Construction and validation of nomogram for the cancer-specific death in children and adolescents‑onset lymphoma

Abstract

Background

Lymphomas in children and adolescents accounted for about 25% of all lymphoma cases. The cancer-specific death (CSD) of children and adolescent-onset remains dismal and varies widely in different individuals. The current investigation aimed to develop a predicting nomogram to evaluate CSD risk in children and adolescents‑onset lymphoma with data from the Surveillance, Epidemiology and End Results (SEER) database.

Methods

The pathological variables of lymphomas patients were extracted from the SEER database. A competing risk model was used to identify independent factors for CSD. The predicting nomogram was also developed. The performance of the nomogram was evaluated with calibration curves, receiver operating characteristic curve and decision curve.

Results

A total of 7349 lymphoma cases were selected in our investigation, which were separated into the training (n = 5144) and testing (n = 2205) cohorts. The results of univariate and multivariate analysis demonstrated age, race, year of diagnosis, pathological subtype, tumor grade, tumor stage, and chemotherapy as independent risk factors for CSD in lymphoma cases. The 1-year, 3-year-, and 5-year AUCs of ROC curves of nomogram for CSD in lymphoma were 0.855, 0.827, and 0.810 in the training cohort and 0.862, 0.829, and 0.812 in the testing cohort, respectively. Further analysis suggested a good agreement between the observed outcome and the predicted probabilities in the calibration curves in training cohort and testing cohort. Moreover, decision curve analysis also indicted good clinical utility of the nomogram models in training cohort and validation cohort.

Conclusion

The nomogram shows good accuracy and reliability in evaluating the risk of CSD in children and adolescents‑onset lymphoma, and it could provide some theoretical support for clinicians to make decisions.

Introduction

Lymphoma is the most prevalent malignancies worldwide, with more than 635,000 newly diagnosed cases and 273,000 deaths in 2022 globally [1]. Lymphoma mainly includes non-Hodgkin lymphoma and Hodgkin lymphoma. The biological and clinical behavior of different cases of lymphoma are quite due to the tumor heterogeneity [2]. Based on the different molecular subtypes, clinicians stratify patients for risk and clinical outcome, which allows us select appropriate treatment and follow-up strategy.

Lymphomas in children and adolescents accounted for about 25% of all lymphoma cases [3]. Children-and-adolescent onset lymphomas cases represent a unique population as they have diverging issues in terms of psychosocial hardships/barriers, economics, and lack of standardization of therapeutic approaches. Most often, the prognosis of children and adolescents‑onset lymphomas cases was worse than adult lymphomas [4]. Characterized by high aggressiveness and recurrence, many children and adolescents‑onset lymphomas cases would experience metastatic disease after initial therapy. Once metastatic disease occurs, both clinical outcomes and quality of life decline sharply. A previous study developed an inflammatory response score for predicting overall and cancer specific survival in malignant lymphoma [5]. However, this study only included 147 lymphoma cases with the longest follow-up time of 8 year, making its result unconvincing. Thus, it is necessary to develop another novel model for predicting cancer-specific death (CSD) in children and adolescents‑onset lymphomas.

A nomogram is a graphical device that implements a regression model in a user-friendly way, allowing the user to relate subject-specific covariates to the probability of an event. Previous studies showed that nomogram have a good performance in predicting the clinical outcome of cancer patients. A competing risk nomogram was developed to predict cancer-specific survival in patients with primary bone diffuse large B-cell lymphoma, with the areas under the curve (AUCs) for 1-, 5-, and 10-year CSS were 0.879, 0.848, and 0.839, respectively [6]. Another nomogram for primary thyroid lymphoma also showed a good performance in predicting the prognosis of patients [7]. Moreover, age-based nomogram has been suggested as a predictor for long-term overall survival in lymphoid tissue lymphoma patients [8].

Using data from the Surveillance Epidemiology End Results (SEER) database comprising 7349 cases, we identified independent prognostic factors associated with CSD in children- or adolescent-onset lymphomas through multivariate Cox analysis; based on this analysis we developed a nomogram to predict CSD risk specifically within these populations. Our results may provide additional data useful for identifying high-risk cases, thus facilitating individualized treatment decisions and follow-up plans.

