Prognostic value of anemia plus thrombocytopenia in primary myelofibrosis

To the Editor:

Median overall survival of people with primary myelofibrosis (PMF) varies significantly, ranging from approximately 14 years in pre-fibrotic PMF to around 7 years in overt PMF [1]. These figures underscore the critical importance of accurate survival prediction to guide timely therapeutic decisions and to provide patients with realistic expectations regarding their long-term prognosis.

Although several prognostic models incorporating clinical, cytogenetic, and molecular data—such as IPSS, DIPSS, DIPSS-plus, MIPSS70, and GIPSS—have been developed [2], their full predictive potential is often difficult to realize in routine clinical practice. Notably, one-third of patients are never assigned a risk category, and in nearly 40% of cases, the assigned category is incorrect [3]. This highlights the need for more accessible and user-friendly prognostic tools.

A recent analysis of 559 patients from the European multi-center ERNEST-2 registry reported that 34 percent of subjects with overt PMF presented with anemia only, 6 percent with isolated thrombocytopenia, and 10 percent with both [4]. Subjects with anemia and thrombocytopenia had a median survival of less than two years. These findings call into question the weight of anemia as an independent risk factor in the widely used IPSS model, which currently also includes age over 65 years, blood blasts ≥1%, white blood cell count >25 × 10⁹/L, and constitutional symptoms [5, 6].

In this study, we sought to validate the findings of the ERNEST-2 registry through a retrospective cohort analysis. Our validation cohort included 530 individuals aged ≥18 years with overt PMF, identified from the institutional Pavia-CSM database at the Center for the Study of Myelofibrosis, IRCCS Policlinico S. Matteo Foundation, Pavia. This database comprises consecutive subjects diagnosed with a myeloproliferative neoplasm since 1998, each evaluated at least once. All participants provided written informed consent, as approved by the institutional ethics committee. Diagnoses were established according to WHO consensus criteria.

Phenotypic classification followed the definitions used in the ERNEST-2 registry. Table 1 displays a comparative analysis between the ERNEST-2 registry and the Pavia cohort. The proportion of subjects with myelodepletive phenotype was comparable between cohorts (49% vs. 51%; p = 0.53). However, notable differences emerged in demographic characteristics. Myelodepletive subjects in the Pavia cohort were younger (mean age: 61 vs. 67 years; p < 0.001) and more often men (76% vs. 69%; p = 0.044). Isolated anemia was more prevalent in the Pavia cohort (69% vs. 49%; p < 0.001), whilst the co-occurrence of anemia and thrombocytopenia was less common (11% vs. 20%; p = 0.004). Additionally, leukopenia was significantly more common in the Pavia cohort (15% vs. 2%; p < 0.001).

Table 1.

Main features of patients with primary myelofibrosis categorized as myeloproliferative and myelodepletive phenotype at diagnosis.

