Beyond traditional survival estimates: conditional survival and individualized prognosis in primary breast lymphoma

Introduction

Primary breast lymphoma (PBL) is a rare extranodal lymphoma, accounting for approximately 0.5% of all breast cancers and 2% of extranodal non-Hodgkin lymphomas (1-4). It predominantly affects women and typically presents as a unilateral breast mass without prior systemic involvement (1). Clinically, PBL often manifests as a palpable lesion, while imaging features are generally nonspecific and can occasionally mimic benign masses (5). Despite advances in diagnostic imaging, histopathologic evaluation, and multimodal therapy, patient outcomes remain highly variable, largely influenced by histologic subtype, disease stage at diagnosis, and treatment approach (6). The rarity of PBL has limited prospective studies, resulting in a lack of robust, evidence-based prognostic guidance for clinical decision-making.

Traditional survival estimates, such as overall survival (OS) and disease-specific survival, are calculated from the time of diagnosis and do not account for changes in risk over time (7). Conditional survival (CS) analysis offers a dynamic approach, estimating the probability of surviving an additional period given prior survival (7-10). This is particularly valuable for PBL, where early mortality is concentrated among high-risk patients, while long-term survivors may experience substantially improved outcomes. Incorporating CS into prognostic assessment allows clinicians and patients to obtain more realistic, time-updated survival expectations, which can guide follow-up intensity and therapeutic decisions. Nomograms, as individualized predictive tools, combine demographic, tumor, and treatment characteristics to provide personalized survival estimates, offering greater clinical utility than conventional staging systems (11,12). However, existing models for PBL do not account for the duration of prior survival (13-15). Integrating CS into prognostic models remains rare, which limits their capacity to support dynamic, evidence-based risk stratification for patients.

Large, population-based registries such as the Surveillance, Epidemiology, and End Results (SEER) program provide longitudinal, multicenter data with sufficient sample size to study rare cancers. Leveraging SEER data, this study aimed to characterize CS patterns in PBL and develop a validated nomogram to support individualized prognosis and clinical decision-making, addressing critical gaps in the current literature for this rare malignancy. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0301/rc).

Methods

Patient cohort

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The dataset for this analysis was obtained from the SEER program, a comprehensive cancer registry that collects information on patient demographics, tumor characteristics, treatment, and survival outcomes. After receiving approval for data use, cases of PBL were identified and retrieved through the SEER*Stat software (https://seer.cancer.gov/). This resource ensures standardized data collection and high-quality follow-up, thereby supporting reliable population-based survival analyses.

Cases were included if they met the following criteria: diagnosed between 2004 and 2021, histologically confirmed as non-Hodgkin lymphoma according to the International Classification of Diseases for Oncology, Third Edition (ICD-O-3), identified as the first primary malignancy, with the primary site restricted to the breast (codes C50.0-C50.9), and limited to female patients. Exclusion criteria comprised cases diagnosed only through autopsy or death certificate, those lacking essential clinicopathological variables, and patients with incomplete or missing survival data. For patients diagnosed with PBL, baseline demographic and clinical characteristics were extracted, including age at diagnosis, race, sex, tumor histology, Ann Arbor stage, and therapeutic interventions [surgery, radiotherapy (RT), and chemotherapy (CT)], as well as socioeconomic factors such as marital status and income. Follow-up information encompassed survival time calculated from diagnosis and vital status at the latest update.

Statistical analysis

To characterize the survival dynamics of patients with PBL, CS analysis was first applied to estimate 10-year CS probabilities, providing time-dependent prognostic information conditional on prior survival. In parallel, the overall mortality risk of the cohort was evaluated using annual hazard rate (AHR) analysis, which allowed visualization of temporal changes in the risk of death.

