0	20	0/0
Risk classification 2000–2009
Favorable ≤0.0	11	91/55	0	NA/NA
Intermediate 0.5–1.0	75	48/17	27	85/40
Poor 1.5–2.0	113	40/10	34	41/18
Very poor ≥2.5	142	23/2	46	15/0
Risk classification 2010–2020
Favorable ≤0.0	41	73/52	6	100/100
Intermediate 0.5–1.0	155	62/29	49	65/39
Poor 1.5–2.0	224	46/12	80	41/15
Very poor ≥2.5	222	22/3	87	30/2
Risk classification 2000–2020
Favorable ≤0.0	52	79/52	6	100/100
Intermediate 0.5–1.0	230	57/25	76	72/39
Poor 1.5–2.0	337	44/11	114	41/16
Very poor ≥2.5	364	22/3	133	25/2

Additional Value of Molecular Mutational Studies in Predicting Survival

A total of 1221 patients had molecular mutational studies. The mutation profile included the assessment of FLT3 and NPM1 mutations, and 28-gene, 53-gene, and 81-gene mutation panels in myeloid neoplasms.⁸¹ After accounting for the variables in the survival risk model, we analyzed the added predictive value of each mutation for survival (Table S4). With the addition of significant mutations to the prognostic model, and on multivariate analysis with backward elimination, the presence of IDH2 (p = .016; HR, 0.717; 95% confidence interval [CI], 0.546–0.940), NPM1 (p = .015; HR, 0.713; 95% CI, 0.542–0.937) and TP53 (p = .047; HR, 1.316; 95% CI, 1.003–1.726) mutations were independently associated with survival (Table 5).

TABLE 5.

Addition of molecular variables to the prognostic model (mutated ≥20 cases)

	No.	Addition to the multivariate			C-indexa, baseline: 0.670	AICca, baseline: 16,276.38
		p	HR	95% CI
Add four molecular results
FLT3 D835	37	.402	0.786	0.448–1.380	0.667	5925.73
IDH2	139	.008	0.693	0.527–0.911
NPM1	198	.045	0.747	0.562–0.993
TP53	180	.080	1.274	0.972–1.671
Add three molecular results, exclusion of FLT3 D835					0.673	6173.06
IDH2	139	.016	0.717	0.546–0.940
NPM1	198	.015	0.713	0.542–0.937
TP53	180	.047	1.316	1.003–1.726

Abbreviations: AICc, conditional Akaike information criterion; CI, confidence interval; C-index, concordance index; HR, hazard ratio.

^aAmong evaluable patients for molecular analysis.

The prognostic model was validated with and without IDH2 mutations (Figure S1A,B), with and without NPM1 mutations (Figure S2A,B), and without TP53 mutations (Figure S3B). The presence of TP53 mutations was associated with adverse outcome regardless of the risk classification (Figure S3A). In patients who received venetoclax therapy, the 3-year survival rates were 58%, 34%, 16%, and 0% in the low, intermediate, poor, and very poor groups, respectively (Figure S4A); without venetoclax, the 3-year survival rates were 50%, 26%, 10%, and 0% in the low, intermediate, poor, and very poor groups, respectively (Figure S4B). The LASSO regression validated all the variables in the Cox model with the minimum λ of 0.003671205 (Figure S5) (Table S5). The estimated β times two were similar to the assigned scores in the Cox model. The 6-year area under the curve was 75.52% and 79.86% in the training and validation groups, respectively. The concordance index (C-index) was 0.642 (SE, 0.009) and 0.662 (SE, 0.015) in the training and validation groups. By the 2017 European LeukemiaNet risk classification by genetics, the C-index was 0.564 (SE, 0.01) using the 952 evaluable patients.

Discussion

Prognostic and predictive models for survival, achievement of CR, and early mortality serve multiple purposes: advising individual patients of their expected outcomes with particular therapies, comparing outcomes across different cancer entities, and comparing outcomes of different forms of therapies, among others.

Several prognostic models have been developed to assess the outcomes among patients with AML receiving intensive chemotherapy.^{45, 46, 82} These were performed across all age groups and also selectively among younger and older patients receiving intensive chemotherapy. These have usually identified similar adverse pretreatment patient and leukemia-associated characteristics, to different degrees, depending on when these studies were conducted (effect of supportive care measures, antibiotics, and AML therapies), whether molecular and cytogenetics studies were fully available, and how large the studies were (large studies identifying more independent adverse variables). The usual adverse factors identified across many studies include older age; poorer performance and organ dysfunctions (cardiac, pulmonary, hepatic, renal) or comorbidities (diabetes, hypertension, infections); and leukemia-associated variables including degree of leukocytosis, percent of marrow or peripheral blasts, elevated uric acid or lactate dehydrogenase, and the presence of adverse cytogenetic or mutational variables. Some of the variables can be both related to patient and leukemia characteristics (e.g., performance status and organ dysfunctions).

