基於台灣癌登資料建構頭頸癌存活預測模型

Abstract

背景 : 頭頸癌是全球常見癌症排名第七大的癌症，發生率為14.24//100,000人，死亡率為6.87/100,000人。在台灣，頭頸癌的發生率為57.68/100,000人，死亡率則為22.34/100,000人，其中頭頸癌患者中更大多為口腔癌，發生率並為十大癌症的第五名且死亡率為十大癌症的第六名。因此，不論在世界又或是在台灣，頭頸癌都是排名前十的癌症，且台灣頭頸癌的發生率和死亡率數值又明顯較其他國家高。 在頭頸癌存活的研究中，頭頸癌的五年整體存活大約介於50%~65%之間，隨著癌症部位及種族的不同，頭頸癌的發病率、死亡率及存活時間都存在差異。因此本研究的目的是希望藉由台灣癌症登記中心的資料，建立屬於台灣人的預測模型，來預測頭頸癌患者的整體存活及腫瘤特定存活，並且使用美國癌症登記資料庫(The Surveillance, Epidemiology, and End Results ,SEER)中亞洲人的資料進行外部驗證，再進一步觀察亞洲和非亞洲患者的臨床結果是否存在差異。

方法 : 本研究主要蒐集由2013至2018年間在台灣癌症登記中心(The Taiwan Cancer Registry,TCR)被診斷為頭頸癌的患者資料，並分別針對有做手術的頭頸癌患者、有做手術的口腔癌患者和沒有做手術的頭頸癌患者建構三種Cox迴歸模型，最後會使用C-index及五年的每年存活預測，來判斷模型的預測能力和精確性。 在外部驗證方面，本研究使用2010至2015年的SEER資料，並針對兩種資料庫共有的變數進行建模，模型分別對非西班牙裔白人、非西班牙裔黑人及亞洲人進行C-index及五年的每年存活預測分析，來觀察亞洲與非亞洲之間的關係及模型的外推性。

結果 : 本研究總共建構了六個模型，其中針對有做手術頭頸癌患者的整體存活模型的解釋，為在控制了其他變數的情況下，年齡每增加一歲的死亡風險會是原來的1.02倍，女性發生死亡的風險是男性的0.74倍，腫瘤大小每增加一毫米的死亡風險會是原來的1.01倍，診斷年每增加一年的死亡風險會是原來的0.97倍，BMI每增加一單位的死亡風險會是原來的0.95倍，與口腔相比，口咽有較高的風險，其餘下咽、主唾腺、喉則都有較低的死亡風險(口咽， HR=1.24,P<0.001；下咽，HR=0.9,P=0.02；主唾腺，HR=0.72 ,P<0.001;喉，HR=0.92,P=0.16)等。 在模型的預測結果中，整體存活模型的C-index為0.7~0.74，腫瘤特定存活模型的C-index為0.76~0.78，兩者的C-index都大於0.7，代表模型有不錯的預測能力。五年的每年存活預測上，整體存活模型都沒有達到統計上的顯著差異，腫瘤特定存活模型則大多於前兩年沒有達到統計上顯著，且實際結果與預測結果的差異比例大多小於5%。在外部驗證方面，亞洲人之C-index明顯較其他種族高，但於五年的每年存活預測上，不論是非西班牙裔白人、非西班牙裔黑人及亞洲人均達到統計上的顯著差異。

結論 : 不論在整體存活模型又或者在腫瘤特定存活模型，本研究所建構的三種預測模型都有良好的預測能力，在台灣頭頸癌患者的存活預測上也有不錯的預測效果。在外部驗證方面，可以發現亞洲人相較於非亞洲人有較好的預測效果，但在每年的存活預測上，不論是亞洲又或者是非亞洲人，大多結果為實際的死亡人數與使用模型所預測的死亡人數有顯著差異，因此無法進行外推。最後，希望藉由本研究讓頭頸癌患者能夠進行自我評估，同時也提供患者與醫生對於是否接受治療在決策上得到更多資訊。

