Assessing drivers of disparate outcomes and applicability of risk stratification in a cohort of patients with myelofibrosis

Dear Editor,

Patients with Philadelphia-chromosome negative fibrotic phase myeloproliferative neoplasms (MPNs), whether it be primary myelofibrosis (PMF) or secondary myelofibrosis (SMF), have an inferior overall survival compared to cellular-phase MPNs due to progressive bone marrow failure and increased risk of leukemic transformation [1, 2]. Current approved therapies for myelofibrosis (MF) include Janus Kinase (JAK) inhibitors, which primarily provide symptomatic benefit and reduction in spleen size; however, only allogeneic hematopoietic stem cell transplant (allo-HCT) has curative potential [1]. Risk stratification to identify those that would benefit from allo-HCT is a vital component of decision-making [3]. Beyond traditional risk stratification tools, there may be other contributors to disparate survival outcomes that need to be better assessed in MF.

Our group has previously characterized the impact of structural racism in acute myeloid leukemia and identified it as the largest mediator of survival disparity in Black and Hispanic patients relative to White patients [4, 5]. Analyses focused on understanding the potential drivers of disparity in MPNs are more limited [6–9]. With this in mind, we aimed to investigate potential drivers of disparate outcomes in patients with MF, including patient characteristics, disease features, treatment modalities, and census-tract level measures. Furthermore, we aimed to characterize the applicability of current MF risk stratification tools in a heterogeneous population of patients.

Methods

We conducted a retrospective study of patients with MF treated at six Chicagoland institutions. Patient demographic, disease, and census-tract data were collected after receiving Institutional Review Board/data use approval at each participating institution. Adult patients with a diagnosis of MF from 2011 onwards were eligible.

We collected demographic data, including age at diagnosis, gender, race/ethnicity, body mass index (BMI), and insurance status. Clinical and hematopathologic characteristics, including blood counts, blast percentage, presence of splenomegaly, disease phenotype, cytogenetic data, and molecular mutation data, were collected for each patient when available. Treatment modalities (JAK inhibitor use, receipt of allo-HCT), disease complications (arterial thrombosis, venous thromboembolism, progression to accelerated/blast-phase (AP/BP), and survival outcomes were also collected. Using the Federal Financial Institutions Examination Council’s Geocoding/Mapping system, we captured each patient’s census-tract based on the address on file at each institution. Census-tract level social determinant data was captured from the American Community Survey from the year of each patient’s diagnosis.

Data were stratified by race/ethnicity (non-Hispanic White [NHW], non-Hispanic Black [NHB], non-Hispanic Other [NHO], and Hispanic). In patients with the appropriate clinical data available, risk stratification by Dynamic International Prognostic Scoring System (DIPSS+) and Mutation and Karyotype Enhanced International Prognostic Scoring System (MIPSS70+) Version 2.0 (v2.0) were calculated [10, 11]. These risk scores were applied to patients with SMF in addition to PMF given data to support their utility in both disease phenotypes [12, 13]. Chi-squared and Fisher’s exact tests were used to compare demographic, disease characteristic, and treatment data by race/ethnicity.

Survival analysis was performed using Kaplan–Meier analysis. Cox proportional hazards models were used to assess survival by race/ethnicity after controlling for age and gender—producing a baseline hazard ratio (HR_B). These models were also used to assess the success of prognostic models in predicting survival in our dataset by race/ethnicity. For race/ethnic cohorts with fewer patients, prognostic evaluation was also performed for simplified high vs low risk categories.

Following Abraham et al. 2022 [4], structural racism was measured using four indicators: tract disadvantage, tract affluence, proportion of NHB individuals in a tract, and proportion of NHW individuals in a tract. Full methodology is detailed in the supplemental information.

Cox proportional hazard models were used to produce hazard ratios by race/ethnicity, controlling for age and gender at baseline (HR_B) and after incorporating insurance status, structural racism, or DIPSS+ (HR_A, or adjusted Hazard Ratio).Mediation analysis was performed using the following equation: [(HR_B- 1)-(HR_A-1)]/(HR_B- 1).

