Prognosis prediction in esophageal signet-ring-cell carcinoma: a competing risk analysis

Esophageal cancer (EC) is a common malignancy of digestive system and is reported to have 604100 new cases globally in 2020, with Asia accounting for 79.7% [1]. Although esophageal cancer is often diagnosed as squamous-cell carcinomas, [2, 3] the incidence of adenocarcinoma has increased over the past two decades [4]. Esophageal signet-ring cell carcinoma (ESRCC) is a special subtype of esophageal adenocarcinoma characterized by the abundant accumulation of intracellular mucin, which replaces and compresses the nucleus to the periphery of the cell, thereby forming the classical signet-ring shape [5]. As reported in the western studies, the incidence of ESRCC is estimated to float from 3.5% to 5% of all esophageal malignancies [6‐8].

The current understanding of prognostic implications of ESRCC is limited and somewhat inconsistent. Owing to most SRC carcinoma deriving from stomach or gastroesophageal junction, the prognosis of SRC carcinoma primarily originated from esophagus is not well studied [9]. Compared to usual-type esophageal adenocarcinoma, available literature pointed out that ESRCC patients tend to present with more aggressive feature, less responsive to induction therapy, and have worse overall survival (OS) and disease-free survival (DFS) after surgery, all of which support SRC histology as a predictive marker for poor prognosis [6, 7, 10‐12].

However, one research from the United Stated claimed that SRC does not necessarily portend a worse prognosis [13]. These discrepancies motivate more efforts to provide an overview of survival analysis and find out potential risk factors affecting long-term prognosis. In addition, large population-based study never specifically spotlighted on the prognosis prediction model for ESRCC. The Surveillance, Epidemiology, and End Results (SEER) database covering around 28% U.S. population provides a relatively large sample-based database for exploring such a rare cancer type [14]. Moreover, there are frequently multiple time-to-event outcomes in clinical follow-up. Death is not only attributed to primary tumor or relapse, but also caused by non-ESRCC-specific reasons, such as toxicity of interventions patients underwent. These different outcomes are in competition and might lead to result bias if only performing Cox proportional hazard model. Therefore, the competing risk analysis is more appropriate when dealing with multiple event outcomes.

To address above issues, this study is to identify risk factors affecting cancer-specific survival (CSS) in patients with ESRCC. Besides, we aim to construct and validate a competing risk nomogram model in SEER database to predict 1-year, 3-year, and 5-year cancer-specific survival.

Given the competing risks event that existed, we retrospectively analyzed the SEER database and constructed a nomogram model to evaluate independent prognostic factors and predict individualized 1-year, 3-year, and 5-year ESRCC-specific survival. This nomogram model showed a favorable discriminative ability, prediction accuracy, and clinical availability in both training and validation cohorts.

The incidence of esophageal adenocarcinoma has significantly increased in the western countries [15]. Especially, ESRCC was anticipated to account for 3–3.5% of all esophageal cancers [6‐8]. This rare cancer was characterized by several clinicopathological features, including age, gender, race, material status, primary site, tumor grade, tumor stage, regional lymph nodes examined, positive lymph nodes, distant metastases, treatment modalities. There are numerous studies have spotlighted on the prediction model for esophageal adenocarcinoma. For instance, one study [16] established and validated a nomogram to predict individual survival of esophageal adenocarcinoma with six variables, including tumor size, T, N, M, age, grade. Also, its nomogram presented superior risk-stratifying capability than AJCC TNM staging system.

