Sézary Syndrome: Survival Trends, Racial Disparities, and Limited Prognostic Value of Routine Registry Variables

ABSTRACT

Introduction

Sézary syndrome (SS) is a rare leukemic cutaneous T‐cell lymphoma with limited contemporary population‐level outcome data.

Methods

Using SEER‐22, we identified 403 adults with first primary SS diagnosed from 2000–2021 and evaluated survival, disparities, and prognostic modeling.

Results

Median overall survival (OS) was 48 months; 5‐year OS was 42.7%. Survival improved from 39.5 months in 2000–2010 to 56.0 months in 2011–2021. Black patients were diagnosed younger and had worse adjusted mortality than White patients. Machine‐learning and conventional models showed limited discrimination.

Conclusion

SS remains highly lethal, with persistent racial disparities and limited prognostic value of routine registry variables.

Trial Registration

The authors have confirmed clinical trial registration is not needed for this submission

1.

Sézary syndrome (SS) is a rare leukemic variant of cutaneous T‐cell lymphoma with historically poor survival, yet contemporary population‐based data remain limited [1, 2, 3, 4, 5, 6]. Although small institutional series suggest a median overall survival (OS) of only a few years, it remains unclear whether outcomes have improved in the modern therapeutic era, whether racial and socioeconomic disparities parallel those observed in mycosis fungoides (MF), and whether routine registry variables can support meaningful prognostic modeling [1, 5, 7, 8, 9]. Using SEER‐22, we evaluated real‐world survival trends and racial disparities in SS over the last two decades and assessed whether routine registry variables could identify prognostic factors and support clinically meaningful survival prediction using conventional and machine‐learning models.

We identified 403 adults with a first primary SS (ICD‐O‐3 9701/3) diagnosed between 2000 and 2021. The median age at diagnosis was 68 years (range, 24–90), and most patients were aged ≥ 60 years. White patients comprised 77% (n = 311) of the cohort and Black patients 19% (n = 77). Because the number of patients in other racial categories was small (n = 15), Asian/Pacific Islander, American Indian/Alaska Native, and unknown race were combined into a single “Other/Unknown” group for analysis. Age at diagnosis differed markedly by race: 46.8% of Black patients were aged < 60 years compared with 25.4% of White patients, whereas 53.2% versus 74.6%, respectively, were aged ≥ 60 years (p = 0.0009). In contrast, sex distribution did not differ by race; 55.3% of White patients and 53.2% of Black patients were male (p = 0.89). Median follow‐up by reverse Kaplan–Meier was 86.0 months (95% CI, 73.0–97.0).

Overall outcomes remained poor. For the entire cohort, median OS was 48 months, and estimated 1‐, 3‐, and 5‐year OS rates were 83.2%, 60.9%, and 42.7%, respectively. We next examined temporal trends by comparing patients diagnosed in 2000–2010 versus 2011–2021, approximating the periods before and after broader availability of modern systemic and targeted therapies [6, 10]. Median OS improved from 39.5 months (95% CI, 31.0–50.0) in 2000–2010 to 56.0 months (47.0–80.0) in 2011–2021. Short‐term survival also improved: 12‐month OS increased from 75.7% to 87.5% (difference, +11.8 percentage points; Holm‐adjusted p = 0.015), and 60‐month OS from 35.0% to 48.4% (+13.4 percentage points; p = 0.032). The improvement at 36 months was of similar magnitude (+10.4 percentage points) but did not reach conventional statistical significance after multiple‐comparison adjustment (p = 0.053). Thus, survival for SS has improved modestly over time but remains substantially inferior to that reported in contemporary MF cohorts [5, 11].