Because lower intensity therapy regimens were only recently broadly used in older patients with AML, no studies have evaluated or proposed risk models in older patients with AML receiving lower intensity chemotherapy. Our study is the first large single institutional experience in AML proposing a risk model for survival with lower intensity chemotherapy. As expected, we identified many of the same independent adverse variables associated with survival, although, again as expected, survival was significantly worse in older patients than younger patients, whether they received intensive chemotherapy (previously published) or lower intensity chemotherapy (current study).⁴⁶ We identified 11 variables accounting for survival, 10 pretreatment ones, and one related to therapy (addition of venetoclax). The latter, the benefit of venetoclax containing regimens, has been assuredly confirmed in randomized trials.

The proposed and validated risk model divided patients into four risk groups, a favorable risk group (5% of patients) with an estimated 3-year survival rate of 52%; an intermediate risk group (21% of patients) with an estimated 3-year survival rate of 25%; a poor risk group (31% of patients) with an estimated 3-year survival rate of 11%; and a very poor risk group (33% of patients) with an estimated 3-year survival rate of 3% (Table 4; Figure 1A,B). This risk model can now be used to advise patients on their expectations with such therapy, particularly among those where intensive chemotherapy is associated with a prohibitive high risk of early (4- and 8-week) mortality rates (see previously published risk model for early mortality with intensive chemotherapy).^{45, 46} It can also serve to consider changing therapy in complete remission among patients expected to have poor survival to attempt to modify their course using a different standard (allogeneic stem cell transplant [SCT], oral azacitidine maintenance therapy) or investigational strategies (HMAs with venetoclax or other targeted or immune therapies) The risk model can also be used to stratify patients on randomized trials or to compare different therapeutic strategies in older AML.

Several studies and international classifications (National Comprehensive Cancer Network, European LeukemiaNet, and others) have highlighted the importance of incorporating mutational studies into the risk modeling of AML.^83–87 Some have proposed to define AML risk groups based only on cytogenetic and mutational abnormalities, thus discounting the effect of simpler and easily collected variables such as age, performance status, organ dysfunctions, and comorbidities.⁸⁸ An important question is, after accounting for such variables and newer therapies (addition of FLT3 or IDH inhibitors), what are the mutations that remain predictive and relevant for survival.^89–95 In a previous analysis of intensive chemotherapy, the prediction of mutational studies was reduced to only three mutations that remained important: NPM1 (favorable), TP53, and PTPN11 (both unfavorable).⁴⁶ Similar findings were reported by Tefferi and colleagues⁸² who found only FLT3 mutations and NPM1 mutations to matter. In this analysis of lower intensity therapy in older AML, we identified similar mutations to be important: IDH2 and NPM1 (favorable) and TP53 (unfavorable). In this study, as in our previous one with intensive chemotherapy, FLT3 mutations were not adverse, perhaps because we have used FLT3 inhibitors as part of our frontline AML regimens for over a decade.⁴⁶ In this study, the addition of PTPN11 mutations to the prognostic model showed a trend for unfavorable survival with a HR of 1.260 (95% CI, 0.836–1.900). Given the relatively small number of patients with PTPN11 mutations and overall unfavorable outcome in older patients with AML, the presence of PTPN11 did not reach statistical significance in this study. The independent prognostic significance of TP53 mutations in AML supports the separation of TP53-mutated myeloid neoplasms as a distinct disease entity.^{96, 97}

This analysis also highlights the improvement of patient outcomes with venetoclax. The 3-year survival rate improved from 7% without venetoclax to 20% with venetoclax. The therapeutic improvements included the addition of venetoclax to HMAs, possible addition of other targeted agents, and the use of triple-nucleosides–venetoclax regimens, as well as considering more frequent use of allogeneic SCT in older AML.⁴⁹

There are several limitations in our study. First, we developed the prognostic model from the data since 2000. Although little progress has been made in lower intensity therapy for older patients over decades, recent therapeutic breakthroughs, such as venetoclax, may affect the accuracy of prediction. However, we validated the accuracy of prediction in patients who received venetoclax. Further, we developed the prognostic models using mutually significant variables both with and without eras to develop the robust pronogstic model over decades. In addition, we confirmed the validity of our model in each era. Second, the accuracy of prediction improves with the direct use of regression terms. However, clinicians have limited time and effort for routine use of prediction models in daily practice. Thus, we approximated the regression terms for scoring and simplified the scores for practical use. Third, the assessment of frailty, comorbidity, and infection severity is limited. However, the accuracy of prediction, C-index, improved significantly; our model was validated with stable C-index in the independent validation group with higher C-index than the C-index with the standard-of-care prognostic model with the 2017 European LeukemiaNet by genetics.

In summary, this is the first large analysis of prognostic factors associated with survival with lower intensity therapy in older AML. The validated model can now serve to advise patients, decide on induction therapy or on treatment modifications in CR, and compare the differential effects of current and future strategies in this setting.