Method

Data source and case selection

Lymphomas cases were identified from 17 registries of the SEER database, representing about 26.5% of the US population in the 2020 census. SEER*Stat software (version 8.4.2) was used to isolate the variables of lymphoma cases, including age at diagnosis, sex, race, median household income, rural-urban distribution, subtype, tumor grade, tumor, stage, radiotherapy and chemotherapy status. Histological types of Lymphomas cases were identified according to the International Classification of Diseases for Oncology (ICD-O), 3rd edition. Ethical review is exempt and the informed consent is not required as all data are from the SEER.

lymphomas cases were selected with the inclusion criteria: (1) The pathological type is lymphomas, (2) First primary cancer; (3) 1–17 years old of age at diagnosis, (4) With complete follow-up information. (5) Died from lymphoma. Those Cases with incomplete survival information or a survival period < 6 months or diagnosed by death certificate or autopsy should be excluded from our study. Those cases diagnosed by death certificate or autopsy may have an incomplete or incorrect follow-up data.

Statistical analysis

A random sampling method was used to separate lymphoma cases into training and testing cohort with a 7:3 ratio. Differences in categorical and continuous variables between groups were evaluated by Chi-squared test and Student’s t-test, respectively. Univariate and multivariate analyses were performed to identify independent prognostic factors for CSD with competing risk model. The competition risk model was developed by Austin et al. [9]. When fitting regression models in the presence of competing risks, the effect of covariates on the cumulative incidence function was modeled differently than in standard Cox models. Compared with standard Cox models, competitive risk model could be able to better assess CSD more accurately. The previous study detailed how this model works [9]. We then constructed and verified a predicting nomogram using the ‘rms’ and ‘survival’ packages in training and testing cohort. The performance of the nomogram was evaluated by the Harrell concordance index (C-index), ROC curve and the calibration curves. Moreover, we then performed decision curve analysis (DCA), which could assess the clinical benefits of the nomogram. SPSS 25.0 software (IBM Corp., USA), R 4.1.1 software (R core team, Austria) and GraphPad Prism Version 9.0 were used to perform statistical analyses. p < 0.05 was considered statistically significant.

Result

The clinical characters of selected lymphoma cases

All lymphoma cases were chosen through a strict process (Fig. 1). As shown in Table 1 and 7349 lymphoma cases were selected for our investigation. These cases were separated into the training (n = 5144) and testing (n = 2205) cohorts. About 63% of lymphoma cases were male. Among all lymphoma cases, 3454 were Hodgkin lymphoma and 3895 were non-Hodgkin lymphoma. About 3238 (44%) lymphoma cases had distant disease. The detailed characteristics of lymphoma cases were shown in Table 1. No significant difference was found in CSD between training and testing cohort (Fig. 2).

Fig. 1.

Flowchart of the study. SEER Surveillance, epidemiology and end results

Table 1.

Clinical characteristics among lymphoma patients in the linked SEER-medicare cohort

Age	Total cohort (n = 7349)	Train cohort (n = 5,144)	Test cohort (n = 2,205)
	13 (9, 16)	13 (9, 16)	13 (9, 16)
Sex
Female	2,743 (37%)	1,974 (38%)	769 (35%)
Male	4,606 (63%)	3,170 (62%)	1,436 (65%)
Household income
$75,000-	3,910 (53%)	2,711 (53%)	1,199 (54%)
$75,000+	3,439 (47%)	2,433 (47%)	1,006 (46%)
Race
White	5,614 (76%)	3,947 (77%)	1,667 (76%)
Black	977 (13%)	683 (13%)	294 (13%)
Others	758 (10%)	514 (10.0%)	244 (11%)
Year of diagnosis
2000–2009	3,679 (50%)	2,549 (50%)	1,130 (51%)
2010–2019	3,670 (50%)	2,595 (50%)	1,075 (49%)
Rural-urban distribution
Metropolitan area	6,548 (89%)	4,586 (89%)	1,962 (89%)
Nonmetropolitan area	801 (11%)	558 (11%)	243 (11%)
Subtype
Hodgkin lymphoma	3,454 (47%)	2,411 (47%)	1,043 (47%)
Non-Hodgkin lymphoma	3,895 (53%)	2,733 (53%)	1,162 (53%)
Tumor grade
B-cell	2,436 (33%)	1,743 (34%)	693 (31%)
T-cell	1,154 (16%)	814 (16%)	340 (15%)
Others	3,759 (51%)	2,587 (50%)	1,172 (53%)
Tumor stage
Localized	1,736 (24%)	1,191 (23%)	545 (25%)
Regional	2,375 (32%)	1,689 (33%)	686 (31%)
Distant	3,238 (44%)	2,264 (44%)	974 (44%)
Ratiotherapy
No	3,685 (50%)	2,570 (50%)	1,115 (51%)
Yes	3,664 (50%)	2,574 (50%)	1,090 (49%)
Chemotherapy
No	1,525 (21%)	1,056 (21%)	469 (21%)
Yes	5,824 (79%)	4,088 (79%)	1,736 (79%)
Survival time	8.8 (4.0, 13.8)	8.6 (4.0, 13.7)	9.2 (4.2, 14.2)
Survival status
Alive	2,205 (30%)	4,751 (92%)	2,065 (94%)
Dead	533 (7.3%)	393 (7.6%)	140 (6.3%)

Fig. 2.