Results

Patient characteristics

Our total cohort contained 548 patients with MF. Patient characteristics are detailed in Supplemental Table 1. Median age by race/ethnicity was 66 years-old (yo) for NHW, 63 yo for NHB, 68 yo for Hispanic, and 64 yo for NHO patients (p = 0.26). Rates of PMF were highest for NHB (79% PMF, 21% SMF) and Hispanic (88% PMF, 12% SMF) patients and lower for NHW (63% PMF, 37% SMF) and NHO (58% PMF, 42% SMF) patients (p < 0.01). Testing for canonical driver mutation (defined as JAK2, CALR, or MPL) was available in 484 patients; 336 (69%) had a JAK2 mutation, 84 (17%) had a CALR mutation, and 22 patients (5%) had a MPL mutation with no significant difference in presence of a driver mutation by race/ethnicity (p = 0.97). Hispanic patients were marginally less likely to be privately insured (26% vs 35%, 37%, and 36% for NHW, NHB, and NHO, respectively, p = 0.14) (Supplemental Table 2).

DIPSS+ scores were evaluable in 471 patients, with NHB patients being marginally more likely to have low risk disease (23% vs 14%, 4%, and 10% for NHW, NHO, and Hispanic, respectively, p = 0.16, Supplemental Table 3A). DIPSS+ was successful at discriminating patients at high vs low risk of mortality in NHW, NHB, and NHO patients (concordance [C] = 0.66, 0.74, and 0.83, respectively). For Hispanic patients, however, DIPSS+ had a concordance of 0.5 (Table 1A). MIPSS70 + v2.0 was evaluable in 463 patients, with no significant difference in prognostic score distribution by race/ethnicity (p = 0.89, Supplemental Table 3B). Similarly, MIPSS70+ v. 2.0 underperformed in stratifying survival for Hispanic and NHW patients relative to NHB, and NHO (C = 0.58 and 0.61 vs. 0.67 and 0.79, respectively) (Table 1B).

Table 1.

Overall survival by race/ethnicity when stratifying by prognostic score.

A: Overall Survival by race/ethnicity by DIPSS+
	Low risk	Intermediate 1	Intermediate 2	High risk	KM analysis	Concordance
NHW (n = 335)	13.75	NR	5.33	3.26	p = <0.01	0.661
NHB (n = 77)	NR	NR	2.93	1.08	p = <0.01	0.735
Hispanic (n = 29)	4.1	NR	6.76	NR	p = 0.73	0.5
NHO (n = 26)	NR	NR	NR	0.97	p = <0.01	0.827
Hispanic (n = 29)	NR		6.74		p = 0.62	0.523
NHO (n = 26)	NR		2.6		p = 0.03	0.686

B: Median Overall Survival (years) by Race by MIPSS + 70 v 2.0
	Very low risk	Low risk	Intermediate risk	High risk	Very high risk	KM analysis	Concordance
NHW (n = 330)	NR	13.75	NR	5.51	5.10	P < 0.01	0.612
NHB (n = 75)	7.12	NR	NR	3.03	1.51	p = 0.02	0.667
Hispanic (n = 29)	NA	NR	NR	6.74	2.83	p = 0.75	0.578
NHO (n = 25)	NR	NR	3.93	3.73	1.38	p = 0.02	0.793
Hispanic (n = 29)	NR		6.74			p = 0.53	0.519
NHO (n = 25)	NR		2.6			p < 0.01	Unable to calculate

NR Median survival not reached, DIPSS+ Dynamic International Prognostic Scoring System, KM Kaplan Meier, NA no patients in this group; MIPSS70+ Version 2.0 Mutation and Karyotype Enhanced International Prognostic Scoring System.

Complication and treatment disparities

Prevalence of venous thromboembolism (VTE) was highest among NHW and NHB patients (27% and 20%, respectively) and lowest for NHO and Hispanic patients (15% and 6%, respectively) (p = 0.049). Arterial thrombus was present in 16% of patients and did not differ by race/ethnicity (p = 0.77, Supplemental Table 4). Fourteen percent of patients had progression to AP/BP.