	Myeloproliferative phenotype			Myelodepletive phenotype			Comparison between myeloproliferative and myelodepletive phenotype
			ERNEST-2 registry vs. Pavia-CSM database			ERNEST-2 registry vs. Pavia-CSM- database	ERNEST-2 registry	Pavia-CSM-database
	ERNEST-2 registry N=284	Pavia-CSM- database N=259	p	ERNEST-2 registry N=275	Pavia-CSM- database N= 271	p	p	p
Age, years, mean (SD)	60.6 (14.0)	54.2 (13.6)	<0.001	67.2 (11.4)	61.3 (11.7)	<0.001	<0.001	<0.001
Male, n (%)	173 (60.9)	174 (67.2)	0.127	189 (68.7)	207 (76.4)	0.044	0.054	0.019
IPSS, n (%)
- low/Int-1	198/280 (70.7)	73/85 (84.7)	<0.001	79/271 (29.2)	29/92 (31.5)	0.56	<0.001	<0.001
- Int2/high	82/280 (29.3)	13/85 (15.3)	<0.001	192/271 (70.8)	63/92 (68.5)	0.56	<0.001	<0.001
Spleen, n (%)
No palpable spleen	48/233 (20.6)	46/257 (17.9)	0.44	43/223 (19.3)	44/268 (16.4)	0.39	0.88	0.29
Palpable spleen, < 10 cm	153/233 (65.7)	159/257 (61.9)	0.52	146 /223 (65.5)	171/268 (63.8)	0.69	0.96	0.65
Palpable spleen, ≥ 10 cm	32/233 (13.7)	52/257 (20.2)	0.056	34/223 (15.2)	53/268 (19.8)	0.18	0.65	0.91
Symptomatic disease at diagnosis
No, n (%)	152/270 (56.3)	162/209 (77.5)	<0.001	128/255 (50.2)	92/203 (45.3)	0.29	0.16	<0.001
Yes, n (%)	118/270 (43.7)	47/209 (22.5)	<0.001	127/255 (49.8)	111/203 (54.7)	0.29	0.16	<0.001
Hematology
Hemoglobin (g/dL), median (IQR)	12.5 (11.7-14.1)	12.6 (11.6-14.3)	*	9.4 (8.2-10.3	9.1 (8-10.4)	*	<0.001	*
WBC x109/L, median (IQR)	10.5 (7.4-16.3)	10 (7.6-13.8)	*	7.3 (4.1-10.8)	5.9 (3.8-10)	*	<0.001	*
WBC >20 x 109/L, n (%)	50/284 (17.6)	24/259 (9.3)	0.005	24/272 (8.8)	18/267 (6.7)	0.36	0.002	0.27
Platelet x109/L, median (IQR)	386 (251-657)	382 (245-636)	*	168 (78-296)	157 (83-300)	*	<0.001	*
Platelet > 450 x109/L, n (%)	126/283 (44.5)	110/259 (42.5)	0.63	44/271 (16.2)	43/269 (16)	0.94	<0.001	<0.001
Driver mutations, n (%)
JAK2V617F	161/224 (71.9)	139/215 (64.5)	0.003	118/202 (58.4)	132/213 (61)	0.59	0.004	0.45
Homozygous	32/93 (34.4)	79/138 (57.2)	<0.001	14/57 (24.6)	44/131 (33.6)	0.22	0.27	<0.001
Heterozygous	61/93 (65.6)	59/138 (42.8)	<0.001	43/57 (75,4)	87/131 (66.4)	0.22	0.27	<0.001
MPL W515	10/215 (4.7)	15/215 (7.0)	0.31	11/186 (5.9)	16/213 (7.5)	0.52	0.66	0.84
CALR	27/197 (13.7)	56/215 (26.0)	0.002	14/153 (9.2)	30/213 (14.1)	0.16	0.24	0.002
Type 1	23/197 (11,7)	41/215 (19.1)	0.038	9/153 (5.9)	22/213 (10.3)	0.14	0.091	0.002
Type 2	4/196 (2.0)	12/215 (5.6)	0.06	5/153 (3.3)	7/213 (3.3)	1	0.51	0.25
Triple negative	8/197 (4.1)	5/215 (2.3)	0.29	12/153 (7.8)	35/213 (16.4)	0.015	0.16	<0.001

^*The comparison cannot be done with the median values without the individual values.

Genotypic distributions also differed between cohorts. In the ERNEST-2 cohort, the JAK2^V617F mutation was more frequent in subjects with anemia only compared with subjects with thrombocytopenia only (65% vs. 32%; p = 0.002), whilst a triple-negative genotype was more common in the thrombocytopenia-only group (35% vs. 3%; p < 0.001). Conversely, in the Pavia cohort, JAK2^V617F mutations were more common in the thrombocytopenia-only subgroup compared with the anemia-only subgroup (82% vs. 61%; p = 0.009). A triple-negative genotype was more common in subjects with both thrombocytopenia and anemia compared with thrombocytopenia only (18% vs. none; p = 0.05).

Median follow-up durations were 5.4 years (Interquartile Range [IQR]: 2.5–9.1) for the ERNEST-2 cohort and 4.1 years (IQR: 1.75–8.75) for the Pavia cohort. In ERNEST-2, 392 patients (70%) died, with mortality rates of 60% in the myeloproliferative group and 81% in the myelodepletive group (p < 0.001). Corresponding death incidence rates were 7.8 per 100 subject-years (95% CI: 6.6–9.1) for myeloproliferative cases and 17.3 per 100 subject-years (95% CI: 15.5–19.8) for myelodepletive cases (p < 0.001). In contrast, the Pavia cohort recorded 219 deaths (42%), with similar frequencies between myelodepletive (41%) and myeloproliferative (43%) groups (p = 0.94).

In the ERNEST-2 cohort, patients with concurrent anemia and thrombocytopenia demonstrated the shortest median survival (1.7 years; IQR, 0.7–4.4). This finding was not mirrored in the Pavia cohort, where the lowest median survival was observed among individuals with severe anemia alone (5.4 years; IQR, 2.2–11.7). Consistently, the ERNEST-2 cohort showed the lowest 5- and 10-year survival rates in subjects with both anemia and thrombocytopenia, whereas in the Pavia cohort, the poorest long-term outcomes were associated with severe anemia alone, followed by anemia with thrombocytopenia (Table 2).