Racial patterns in OS were also notable. When all eras were combined, median OS was 37.0 months (95% CI, 27.0–63.0) in Black patients, 49.0 months (44.0–61.0) in White patients, and 36.0 months (31.0–NA) in the Other/Unknown group. In addition, when all eras were combined, 12‐month OS was similar in Black and White patients, 83.9% (95% CI, 76.0–92.7) and 82.9% (78.8–87.3), respectively; but the curves separated over time. At 36 months, OS was 51.5% (40.8–65.1) in Black patients versus 63.9% (58.4–69.8) in White patients; at 60 months, corresponding estimates were 36.3% (26.1–50.5) and 44.0% (38.1–50.9). Stratified by era, median OS in Black patients increased from 29.0 months (95% CI, 23.0–50.0) before 2011 to 47.0 months (39.0–NA) after 2011. In White patients, median OS increased from 44.0 months (33.0–55.0) before 2011 to 47.0 months (47.0–116.0) after 2011. One‐year OS improved from 75.0% to 88.2% in Black patients and from 77.0% to 88.6% in White patients, while 5‐year OS improved from 25.0% to 42.6% and from 37.2% to 49.8%, respectively. However, these temporal gains reached statistical significance among White patients but not among Black patients (Figure 1). A four‐era analysis (≤ 2005, 2006–2010, 2011–2017, and ≥ 2018), defined to approximate major therapeutic milestones, showed a similar pattern (Figure S1).

FIGURE 1.

Age distribution and overall survival of patients with Sézary syndrome by race and calendar period. (A) Age distribution at diagnosis by race. Because the numbers of patients in racial categories other than White and Black were small, Asian/Pacific Islander, American Indian/Alaska Native, and unknown race were combined into a single Other/Unknown group for analysis. (B) Kaplan–Meier overall survival curves by race for the full cohort, with shaded 95% confidence intervals and numbers at risk shown below. (C) Kaplan–Meier overall survival curves comparing patients diagnosed in ≤ 2010 versus 2011+, shown for the full cohort (All races) and separately for White and Black patients. Dashed guide lines indicate median survival. P values were calculated using the log‐rank test.

To quantify these differences, we fit sequential Cox models. In crude analyses, Black race was associated with a nonsignificant 19% higher hazard of death compared with White race (HR 1.19, 95% CI 0.86–1.63). After adjustment for age, sex, and calendar period, the hazard ratio increased to 1.67 (1.20–2.33), indicating that part of the disparity was masked by the younger age distribution among Black patients [7, 8]. Additional adjustment for socioeconomic indicators, including county‐level income and rural–urban continuum, modestly attenuated the association (HR 1.57, 95% CI 1.10–2.23), and further adjustment for SEER Summary Stage yielded an HR of 1.56 (1.09–2.22). In a fully adjusted landmark model incorporating time‐to‐treatment interval, Black race remained independently associated with higher mortality (HR 1.49, 95% CI 1.03–2.15). By contrast, the small Other/Unknown race group showed wide confidence intervals overlapping unity in all models. Time‐windowed models similarly demonstrated increased hazards for Black patients over 0–36 and 0–60 months from diagnosis, but not within the first year, consistent with the delayed divergence of the survival curves.

Socioeconomic and geographic factors also showed graded associations with outcome. Patients residing in lower‐income counties and non‐metropolitan areas experienced worse OS than those in the highest‐income and metropolitan strata (data not shown). However, these factors did not fully account for the excess risk observed in Black patients; the association persisted after simultaneous adjustment for income, urban–rural status, stage, and treatment timing. Together, these findings suggest that both access‐related and disease‐ or care‐related factors may contribute to the observed disparities [7, 8, 9, 12, 13].

We next evaluated the potential for registry‐based risk stratification in SS by fitting four survival models in training and held‐out test sets: a standard multivariable Cox model, elastic‐net penalized Cox, random survival forest, and gradient‐boosted Cox. Model development incorporated the full set of clinically relevant variables available in SEER, including age, race, sex, calendar period, county‐level household income, rural–urban residence, SEER Summary Stage, and time‐to‐treatment interval. SEER Summary Stage is a registry‐derived extent‐of‐disease variable that broadly classifies tumors as localized, regional, distant, or unknown based on available information at diagnosis. For SS, this variable does not correspond to the disease‐specific International Society for Cutaneous Lymphomas/European Organization of Research and Treatment of Cancer (ISCL/EORTC) TNMB staging system and does not capture quantitative blood involvement, skin tumor burden, nodal pathology, or visceral disease in sufficient detail. Among these registry‐derived variables, only a limited subset showed meaningful independent prognostic association, with older age and Black race emerging as the most consistent adverse predictors of OS. By contrast, other variables such as sex, stage, and county‐level socioeconomic categories were not clearly independently associated with outcome, while estimates for smaller racial subgroups were imprecise because of limited sample size. Consistent with this sparse covariate‐level signal, all four modeling approaches showed similarly limited discrimination in the test set, with Harrell's C‐indices near 0.5 and low time‐dependent Uno AUCs at 12, 36, and 60 months. Calibration curves and Brier scores showed only minimal improvement over unstratified prediction. In contrast, application of the same modeling pipeline to a much larger MF cohort from SEER yielded substantially stronger performance, with C‐indices approaching 0.84 and consistently higher time‐dependent AUCs (data not shown). Together, these findings suggest that registry‐based modeling can identify population‐level patterns in SS, but routine SEER variables alone are insufficient for individualized prognostic prediction.