We also examined JAK inhibitor use by race/ethnicity and by insurance status. JAK inhibitors were prescribed at higher rates in NHW, NHO, and Hispanic patients relative to NHB patients (57%, 55%, and 53% vs 40% respectively, p = 0.03). Notably, after excluding low risk disease by DIPSS + , JAK inhibitor prescription did not differ significantly by race/ethnicity or insurance status (p = 0.27 and p = 0.89, Supplemental Tables 5 and 6). Rates of allo-HCT were numerically lower in NHB patients than NHW, NHO, and Hispanic patients (14% vs 22%, 21%, and 21% respectively, p = 0.46, Supplemental Table 7). In age and gender-adjusted progression-free survival models, the disparity in hazard ratios was smaller relative to overall survival models (Supplemental Table 8).

Survival analysis and mediation analysis

Six patients had incomplete follow-up data and were excluded from survival analysis, resulting in a total sample of 542 patients with an overall median survival of 7.86 years. Unadjusted median survival was 7.84 and 9.48 years for NHW and Hispanic patients, respectively, but was not reached for NHB or NHO patients (Supplemental Fig. 1). After controlling for age and gender, compared with NHW patients, the hazard of death was 42% greater for NHB patients (HR = 1.42; confidence interval [CI]: 0.97, 2.09), 46% greater in NHO patients (HR = 1.46, CI: 0.82, 2.58), and roughly equal for Hispanic patients (HR = 0.91; CI: 0.50, 1.64). After further adjusting for tract-level structural racism, the survival disparity between NHW and NHB was fully accounted for with the HR for NHB decreasing to 0.98 (CI: 0.55, 1.74). Tract-level structural racism accounted for nearly 20% of the disparity between NHW and NHO patients, with the HR for NHO decreasing to 1.37. Risk stratification at diagnosis did not mediate the disparity between NHW and NHB or Hispanic patients, but mediated 30% of the disparity for NHO patients, with HR decreasing to 1.32. Finally, insurance status, prescription of JAK inhibitors, and receipt of allo-HCT did not account for any disparity between NHW and NHB or NHO (Table 2).

Table 2.

Mediation analysis of Cox proportional hazards.

	NHB	Hispanic	NHO
Baseline HR	1.42	0.91	1.46
95% CI	0.97, 2.09	0.50, 1.64	0.82, 2.58
SR-adjusted HR	0.98	0.73	1.37
95% CI	0.55, 1.75	0.39, 1.39	0.77
Mediation proportion	104%	−190%	19%
DIPSS+-adjusted	1.79	0.76	1.32
95% CI	1.20, 2.65	0.39, 1.50	0.69, 2.52
Mediation proportion	−86%	−158%	30%
Insurance-adjusted	1.43	0.86	1.46
95% CI	0.98, 2.10	0.47, 1.58	0.82, 2.59
Mediation proportion	−2%	−46%	−1%
JAK inhibitor- adjusted	1.42	0.91	1.46
95%CI	0.97, 2.09	0.50, 1.64	0.82,2.58
Mediation proportion	0%	0%	0%
Allo-HCT-adjusted	1.53	0.89	1.50
95% CI	1.04, 2.25	0.49, 1.61	0.84, 2.65
Mediation proportion	−25%	−17%	−8%

HR Cox-Hazard model hazard ratio, SR tract structural racism, allo-HCT allogeneic hematopoietic stem cell transplant, NHB Non-Hispanic Black, NHO Non-Hispanic Other.

Discussion

To our knowledge, this is the largest multicenter study of MF that incorporates individual patient characteristics with residential data. We observed an increased risk of mortality in NHB patients relative to NHW patients after controlling for age and gender. Further, we found that this disparity in survival was entirely mediated by census-tract level measures of structural racism. Our study specifically characterizes the impact of census-tract measures that mediate survival disparity in MF even when accounting for traditional risk stratification tools.