Subsequently, the entire cohort was randomly partitioned into training and validation sets at a ratio of 7:3. In the training set, prognostic variable selection was performed through two approaches: the least absolute shrinkage and selection operator (LASSO) regression and conventional Cox proportional hazards regression. To compare the performance of these two methods, time-dependent receiver operating characteristic (ROC) analyses were conducted, and the variable set demonstrating superior predictive ability was chosen for model development. Based on the final predictors, a nomogram model integrating CS analysis was constructed to provide individualized, dynamic prognostic estimates. Model-derived risk scores were then calculated for each patient, which were used to stratify the population into high- and low-risk groups. The distribution of risk across patients was visualized through risk score plots, facilitating assessment of heterogeneity in survival risk. Model performance was comprehensively evaluated in both the training and validation cohorts. Calibration was assessed using calibration plots at multiple time points, while discrimination was quantified by time-dependent ROC curves. In addition, decision curve analysis (DCA) was performed to evaluate the clinical usefulness of the nomogram by quantifying its net benefit across a range of threshold probabilities, representing the risk levels at which a clinician might choose to intervene or intensify follow-up. The performance of the nomogram was compared with two default strategies: treating all patients and treating no patients. A higher net benefit of the nomogram relative to these strategies indicates potential clinical value. DCA was conducted separately in the training and validation cohorts to assess the stability and generalizability of the model’s clinical applicability.

All statistical analyses were performed using R software, and a two-tailed P value <0.05 was considered statistically significant.

Results

The baseline demographic and clinical characteristics of the 1,552 patients with PBL are summarized in Table 1. Patients were randomly assigned into the training cohort (n=1,086) and the validation cohort (n=466). Overall, patients aged over 60 years accounted for 67.5% of the study population, and more than 80% of patients were White. Diffuse large B-cell lymphoma (DLBCL) was the most common histological subtype (39.0%), followed by extranodal marginal zone B-cell lymphoma (EMZL, 26.4%). With respect to disease stage, nearly half of the patients presented with stage I/II disease (48.9%), while 14.5% had stage III/IV disease and 36.6% had unknown staging information. The majority of patients underwent surgery (74.5%), while 25.5% did not. RT was administered in one-third of patients (33.0%), and CT was applied in 44.9%.

CS analysis revealed that the OS rates of patients with PBL were 85%, 77%, and 58% at 3, 5, and 10 years, respectively. Notably, the 10-year CS probability increased progressively with longer elapsed survival time. Specifically, for patients who had already survived 1 to 9 years, the 10-year CS rose from 63% to 65%, 68%, 71%, 76%, 80%, 84%, 89%, and 95%, indicating a substantial improvement in long-term prognosis with increasing survival time (Figure 1).

Figure 1 Conditional survival probabilities of patients with PBL. Conditional survival estimates were calculated at different time points, demonstrating improved long-term survival with increasing years survived after diagnosis. PBL, primary breast lymphoma.

Analysis of the AHR demonstrated that the risk of death was highest during the first year following diagnosis (8.05%) and sharply declined to 3.99% in the second year. After this initial decrease, the AHR gradually stabilized, showing a slow downward trend until reaching 2.36% by the 10th year, highlighting that the early post-diagnosis period carries the greatest mortality risk (Figure 2).

Figure 2 AHR of death among patients with PBL. The hazard rate peaked within the first year after diagnosis and declined thereafter, reflecting the early high-risk period. AHR, annual hazard rate; PBL, primary breast lymphoma.

Then, the cohort was randomly divided into a training set and a validation set at a 7:3 ratio. In the training cohort, LASSO regression using the 1-SE criterion identified age, histology, RT, and marital status as key prognostic factors (Figure 3A). In comparison, univariate Cox analysis suggested age, race, histology, stage, surgery, RT, marital status, and income as significant variables, of which multivariate adjustment retained age, histology, stage, RT, marital status, and income (Figure 3B). ROC analysis of the two variable sets demonstrated comparable discriminatory performance for predicting prognosis (Figure 4). To achieve a more parsimonious model without compromising predictive accuracy, the LASSO-selected variables were chosen to construct a CS nomogram (Figure 5). Using the nomogram-derived risk scores, patients were stratified into high- and low-risk groups in both the training and validation cohorts, revealing marked differences in survival outcomes (Figure 6A,6B). Further visualization of risk score distributions in relation to vital status indicated a higher proportion of deaths among patients classified as high-risk, underscoring the model’s effectiveness in risk stratification (Figure 6C,6D).

