目的 胰十二指肠切除术是治疗胰头癌常见的手术方式之一，也是腹部外科最复杂的手术之一。本文通过回顾美国监测、流行病学和最终结果(the Surveillance, Epidemiology, and End Results, SEER)数据库可获得的数据来筛选并分析接受胰十二指肠切除术的胰头癌患者的预后因素，并建立nomogram预测模型为临床提供参考。

方法 我们筛选出1973年至2015年间被诊断为胰头癌患者的数据。使用倾向性得分匹配（Propensity score matching，PSM）控制混杂因素。使用 Kaplan-Meier（log-rank检验）曲线比较生存率。采用多因素Cox回归模型筛选并确定危险因素，绘制了预测nomogram图。通过校准图和ROC曲线来确定列线图的预测精度。

结果  2010-2015年被诊断为胰头癌并且接受胰十二指肠切除术的4099例患者数据被完整获取，其术后中位生存期为22个月，术后1年、3年和5年的生存率分别为74.2%、36.5%和26.2%; 疾病特异性生存期的中位数为24.0个月，术后1年、3年和5年的疾病特异性生存率分别为71.1%、32.6%和21.9%。Cox比例风险回归模型表明，患者的诊断年龄，保险状态、性别、组织学类型、组织分化程度、T期、N期、 肿瘤大小、 区域淋巴结清扫范围、 术后放疗、术后化疗是影响预后的独立因素。为了进一步评估辅助治疗方式（术后放疗和术后化疗）对术后生存的影响，分别采用倾向性评分匹配，以消除原始数据中不均衡的协变量带来的偏倚。在消除年龄、保险状态、性别、分化程度、组织学类型、肿瘤大小、T分期、N分期、术后化疗等潜在混杂因素对术后放疗效果的干扰后，发现接受术后放疗的患者比未接受放疗的患者拥有更好的生存预后和胰腺癌的疾病特异生存预后，HR=0.809,95%CI(0.731~0.894),P<0.001和HR=0.814,95%CI（0.732~0.904），P<0.001;而在消除年龄、保险状态、组织学类型、组织分化程度、肿瘤大小、T分期 、N分期、区域淋巴结清扫范围、术后放疗等潜在混杂因素对术后化疗的干扰后，发现接受术后化疗的患者比未接受术后化疗的患者拥有更好的生存预后和胰腺癌的疾病特异生存预后，HR=0.703,95%CI(0.633~0.78),P<0.001和HR=0.736,95%CI（0.658~0.822）, P<0.001。最后，Cox比例风险回归模型的系数β被用来构造nomogram图。校准图和ROC曲线显示该nomogram预测模型具有良好的预测能力。

结论  患者的年龄，医疗保险状态，性别，病理的组织类型及分化程度， T分期、N分期及肿瘤大小，术中的淋巴结清扫范围，术后放疗及术后化疗是影响胰十二指肠切除术治疗胰头癌的因素。术后放疗及术后化疗可以提高患者的生存率，但术前放疗并未被发现有改善术后生存预后的作用。这仍需要未来更多的高质量前瞻性研究去进一步验证。本文建立的nomogram预测模型具有良好的预测精度，有助于临床医师评估和预测胰头癌患者胰十二指肠切除术后生存情况，为临床提供更好指导。

Objective: Pancreaticoduodenectomy is one of the common surgical procedures for the treatment of pancreatic periampullary cancer, which is considered one of the most complex procedures in abdominal surgery. In this paper, we analyzed and screened the prognostic factors of patients with pancreatic head cancer receiving pancreaticoduodenectomy by reviewing the available information from the Surveillance, Epidemiology, and End Results (SEER) database in the United States, and established a nomogram prediction model for clinical reference.

Methods We screened data from the SEER database of patients diagnosed with pancreatic head cancer between 1973 and 2015. Propensity score matching (PSM) was used to control for confounding factors. Kaplan-Meier (log-rank test) curves were used to compare overall survival. Multivariate Cox regression models were employed to identify and screen for risk factors, and a nomogram prediction model was developed, whose predictive accuracy was verified by calibration plots and ROC curves.

Results 4099 patients diagnosed with pancreatic head cancer and received pancreaticoduodenectomy from 2010 to 2015 were obtained intact with a median overall survival (OS) of 22 months, and postoperative survival rates of 74.2%, 36.5% and 26.2% at 1, 3 and 5 years, respectively, and with a median pancreatic cancer specific survival (CSS) of 24.0 months, and postoperative CSS rates of 71.1%、32.6% and 21.9% at 1, 3 and 5 years, respectively. Cox proportional risk regression models showed that age, insurance status, gender, histology, degree of differentiation, T stage, N stage, tumor size, extent of regional lymph node dissection, postoperative radiotherapy, and postoperative chemotherapy were independent factors affecting prognosis. To further assess the effect of adjuvant therapy (postoperative radiotherapy and postoperative chemotherapy) on postoperative survival, propensity score matching was used separately to eliminate bias from other unbalanced covariates in the raw data. After eliminating potential confounding factors such as age, insurance status, gender, degree of differentiation, histology, tumor size, T-stage, N-stage, and postoperative chemotherapy on the effect of postoperative radiotherapy, patients receiving postoperative radiotherapy were found to have a better survival prognosis and disease-specific survival prognosis than patients not receiving radiotherapy, HR=0.809,95% CI (0.731-0.894), P <0.001 and HR=0.814,95%CI (0.732~0.904), P<0.001; After eliminating potential confounding factors such as age, insurance status, gender, degree of differentiation, histology, tumor size, T-stage, N-stage, and postoperative radiotherapy on the effect of postoperative chemotherapy, patients receiving postoperative chemotherapy were found to have a better survival prognosis and disease-specific survival prognosis than patients not receiving chemotherapy, HR=0.703, 95%CI(0.633~0.780), P<0.001 and HR=0.736, 95%CI(0.658~0.822), P<0.001. Finally, the coefficients β of the Cox proportional risk regression model are used to build the nomogram. calibration plots and ROC curves show that the nomogram has good predictive power.

Conclusion Age at diagnosis, health insurance status, gender, histological type and degree of differentiation, T-stage, N-stage and tumor size, extent of intraoperative lymph node dissection, postoperative radiotherapy and postoperative chemotherapy are factors that influence pancreaticoduodenectomy for pancreatic head cancer. Postoperative radiotherapy and postoperative chemotherapy can improve patient survival, but preoperative radiotherapy has not been found to have an improved prognosis for postoperative survival. This still needs to be further validated by more high-quality prospective studies in the future. The nomogram prediction model developed in this paper has good prediction accuracy, which can help clinicians to evaluate and predict the survival of pancreatic head cancer patients after pancreaticoduodenectomy and provide better clinical guidance.