However, a majority of them performed traditional methods to analyze the survival data, such as Kaplan–Meier curve and Cox regression model, both of which are only suitable for single time-to-event outcome. Regarding time-to-event analysis, multiple reasons can lead to death, especially for patients with a long survival time. Notably, the occurrence of one kind of death might hinder the observation of other sorts of deaths [17]. For example, one study suggested that non-lung-cancer-specific death would confound the prediction of lung-cancer-specific death. Such a bias would increase along with age [18, 19]. Therefore, the risk of interest events might be overestimate by Kaplan–Meier method and the hazard ratio would be mistakenly evaluated by Cox regression model [20]. The competing risks caused by non-ESRCC might be ignored and censored. As reported, [21] if the percentage of competing event exceeded 10%, the Kaplan–Meier curve and Cox regression model would lead to severe result biases. Consequently, introducing CSS with competing risk model when predicting the prognosis of ESRCC patients is particularly essential.

To address above issues, competing risk model, a proportional sub-distribution hazards (SH) regression model, has been frequently conducted in survival analysis. But few researches focused on ESRCC-specific survival. To fulfill the research gaps, we estimated 5-year cumulative incidence of ESRCC in the SEER database from 2010 to 2015 and constructed a well-calibrated prognostic nomogram to predict 1-year, 3-year, and 5-year CSS. Consisting with prior studies of ESRCC, predictive parameters included the gender, lung metastases, liver metastases, and receiving surgery.

In the training cohort, gender made statistical difference in ESRCC-specific mortality. Male patients tended to have a poorer CSS than female patients (HR = 2.309, 95% CI: 1.259–4.236, P = 0.007). But the validation cohort showed there was no association between gender and prognosis in ESRCC patients (HR = 1.641, 95% CI: 0.893–3.017, P = 0.11). The selection bias might explain this paradoxical finding. Female patients were significantly less than male patients in our study, consistent with previous findings indicating a higher incidence in males [22‐26]. Sex-specific variants and androgen/estrogen balance associated with esophageal adenocarcinoma might improve the understanding of the reason for male predominance [27, 28]. Therefore, selection bias cannot be completely avoided owing to the different levels of disease exposure. With a limited enrollment of female patients, the median survival time is easily affected by individual difference. Overall, on the one hand, our nomogram model combined gender with other parameters showed a good discrimination and accurate prediction, supporting gender as a predictor for CSS. On the other hand, our finding indicated a poorer prognosis in men but conflicted with some studies manifesting no sex difference in prognosis observed for esophageal adenocarcinoma [28‐30]. Prospective studies are urgently required to further clarify the sex difference in prognosis.

Liver (coefficient = 0.588) and lung (coefficient = 0.94) distant metastases were two predominant predictors of this competing risk nomogram, both of which acted as independent risk variables for CSS. It is because distant metastases tend to indicate high malignancy and strong invasion ability of tumor, which is connected with limited treatment options, decreased survival, and worse prognosis. A real-world study [31] reported postoperative pathologic stage might be an independent factor associated with distant metastases (P = 0.004, HR = 0.15, 95% CI: 0.041–0.552). Particularly, T0-T2 stage had a lower risk of metastases (P = 0.011, HR = 0.119, 95% CI: 0.023–0.610). Consistently, in our multivariate SH model, T3 (P < 0.001, HR = 4.336, 95% CI: 1.883–9.987) and T4 (P = 0.004, HR = 3.173, 95% CI: 1.435–7.017) stage showed a higher risk of cancer-specific mortality in the validation set. But in the training set, the wide 95% CI and P-value were unsatisfied. It is possible that T stage functions as a risk factor for distant metastasis but contributes less to ESRCC-specific survival.