Figure 3 Variable selection for prognostic modeling. (A) LASSO regression identified the most relevant prognostic variables using the one-standard-error (1-SE) criterion. (B) Forest plot of univariate and multivariate Cox regression identifying independent prognostic factors. AUC, area under the curve; CI, confidence interval; LASSO, least absolute shrinkage and selection operator.

Figure 4 Comparison of model performance between LASSO- and Cox-based approaches. Time-dependent ROC curves for predicting 3-, 5-, and 10-year survival showed that the simplified LASSO-based model achieved comparable predictive accuracy to the Cox-based selection. CI, confidence interval; CT, chemotherapy; DLBCL, diffuse large B-cell lymphoma; EMZL, extranodal marginal zone B-cell lymphoma; FL, follicular lymphoma; HR, hazard ratio; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic; RT, radiotherapy.

Figure 5 Construction of the conditional survival-nomogram. The nomogram incorporated age, histology, radiotherapy, and marital status to predict individualized conditional survival probabilities at different time horizons. CS, conditional survival; DLBCL, diffuse large B-cell lymphoma; EMZL, extranodal marginal zone B-cell lymphoma; FL, follicular lymphoma; RT, radiotherapy.

Figure 6 Risk stratification based on the CS-nomogram. (A) Patients were classified into distinct risk groups according to their prognostic scores. (B) Correlation between the risk score and mortality events, showing increasing deaths with higher scores. (C) Distribution of risk scores and patient survival status in training cohort. (D) Distribution of risk scores and patient survival status in validation cohort. CS, conditional survival.

Calibration of the CS-nomogram was assessed in both training and validation cohorts. As shown in Figure 7A,7B, the predicted 3-, 5-, and 10-year survival probabilities closely aligned with observed outcomes, indicating good agreement between predicted and actual survival. Discrimination performance was evaluated using time-dependent ROC curves. In the training set (Figure 7C), the area under the curve (AUC) values were 0.787 [95% confidence interval (CI): 0.747–0.827] at 3 years, 0.780 (95% CI: 0.744–0.815) at 5 years, and 0.814 (95% CI: 0.782–0.846) at 10 years. Comparable results were observed in the validation cohort (Figure 7D), with AUC values of 0.747 (95% CI: 0.682–0.812), 0.741 (95% CI: 0.682–0.800), and 0.762 (95% CI: 0.705–0.819) at 3, 5, and 10 years, respectively, demonstrating robust predictive ability across cohorts. Clinical utility of the nomogram was further evaluated via DCA. Figure 7E,7F showed that the DCA demonstrated that the nomogram provided a higher net benefit than treat-all or treat-none strategies across a wide range of clinically relevant threshold probabilities in both cohorts, supporting its potential usefulness for risk stratification in clinical practice.

Figure 7 Performance evaluation of the CS-based nomogram. Calibration curves for 3-, 5-, and 10-year OS in the training (A) and validation (B) cohorts, demonstrating close agreement between the nomogram-predicted probabilities and the observed outcomes. Time-dependent ROC curves for 3-, 5-, and 10-year OS in the training (C) and validation (D) cohorts, indicating the discriminative ability of the nomogram over time. DCA in the training (E) and validation (F) cohorts, showing the net clinical benefit of using the nomogram for risk stratification across a range of threshold probabilities, compared with default strategies of treating all patients or no patients. AUC, area under the curve; CI, confidence interval; CS, conditional survival; DCA, decision curve analysis; OS, overall survival; ROC, receiver operating characteristic.