Table 2.

Comparison of cumulative incidence of death at 5 and 10 years between ERNEST-2 registry and Pavia-CSM database in the category of severe anemia only and anemia plus thrombocytopenia.

	Time (years)	Severe anemia only		Anemia plus thrombocytopenia
		ERNEST-2 registry	Pavia-CSM database	ERNEST-2 registry	Pavia-CSM database
Cumulative incidence of death (95% CI)	5	57 (45.1-69.5)	44.8 (31.7-60.4)	77.9 (65.8-88)	45.3 (32.9-68.6)
	10	86 (75.3-93.6)	71.2 (54.6-85.9)	85.9 (74.9 93.8)	76.8 (51.9-94.8)
Blast transformation (95% CI)	5	14.3 (4.5-24.1)	19.6 (10.4-35.2)	22.3 (10.9-33.6)	22.2 (12.1-38.7)
	10	17.8 (6.1-29.6)	59.4 (37.2-82.5)	27.4 (17.8-37.0)	37.8 (16.2-72.1)

Competing risk analyses revealed a higher cumulative incidence of BP transformation in myelodepletive versus myeloproliferative phenotypes across both cohorts: ERNEST-2, 17.0% (95% CI: 12.6–21.7) vs. 10.6% (95% CI: 7.7–15.3; p = 0.028); Pavia, 32.8% (95% CI: 24.8–39.6) vs. 28.2% (95% CI: 20.8–34.1; p = 0.032). Notably, in ERNEST-2, the highest BP transformation rates were linked to anemia plus thrombocytopenia, whereas in Pavia, severe anemia alone conferred the greatest risk (Table 2).

To evaluate the predictive performance of anemia plus thrombocytopenia across cohorts, receiver operating characteristic (ROC) curve analysis was performed. The area under the curve (AUC) for 10-year survival was 0.52 in ERNEST-2 and 0.50 in the Pavia cohort, indicating poor discriminative capacity for this phenotype in both settings.

Risk of bias and applicability of the anemia plus thrombocytopenia phenotype were assessed using the Quality in Prognosis Studies (QUIPS) tool. Five domains were evaluated: study participation, attrition, prognostic factor measurement, outcome measurement, and confounding. Studies were rated “low risk” only if all domains met this criterion; otherwise, they were classified as “high risk.” ERNEST-2 was rated high risk due to limited transparency in patient selection, exclusion of leucopenia as a predictor based on data-driven decisions, and omission of key confounders such as transplantation status. Similarly, the Pavia-CSM database was deemed high risk due to an insufficient number of deaths relative to cohort size consistent with incomplete follow-up, and failure to account for transplantation status.

In conclusion, both ERNEST-2 and Pavia cohorts present important concerns regarding applicability. The anemia plus thrombocytopenia phenotype offers limited theoretical and statistical validity as a survival prognostic marker in overt-PMF. Our study highlights the challenge of predicting outcomes in PMF and reinforces the need for large, rigorously curated datasets as emphasized by QUIPS standards. The lack of prognostic discrimination for anemia and thrombocytopenia likely reflects failure to integrate molecular risk co-variates which have fundamentally reshaped PMF risk stratification. These findings highlight the need for hybrid prognostic models that incorporate clinical and molecular features.

Author contributions

GB had the idea for the study, performed the first analysis of the PV-CSM database, wrote the first draft of the manuscript. ADeS validated the statistical analysis of manuscript, did the comparisons within the two databases, and did the bias analysis according to the QUIPS score. RPG and VR performed critical revision of the manuscript for important intellectual content.

Data availability

Anonymized data used in this study are available on reasonable request from the Corresponding Author and consistent with the laws of Italy.

Competing interests

RPG is a consultant to BeiGene Ltd., Fusion Pharma LLC, LaJolla NanoMedical Inc., Mingsight Parmaceuticals Inc., Kite Pharma and CStone Pharmaceuticals; Advisor to Antegene Biotech LLC, Medical Director and FFF Enterprises Inc.; Partner in AZACA Inc.; Board of Directors, RakFond Foundation for Cancer Research Support; Scientific Advisory Board, StemRad Ltd.

Ethics approval

The research was conducted in accordance with the World Medical Association Declaration of Helsinki. All subjects included in the Pavia-CSM database gave written informed consent approved by the IRCCS Policlinico S. Matteo Foundation Institutional Ethics Committee. The Ethics Committee of the Hospital also approved a written informed consent for patients to donate samples for molecular research (reference number 20110004143 of the 26.9.2011).