The clinical outcome of lymphoma cases in training and validation cohort

Construction and validation of the prediction nomogram

Table 2 showed the result of univariate and multivariate analyses, demonstrating age, race, year of diagnosis, pathological subtype, tumor grade, tumor stage, and chemotherapy as independent risk factors for CSD in lymphoma cases (all p < 0.05).

Table 2.

Univariate and multivariate logistic regression for the risk factors of cancer-specific death

	Univariate analysis		Multivariate analysis
	HR (95% CI)	p-value	HR (95% CI)	p-value
Age
1–13	Ref.		Ref.
14–17	1.12 (1.05–1.33)	p = 0.018	1.44 (1.21–1.71)	p < 0.001
Race
White	Ref.		Ref.
Black	1.59 (1.28–1.98)	p < 0.001	1.54 (1.23–1.91)	p < 0.001
Other	1.02 (0.76–1.37)	p = 0.884	1.02 (0.76–1.37)	p = 0.898
Sex
Female	Ref.
Male	1.18 (0.95–1.36)	p = 0.886
Year of diagnosis
2000–2009	Ref.		Ref.
2010–2019	0.55 (0.45–0.68)	p < 0.001	0.50 (0.41–0.62)	p < 0.001
Rural-urban distribution
Metropolitan area	Ref.
Nonmetropolitan area	1.23 (0.96–1.58)	p = 0.108
Household income
$75,000+	Ref.
< $75,000	1.13 (0.95–1.34)	p = 0.154
Subtype
Hodgkin lymphoma	Ref.		Ref.
Non-Hodgkin lymphoma	2.37 (1.96–2.86)	p < 0.001	2.21 (1.70–2.87)	p < 0.001
Tumor grade
B-cell	Ref.		Ref.
T-cell	1.94 (1.56–2.40)	p < 0.001	1.68 (1.35–2.08)	p < 0.001
Others	0.73 (0.59–0.89)	P = 0.002	0.95 (0.73–1.23)	P = 0.698
Tumor stage
Localized	Ref.		Ref.
Regional	1.12 (0.86–1.46)	p = 0.395	1.51 (1.15–1.98)	p = 0.003
Distant	1.88 (1.48–2.37)	p < 0.001	2.06 (1.63–2.62)	p < 0.001
Ratiotherapy
No	Ref.
Yes	0.87 (0.72–1.15)	p = 0.725
Chemotherapy
No	Ref.		Ref.
Yes	0.76 (0.59–0.92)	p < 0.001	0.69 (0.57–0.90)	p < 0.001

Using these pathological variables, we then developed and verified the prediction nomogram in training cohort (Fig. 3A) and testing cohort (Fig. 3B). Based on the points of each pathological variable, we could calculate the total score of lymphoma case, which allows us to predict the probability of the CSD in lymphoma.

Fig. 3.

Construction and validation of the prediction nomogram. The nomogram for training (A) and validation (B) cohort. ROC curve (C, D), calibration curve (E, F) and DCA curve (G, H) for training and validation cohort. ROC area under receiver operating characteristic, AUC areas under the curve, DCA decision curve analysis

Evaluation of the performance of the prediction nomogram

The 1-year, 3-year- and 5-year AUCs of ROC curves of nomogram for CSD in lymphoma were 0.855, 0.827 and 0.810 in the training cohort and 0.862, 0.829 and 0.812 in the testing cohort, respectively (Fig. 3C and D). And the C-index in in training cohort and testing cohort was 0.83 (95% CI: 0.79–0.87) and 0.82 (95% CI: 0.79–0.86), respectively. As shown in Fig. 3E and F, further analysis suggested a good agreement between the observed outcome and the predicted probabilities in the calibration curves in training cohort and testing cohort. Moreover, DCA also indicated good clinical utility of the nomogram models in training cohort and validation cohort (Fig. 3G and H). Based on the total score of each lymphoma case, we divided them into two groups. As shown in Fig. 4A and B, there are poor clinical outcomes in lymphoma cases with high-risk score in training cohort and validation cohort (p < 0.001).

Fig. 4.