The studies that initially validated MF risk stratification models (DIPSS+ and MIPSS70+ v. 2.0) did not report racial/ethnic makeup of their cohorts [10, 11]. Gangat et al. conducted an analysis comparing a multi-center cohort of Black patients with PMF to a Mayo-specific cohort of non-Black patients and reported higher rates of high-risk disease by DIPSS and DIPSS+ stratification while still having similar overall survival across cohorts [8]. We found that both DIPSS+ and MIPSS70+ v. 2.0 models underperformed in Hispanic patients relative to NHW and NHB patients. This suggests that current prognostic tools may be insufficient to predict outcomes accurately in this group. Due to the racial heterogeneity of Hispanic populations, further research is necessary to better characterize this performance disparity.

Limitations of our study include the inability to quantify structural racism across the patient’s life course and lack of longitudinal data necessary to re-calculate dynamic scores such as DIPSS+. Further, heterogenous populations were aggregated into NHO due to sample size. Lastly, patients with PMF and SMF were all analyzed as a single cohort and risk scores were only characterized at diagnosis.

In conclusion, our study demonstrates an enduring disparity in survival for NHB patients relative to NHW patients, which is largely driven by structural racism. Future work is needed to elucidate the precise mechanisms by which structural racism produces these disparities.

Author contributions

AP and AAP designed the study plan, collected data, performed data analysis, and wrote the manuscript; GR performed data analysis and reviewed/revised the manuscript; AS, MB, AD, CA, AD, AN, SZ, AO, NA, AE, VS, IA, SBT, MZ, MLL, DR, JK.A., WS, IK, and OO collected data and reviewed/revised the manuscript.

Data availability

The datasets generated during and/or analyzed during the current study are not publicly available due to the privacy of protected health information but are available from the corresponding author on reasonable request and in accordance with institutional guidelines regarding data sharing.

Competing interests

Tsai: Jazz, Sanofi: Honoraria; Bristol Myers Squibb: Membership on an entity’s Board of Directors or advisory committees. Rondelli: Vertex Pharmaceuticals: Honoraria. Altman: AbbVie, Aptitude Health, Astellas Pharma, BioSight, Bluebird Bio, Curio, Daiichi Sankyo, Dark Blue Therapeutics, Gilead, Kura Oncology, Kymera, Stemline Therapeutics, Treadwell Therapeutics, Rigel, Syros: Honoraria; AbbVie, Agios, ALX Oncology, Amgen, Amphivena, Aprea AB, Aptose Biosciences, Astellas Pharma, BioSight, Boehringer Ingelheim, Bristol Myers Squibb, Celgene, Cyclacel, Fujifilm, Immunogen, Kartos Therapeutics, Kura Oncology, Loxo, Pfizer: Research Funding. Stock: Kura: Research Funding; Adaptive: Consultancy, Honoraria; Kura, Servier, Newave, Adaptive, Jazz, Asofarma: Consultancy, Honoraria, Membership on an entity’s Board of Directors or advisory committees. Odenike: AbbVie (Inst); Agios (Inst); Aprea AB (Inst); Astex Pharmaceuticals (Inst); AstraZeneca (Inst); Bristol-Myers Squibb (Inst); Celgene (Inst); CTI BioPharma Corp (Inst); Daiichi Sankyo (Inst); Incyte (Inst); Janssen Oncology (Inst); Kartos Therapeutics (Ins: Research Funding; AbbVie; Blueprint Medicines; BMS; Bristol-Myers Squibb/Celgene (Inst); Celgene; CTI; Impact Biomedicines; Kymera; Novartis; SERVIER; Taiho Pharmaceutical; Treadwell Therapeutics: Honoraria. Patel: honoraria from AbbVie, Astellas, Amgen, Jazz, Sobi, Syndax; research funding (institutional) from Pfizer, Incyte, Servier, Sumitomo.

Ethics approval and consent to participate

The study was approved by the institutional review board (IRB) and ethical committee clearance was obtained (IRB23-0006). All clinical investigations were conducted according to the principles of the Declaration of Helsinki. Regarding informed consent, a waiver of consent was granted by the IRB as part of protocol approval due to this being a retrospective study that would be infeasible to carry out without a waiver of consent.

Supplementary information

The online version contains supplementary material available at 10.1038/s41408-025-01366-x.