Discussion

PBL remains a rare and clinically challenging malignancy, with limited prospective evidence to guide prognostic assessment and individualized management. Consistent with previous reports (2,6), our population-based analysis demonstrated substantial heterogeneity in survival outcomes, reflecting the influence of histologic subtype, disease stage, and treatment strategy. The highest risk of mortality occurred during the first year after diagnosis, highlighting the critical importance of early risk identification and timely therapeutic intervention. Beyond this initial period, the AHR declined markedly, suggesting that long-term survivors may achieve progressively favorable outcomes—a pattern that underscored the value of dynamic prognostic assessment. According to World Health Organization (WHO) and International Consensus Classification (ICC) 2022 classifications (16), PBL includes biologically distinct entities such as DLBCL, EMZL, and follicular lymphoma (FL), which differ in clinical behavior and prognosis. Pooling these subtypes is necessary for statistical power but may mask subtype-specific patterns.

Our CS analysis provided novel insights into the time-dependent nature of survival in PBL. Notably, the probability of surviving 10 years increased steadily among patients who had already survived 1 to 9 years, demonstrating a substantial improvement in long-term prognosis with elapsed survival time. This dynamic perspective addressed a key limitation of conventional survival metrics, which are fixed at the time of diagnosis and may underestimate the evolving survival probability for patients who surpass early high-risk periods. Incorporating CS into prognostic evaluation enables clinicians to provide more accurate, individualized counseling and to tailor follow-up strategies according to patient-specific risk trajectories.

In our multivariate Cox regression analysis, age, histology, Ann Arbor stage, RT, marital status, and income were identified as independent prognostic factors for PBLs. Advanced age was associated with poorer OS, likely reflecting diminished physiological reserve, higher comorbidity burden, and reduced tolerance to intensive therapies in older patients (6). Among clinicopathological variables, both the Ann Arbor stage and histological subtype were strongly correlated with survival. The Ann Arbor staging system, widely applied to both Hodgkin and non-Hodgkin lymphomas, demonstrated significant prognostic value in our cohort, with higher stages associated with worse OS, consistent with previous studies (17). Histological subtypes also exhibited distinct outcomes: DLBCL was the most prevalent (39%) and carried the poorest prognosis, followed by FL (15.3%) and EMZL (26.4%), in line with prior reports (2,6,13). Marital status was identified as an independent prognostic factor, which may reflect differences in social support, healthcare access, and treatment adherence rather than a direct biological effect. Married patients are more likely to receive timely and continuous care, whereas unmarried individuals may experience delays or interruptions. However, given the lack of psychosocial data in SEER, this finding should be interpreted as an association rather than a causal relationship.

The optimal treatment approach for PBL remains a subject of debate and is largely determined by histology and stage (18-21). Current some studies recommend chemo-immunotherapy combined with RT, while surgery is generally reserved for diagnostic biopsy (22). Although radical surgery has been reported to improve local control in limited series, small sample sizes and insufficient statistical power preclude definitive conclusions (1,3). CHOP-based regimens remain standard for primary DLBCL of the breast, with targeted therapies increasingly incorporated in recent years (23-25). Interestingly, CT did not demonstrate a significant survival benefit in our cohort. Although CT was not identified as a significant prognostic factor in this cohort, this should not be interpreted as evidence against its efficacy. Established regimens such as CHOP and R-CHOP have demonstrated clear survival benefits. The lack of association in this study likely reflects data limitations and confounding. Several factors may account for this observation. First, the SEER database lacks detailed treatment information such as CT regimens, dosage, cycles, and treatment compliance, which may obscure its actual effect. Second, PBL is a heterogeneous disease with varying histological subtypes, and the sensitivity to CT differs across subgroups, potentially diluting survival benefits in the pooled analysis. Third, earlier CT protocols for PBL may have provided only modest survival benefit, and the subsequent introduction of rituximab-based immunochemotherapy and advances in RT or targeted therapy have likely played a greater role in improving outcomes, thereby attenuating the apparent contribution of CT alone. And the relatively high proportion of stage I/II lymphomas and the low proportion of stage III/IV cases (14.5%) might have further diminished the apparent survival benefit associated with CT. Moreover, selection bias may exist, as elderly patients or those with significant comorbidities were less likely to receive CT, thereby confounding survival comparisons. In contrast, RT was associated with a pronounced survival advantage, supporting its role in local tumor control and overall outcome improvement (17). The interpretation of treatment effects should be approached with caution. The SEER database lacks detailed information on treatment regimens, dosing, and response, and is subject to selection bias. Therefore, these findings should be considered exploratory rather than causal. Given the rarity of PBL, combining different histological subtypes was necessary to ensure adequate sample size and statistical power. Future studies should aim for subtype-specific analyses and prospective investigations to refine treatment strategies.