Several conclusions follow. First, this population‐based analysis provides contemporary national survival estimates for SS across more than two decades. SS remains highly lethal: median OS is approximately 4 years, and 5‐year OS remains below 45% despite modest secular improvement. Second, Black patients not only develop SS at substantially younger ages but also experience significantly higher mortality than White patients even after accounting for age, sex, calendar period, socioeconomic factors, stage, and treatment timing. These patterns mirror disparities previously described in MF and other lymphomas and raise concern that access to specialized cutaneous lymphoma care and newer agents may not be equitably distributed.

Third, our negative prediction findings provide an important methodological benchmark for future SS modeling studies. The limited discrimination of Cox, penalized Cox, random survival forest, and gradient‐boosted Cox models should not be interpreted as evidence that SS cannot be risk‐stratified clinically. Rather, these findings suggest that routine SEER variables are insufficient to capture the disease‐specific prognostic information needed for meaningful individualized prediction [2, 3, 10, 11].

This interpretation should be considered in light of several registry‐based limitations. SS is a rare disease that requires expert clinicopathologic diagnosis and detailed assessment of skin, blood, nodal, and visceral involvement, whereas the present study relied on ICD‐O‐3 coding in SEER. SEER does not capture the full ISCL/EORTC TNMB staging framework, quantitative blood tumor burden, flow cytometry findings, T‐cell clonality, skin disease burden, nodal pathology, central pathology review, treatment sequence, comorbidities, infectious complications, longitudinal response, or molecular features. In addition, diagnostic and staging criteria for SS evolved during the study period, including broader incorporation of blood classification and flow cytometric assessment of blood tumor burden. Therefore, the observed improvement in survival over time may partly reflect stage migration and changes in diagnostic ascertainment, in addition to improvements in systemic therapy, supportive care, and access to specialized cutaneous lymphoma management [14, 15, 16].

Prior clinically granular prognostic frameworks support this interpretation. The revised ISCL/EORTC staging system and the Cutaneous Lymphoma International Prognostic Index (CLIPi) have shown that outcomes in MF/SS are influenced by disease‐specific features such as skin tumor burden, nodal/visceral involvement, blood stage, LDH, large‐cell transformation, folliculotropic disease, age, and sex [14, 17]. Many of these variables are unavailable or incompletely represented in SEER. Thus, the poor performance of increasingly complex registry‐based models in the present study likely reflects sparse covariate‐level prognostic signal in SEER rather than the absence of clinically meaningful prognostic heterogeneity in SS.

These findings support the need for collaborative, disease‐specific data collection. The PROCLIPI international registry provides an example of a prospective, multi‐institutional effort capturing granular clinical, hematologic, pathologic, imaging, treatment, response, quality‐of‐life, and survival data in cutaneous lymphoma [18]. Such expert‐adjudicated, clinically and biologically annotated datasets will likely be required to define the drivers of survival in SS and support clinically useful prognostic models.

Taken together, these findings position SS as a disease with persistently poor real‐world survival, clear sociodemographic gradients, and limited amenability to prognostic modeling using current registry data. Improving outcomes will likely require not only more effective systemic therapies but also more equitable access to expert care and prospective, biologically annotated cohorts capable of supporting truly informative risk‐stratification tools.

Author Contributions

T.Z. contributed to the literature review and writing. T.Z. contributed to the figure design. T.Z. and W.X. contributed to the conception and design. T.Z. and W.X. contributed to the proofreading and edition. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (82200206).

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: jha270332‐sup‐0001‐Figure 1.pdf

Data Availability Statement

The authors have nothing to report.