The clinical outcome of lymphoma cases in different group. The clinical outcome of lymphoma cases with different risk score in training (A) and validation (B) cohort

Discussion

The cancer death prediction model developed using traditional survival analysis methods considers cancer-specific death cases as the endpoint, and deaths caused by other diseases are regarded as missing data and are not included in statistical analysis. In reality, lymphoma patients often die from a variety of causes, with competing risks. The possibility of CSD may be affected by non-cancer-specific death. In traditional survival analysis, the effect of competing risk factors on death of cancer cases was ignored, exaggerating the risk of CSD [10]. Cancer-specific death based on traditional survival analysis methods is higher compared to that derived from competitive risk models [11]. To avoid that, a statistical method named competing risk model was developed by Austin et al. [9]. As the effects of other risk factors on CSD were considered in the competing risk model, it is more in line with clinical practice to study the prognosis of patients. In our investigation, we first used a competitive risk model to identify the risk factors for the CSD in lymphoma. The results of univariate and multivariate analysis demonstrated age, race, year of diagnosis, pathological subtype, tumor grade, tumor stage, and chemotherapy as independent risk factors for CSD in lymphoma cases. Moreover, we then developed a prediction nomogram, which had a good performance in predicting the CSD in lymphoma.

It should be noted there are some potential biases and confounding factors that may influence the nomogram’s accuracy. After the nomogram is constructed, the regression coefficients of the model will be fixed, but the individual patient characteristics will continue to change. In the real world, clinical parameters used for constructing the nomogram may affect each other, which also ampact the accuracy of nomogram. Since there are no existing models in the current literature, we cannot make a comparison. A high AUC does not mean high practical value and whether nomogram is suitable for clinical application need to be evaluated by DCA. Based on the results of DCA, the current nomogram model performs well in clinical practice.

There is evidence that age was involved in the origin, tumor progression and clinical outcome of lymphoma [12]. When making clinical decision, age is taken as one of the important factors [13]. Compared with younger patients, older patients often present with lower tumor grades, higher incidences of metastasis and recurrence in cancers [14, 15]. There are more underlying heart diseases, cerebrovascular diseases, and diabetes in older patients compared with younger patients, which affects both drug therapy effectiveness and patient longevity.

Racial disparities have been shown in the clinical outcomes of many types of cancer. Black women with ERBB2-negative breast cancer had a worse recurrence-free survival compared to White women [16]. Moreover, Whites have a more favorable clinical outcome compared to blacks in head and neck cancer [17]. Similarly, in colorectal cancer, patients at an advanced stage have worse clinical outcome if they are black [18]. In our study, the results also suggested being Black as a risk factor for lymphoma cases.

Some limitations could be found in our investigation. Bias in cases selection could not be avoided. We could not obtain certain social information about patients, such as family history, menstrual history, fertility, genomic status, and concomitant diseases, which may affect the result of our study. Furthermore, all the analyses were performed based on the population information of the United States, which may be different from that in other countries. And it would be better to perform external validation using non-U.S. cohorts. The International Prognostic Index (IPI) is a validated scoring system to predict the survival rate of patients with newly diagnosed aggressive lymphoma. In order to further verify the possibility of the clinical application of our nomogram, it is necessary to compare the nomogram with the IPI score. Some patients who have received cancer related treatment at first coming may also be include in our study, which could affect our results.

Conclusion

All in all, the nomogram shows good accuracy and reliability in evaluating the risk of CSD in children and adolescents‑onset lymphoma, and it could provide some theoretical support for clinicians to make decisions.

Abbreviations

SEER

The Surveillance, Epidemiology and End Results

CSD

Cancer-specific death

ROC

Receiver operating characteristic

AUC

Area under curve

DCA

Decision curve analysis

Author contributions

Ting Ding: Writing-Original draft preparation and Investigation; Hongyan Peng: Formal analysis; Si Dong: Data curation; Junquan Zeng and Weifang Gao: Formal analysis and Data curation; Jinmei Shao and Yongliang Zheng: Writing-Review and editing, Project administration.All authors reviewed the manuscript.

Funding

This study was funded by Jiangxi Provincial Natural Science Foundation (NO.20242BAB20374); Jiangxi Educational Department Foundation of China (NO.GJJ2209312 and GJJ2401513); Science and technology program of Jiangxi Provincial Health Commission (NO.202410377, 202310853 and 20203789); Science and technology project of Jiangxi Provincial Administration of Traditional Chinese Medicine (NO.2023A0311), and Ji’an science and technology plan special topic (20222–201786).

Data availability

The datasets generated and/or analyzed during the current study are available in the the Surveillance, Epidemiology, and End Results Program (SEER) database (Incidence SEER 17 Registries Nov 2021 Sub) at https://seer.cancer.gov/data-software/documentation/seerstat/nov2021/.

Declarations

Ethics approval and consent to participate

Our study is not involved in the use of human subjects and animals. No ethical approval and informed consents are required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.