The development of a CS-based nomogram represents an additional advancement in individualized risk prediction. By integrating age, histology, RT, and marital status—variables identified through rigorous LASSO selection—our model offered favorable discrimination and calibration across training and validation cohorts. Importantly, this approach captured both baseline prognostic factors and the dynamic impact of prior survival, overcoming limitations of existing models that do not account for elapsed survival time. The ability to stratify patients into high- and low-risk groups with clear differences in observed outcomes supported the clinical applicability of this tool. DCA further confirmed its potential utility, demonstrating net benefit across a range of threshold probabilities.

Several strengths enhanced the reliability of our findings. The use of the SEER database ensured a large, population-based cohort with standardized data collection and long-term follow-up, which is particularly valuable for studying a rare malignancy such as PBL. Moreover, the combination of CS analysis with nomogram development represents a novel approach that integrates dynamic survival estimation with individualized prognostic modeling, providing actionable information for both clinicians and patients.

Nevertheless, several limitations should be acknowledged. The SEER database lacks detailed information on systemic therapy regimens, response to treatment, and comorbidities, which may influence survival outcomes. The inclusion of both aggressive and indolent lymphoma subtypes introduces heterogeneity. Due to limited sample size and event numbers within subgroups, subtype-specific analyses were not feasible and should be addressed in future studies. Due to the limitations of SEER, systemic lymphoma with secondary breast involvement cannot be completely excluded. Although we restricted inclusion to cases with the breast as the primary site and first malignancy, misclassification remains possible. Additionally, external validation in independent cohorts is warranted to confirm the generalizability of our CS-nomogram. Future studies incorporating molecular, genomic, and treatment-response data may further refine prognostic precision and guide personalized management strategies for PBL.