Surgery (coefficient = -0.654) is the only protective factor determined in the final competing risk nomogram model. Our study suggested ESRCC patients without surgery tend to have a shorter CSS and worse prognosis than those receiving resection. Both multivariate SH analysis and cumulative incidence curves supported this finding. As known, [32] surgery is the primary option for esophageal tumor, even though only 34.6% of patients with esophageal adenocarcinoma received complete resection. Surgical methods include esophagectomy only and esophagectomy plus other excisions. A prior study supported the OS (HR = 0.366, P < 0.001) and CSS (HR = 0.36, P < 0.001) of esophageal adenocarcinoma patients who underwent surgery were significantly longer than those without surgery, which is exactly consistent with our prediction model [32]. However, a retrospective analysis [33] put forward an interesting finding. Despite receiving modern induction therapy and surgery, the five-year OS of patients with ESRCC was only 31.3%, considerably less than historically reported for esophageal adenocarcinoma. We could infer that esophagectomy is an independent protective factor for ESRCC-specific survival, although it brings fewer clinical benefits to ESRCC patients than those without the SRC signature. Notably, either chemotherapy or radiotherapy showed limited prognostic values affecting CSS in our study. Due to the lack of enough evidence, the current standard care for ESRCC is still debatable [34]. Whether advanced ESRCC patients should firstly undergo surgery resection or multimodality therapy is still unclear. However, our study provided necessary evidence that surgery could significantly improve CSS for operable ESRCC patients. It highlighted the need to increase exposure to esophagectomy during early clinical decision-making. Some patients might be unresectable during planned esophagectomy or progressed during induction therapy [33], so surgical resection should be performed as early as possible after initial diagnosis. Novel biomarkers are urgently required to predict the efficacy of esophagectomy. Further clinical evidence-based data are required to compare the benefits of chemoradiation plus surgery versus surgery alone in prospective studies.

Overall, signet-ring-cell histology is a rare subtype of esophageal adenocarcinoma, which is known to have less well response to induction therapy and have decreased OS compared to patients with non-SRC histology [9]. With the increasing incidence of ESRCC, especially for western countries, more attention should be paid to construct prognosis prediction model. Since ESRCC are commonly associated with many complications, there are a sum of possibilities leading to biased results. Therefore, a competing risk model is of particular significance. To our best knowledge, this is the first study building the competing risk nomogram for prognosis prediction of patients with ESRCC. The performance of this nomogram appears to have effective discrimination ability, adequate model calibration, and a satisfying clinical net benefit. Our study provides several significant implications for clinical practice. Firstly, based on this large population-based study, it is necessary to increase exposure to resection in the early clinical decision-making, which may bring more benefits to ESRCC patients compared to other treatment strategies. Secondly, our nomogram model is an efficient approach to predict 1-, 3-, and 5-year CSS for patients with ESRCC. All variables could be easily obtained from clinical work. By calculating the score of each feature, this model could assist clinicians to make an accurate, comprehensive, and quick prognosis judgement and develop individualized therapy for each ESRCC patient. Thirdly, this competing risk model is an economic tool to avoid medical overuse, consisting of overdiagnosis and overtreatment [35‐37]. For unresectable ESRCC patients accompanying with more than one risk factor, overtreatment may place patients at risk of unnecessary adverse events. It is therefore clinicians could adopt palliative symptomatic treatment, aiming to alleviate patients’ pain and release economic burdens.

Our study has several outstanding advantages. First, our study is based on the large sample size provided by SEER database, ensuring the nomogram model robust enough. Second, cumulative incidence curve and multivariate SH model were employed to precisely estimate CSS by controlling the competing events. Third, we conducted decision curve analysis to reflect net benefit of the nomogram model, which is usually ignored in other similar reports. Meanwhile, the study has some limitations that deserves discussion. First, some recognized prognostic factors (such as genetic mutations) were not available in the SEER database. We failed to involve these parameters in our competing risk analysis. Second, since ESRCC is extremely rare in China, we lack enough data from the real world to make external validation. Third, most patients in our study were white and came from American. The universality of our competing risk model might be restricted. Finally, we only performed the internal validation for the predictive model, additional external validation is also required to guarantee its reliability and suitability.

In conclusion, a competing risk nomogram for ESRCC was successfully constructed and internally validated. This model is expected to predict 1-year, 3-year, and 5-year cancer-specific survival, and help oncologists and pathologists in clinical decision making and health care management for ESRCC patients. Future studies of ESRCC connected to molecular mutations, are vital as well for developing individualized therapy and improving cancer-specific survival.