Background: Head and neck cancer is the seventh most common cancer in the world, with an incidence rate of 14.24/100,000 people and a mortality rate of 6.87/100,000 people. In Taiwan, the incidence rate of head and neck cancers is 57.68/100,000, and its mortality rate is 22.34/100,000. Among them, oral cancer ranks fifth in terms of incidence rate among the top ten cancers, and ranks sixth in terms of mortality rate among the top ten cancers. Whether worldwide or in Taiwan, head and neck cancer ranks among the top ten cancers, and the incidence and mortality rates of head and neck cancer in Taiwan are significantly higher than in other countries. The five-year overall survival rate ranges from 50% to 65%, however, the incidence, mortality, and survival time of head and neck cancer vary depending on cancer sites and race. Therefore, the purpose of this study is to develop prediction models specific to Taiwanese patients, using data from the Taiwan Cancer Registry, towards predicting overall survival and head and neck cancer-specific survival. The study will also include external validation using the American Cancer Registry Database (The Surveillance, Epidemiology, and End Results, SEER) to examine potential differences in clinical outcomes between Asian and non-Asian patients. Method: Patients diagnosed with head and neck cancer at the Taiwan Cancer Registry (TCR) between 2013 and 2018 were included in this study. Three Cox regression models were constructed for the following groups: head and neck cancer patients who underwent surgery, oral cancer patients who underwent surgery, and head and neck cancer patients who did not undergo surgery. The predictive ability and accuracy of the model were assessed using discrimination analysis via c-index and five-year annual survival prediction (calibration analysis). For external validation, head and neck cancer patients from the SEER data recruited between 2010 and 2015 were utilized for fitting models with variables that were common to both databases, as predictors. The models were evaluated using discrimination and calibration analyses to examine the relationship between Asian and non-Asian populations and assess the models’ generalizability. Results: A total of six models were constructed in this study. Among them, the overall survival model for patients with head and neck cancer who underwent surgery showed the following results: while controlling for covariates, the risk of death increased by 1.02 times for every one-year increase in age. The risk of death for females was 0.74 times to that of males. Additionally, the risk of death increased by 1.01 times for each millimeter increase in tumor size, while the risk of death decreased by 0.97 times for each year of diagnosis and increased with each unit increase in BMI. When compared to the oral cavity, the oropharynx had a higher risk of death, while the hypopharynx, main salivary gland, and larynx had lower mortality risks (oropharynx, HR=1.24, P<0.001; hypopharynx, HR=0.9, P=0.02; main salivary gland, HR=0.72, P<0.001; larynx, HR=0.92, P=0.16) and so on.The C-index of the overall survival model ranged from 0.7 to 0.74, and the cancer-specific survival model ranged from 0.76 to 0.78. Both c-index values exceeded 0.7, indicating that the models exhibited good predictive ability. The overall survival model did not show a statistically significant difference between observed and predicted survival via the calibration analysis. Most of the cancer-specific survival models also did not exhibit a statistically significant difference for the first two years, while the difference between the actual and predicted survival were < 5% for the latter years. In external validation, Asians exhibited a significantly higher c-index compared to other racial groups. However, in calibration analyses, statistically significant differences were observed between observed and predicted survival among non-Hispanic whites, non-Hispanic blacks, and Asians. Conclusions: Regardless whether it was the overall survival model or the cancer-specific survival model, the three prediction models developed in this study exhibit good predictive capabilities. The models demonstrate their effectiveness in predicting the survival of head and neck cancer patients in Taiwan. in the SEER data it was further observed that the Asians show better prediction performance compared to non-Asians. However, in in calibration analyses, both Asians and non-Asians experience significant difference between observed and predicted numbers of deaths, which makes extrapolation unreliable. Lastly, it is hoped that this study will enable patients with head and neck cancer to engage in self-assessment, thereby providing both patients and doctors with valuable information to make informed therapeutic decisions.