Conclusions

In conclusion, our study provided the population-based CS analysis for PBL and introduces a validated, individualized nomogram that incorporates elapsed survival into risk prediction. These findings offered clinicians a dynamic, patient-centered tool for prognostic counseling, follow-up planning, and therapeutic decision-making, addressing a critical gap in the management of this rare lymphoma.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0301/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0301/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1-0301/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Cheah CY, Campbell BA, Seymour JF. Primary breast lymphoma. Cancer Treat Rev 2014;40:900-8. [Crossref] [PubMed]
2. Mustafa Ali MK, Samhouri Y, Law JY, et al. Patterns of treatment and their outcomes in primary breast lymphoma; a comprehensive population-based analysis. Breast Cancer 2022;29:1022-31. [Crossref] [PubMed]
3. Radkani P, Joshi D, Paramo JC, et al. Primary breast lymphoma: 30 years of experience with diagnosis and treatment at a single medical center. JAMA Surg 2014;149:91-3. [Crossref] [PubMed]
4. Thomas A, Link BK, Altekruse S, et al. Primary Breast Lymphoma in the United States: 1975-2013. J Natl Cancer Inst 2017;109:djw294. [Crossref] [PubMed]
5. Feng K, Zhao S, Shang Q, et al. An overview of the correlation between IPI and prognosis in primary breast lymphoma. Am J Cancer Res 2023;13:245-60.
6. Maccio U, Rets A, Grosse C, et al. Clinical and pathological features of lymphomas in the breast: a comprehensive multicentric study. Sci Rep 2025;15:29032. [Crossref] [PubMed]
7. Miura JT, Lindner H, Karakousis GC, et al. Conditional survival estimates for merkel cell carcinoma reveal the dynamic nature of prognostication. J Surg Oncol 2022;126:348-55. [Crossref] [PubMed]
8. Zhao J, Gao J, Liu W, et al. Dynamic conditional survival nomogram for primary hepatocellular carcinoma: a population-based analysis. Discov Oncol 2025;16:854. [Crossref] [PubMed]
9. Hieke S, Kleber M, König C, et al. Conditional Survival: A Useful Concept to Provide Information on How Prognosis Evolves over Time. Clin Cancer Res 2015;21:1530-6. [Crossref] [PubMed]
10. Zabor EC, Gonen M, Chapman PB, et al. Dynamic prognostication using conditional survival estimates. Cancer 2013;119:3589-92. [Crossref] [PubMed]
11. Balachandran VP, Gonen M, Smith JJ, et al. Nomograms in oncology: more than meets the eye. Lancet Oncol 2015;16:e173-80. [Crossref] [PubMed]
12. Iasonos A, Schrag D, Raj GV, et al. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol 2008;26:1364-70. [Crossref] [PubMed]
13. Peng F, Li J, Mu S, et al. Epidemiological features of primary breast lymphoma patients and development of a nomogram to predict survival. Breast 2021;57:49-61. [Crossref] [PubMed]
14. Yu Y, Xu Z, Shao T, et al. Epidemiology and a Predictive Model of Prognosis Index Based on Machine Learning in Primary Breast Lymphoma: Population-Based Study. JMIR Public Health Surveill 2023;9:e45455. [Crossref] [PubMed]
15. Liu Y, Han H, Wei H, et al. Predictive Modelling of Overall Survival in Adult Patients with Primary Diffuse Large B-cell Lymphoma of the Breast Using the Surveillance, Epidemiology, and End Results (SEER) Database. Recent Pat Anticancer Drug Discov 2024;19:373-82. [Crossref] [PubMed]
16. Alaggio, R., Amador, C., Anagnostopoulos, I. et al. The 5th edition of the World Health Organization Classification of Haematolymphoid Tumours: Lymphoid Neoplasms. Leukemia 2022;36:1720-48.
17. Liu MT, Hsieh CY, Wang AY, et al. Primary breast lymphoma: a pooled analysis of prognostic factors and survival in 93 cases. Ann Saudi Med 2005;25:288-93. [Crossref] [PubMed]
18. Zhou W, Yu G, Liu L, et al. Primary diffuse large B-cell lymphoma of the breast: A retrospective study of outcomes and insulin resistance. Saudi Med J 2023;44:38-44. [Crossref] [PubMed]
19. Abbondanzo SL, Seidman JD, Lefkowitz M, et al. Primary diffuse large B-cell lymphoma of the breast. A clinicopathologic study of 31 cases. Pathol Res Pract 1996;192:37-43.
20. Aviv A, Tadmor T, Polliack A. Primary diffuse large B-cell lymphoma of the breast: looking at pathogenesis, clinical issues and therapeutic options. Ann Oncol 2013;24:2236-44. [Crossref] [PubMed]
21. Franco Pérez F, Lavernia J, Aguiar-Bujanda D, et al. Primary Breast Lymphoma: Analysis of 55 Cases of the Spanish Lymphoma Oncology Group. Clin Lymphoma Myeloma Leuk 2017;17:186-91. [Crossref] [PubMed]
22. Mouna B, Saber B. Primary malignant non-Hodgkin’s lymphoma of the breast: a study of seven cases and literature review. World J Surg Oncol 2012;10:151. [Crossref] [PubMed]
23. Teng LC, Liao YM, Gau JP, et al. Clinical Features and Outcomes of Primary Breast Diffuse Large B-Cell Lymphoma: A Matched-Pair Study. Clin Med Insights Oncol 2023;17:11795549231203142. [Crossref] [PubMed]
24. Xu XT, He DL, Tian MX, et al. Prognostic Value of Sarcopenia in Patients With Diffuse Large B-Cell Lymphoma Treated With R-CHOP: A Systematic Review and Meta-Analysis. Front Nutr 2022;9:816883. [Crossref] [PubMed]
25. Zhang T, Zhao H, Cui Z, et al. A multicentre retrospective study of primary breast diffuse large B-cell and high-grade B-cell lymphoma treatment strategies and survival. Hematology 2020;25:203-10. [Crossref] [PubMed]