Establishment and Validation of a Clinically Predictive Nomogram Model for Thyroid Carcinoma Patients

Abstract


Background
The incidence of thyroid carcinoma has been rising yearly around the globe and the whole incidence of all thyroid carcinoma patients increased 3% annually according to Hyeyeun Lim's research article published on JAMA (1). Despite the increasing incidence, mortality has declined gradually due to the development of clinical treatment (2). The prognosis of thyroid carcinoma can be affected by various factors including gender, age, histology type, tumor (T-stage), node (N-stage), metastasis (M-stage), treatment strategy and so on. The current 8 th edition of AJCC Cancer Staging Manual already provided with rather an effective classi cation, but given these speci c stages as well as other factors, how do we predict the actual prognosis within certain years both visually and accurately? Which becomes imperative to build a clinical prediction model to nd out which ones of those various factors are the most signi cant in terms of prognosis and explain it to patients with an easily understandable way, nomogram. Nomogram is a statistical tool which can transform a clinically predictive model into a visual graph which produces a numerical probability of a certain clinical event such as death or recurrence (3). Nomogram is even more accurate and applicable in clinical prediction compared to conventional staging and has been widely used in medicine eld, especially in oncology (4). Unfortunately, there hasn't been any effectively predictive nomogram models taking all useful factors (factors can maximize patients' net bene t) into consideration for thyroid carcinoma patients based on huge population data (5)(6)(7), whereas our study presented such accurate tool timely.

Study population
The design of this model was based on patients from Surveillance, Epidemiology, and End Results (SEER)

Program supported by the Surveillance Research Program (SRP) in NCI's Division of Cancer Control and
Population Sciences (DCCPS). As shown in Table 1, a total of 134,962 patients were diagnosed with thyroid carcinoma from 2004 to 2015, 32,783 of which were male and 102,179 of which were female.
Each variable was strati ed by gender and its percentages did not contain N/A cases. T0 was excluded from our study because there were only 290 cases identi ed and didn't have statistical value. Also, TX, T NOS, NX, N NOS, as well as MX were excluded because they didn't have signi cance for our study. All results and graphs were produced by R Project 3.6.1, Empower Stats 2.20 and IBM SPSS statistics 23.

Variable selection and univariate analysis
Gender is considered a risk factor which affects the outcome of thyroid carcinoma and it's well-known to us that female has a better chance of survival than male. The impact of gender has always been controversial as some researchers claim that there is a signi cant statistical difference in terms of survival between male and female(8), yet others say there is indeed a difference when gender is considered as an independent factor and this so-called signi cant difference will vanish in multivariate analysis (9). That's why, in our study, we analyzed gender in uence in both univariate and multivariate model to explore whether this factor can be an independent predictive variable.
Age as a risk factor is introduced into the 8 th edition of the AJCC Cancer Staging Manual and is divided into two groups using 55 years old as a threshold instead of the previous 45 years old in the 7 th edition(10). This is very important for patients between 45-55 years old in the purpose of preventing overstaging in low-risk patients and preventing over-aggressive treatment (11).
Among the four main types of thyroid carcinoma, ATC (Anaplastic thyroid carcinoma) is the one with the rarest incidence and accounts for the majority of deaths from thyroid carcinoma despite its rare morbidity due to its malignant character (12). By contrast, PTC (Papillary thyroid carcinoma) is the commonest type with an excellent prognosis (survival rates of >95% at 25 years) and can be especially found among women (13). FTC (Follicular thyroid carcinoma) is another less common type of well-differentiated thyroid carcinoma. MTC is an aggressive form of thyroid carcinoma causing about 8% to 15% of all thyroid cancer-related deaths (14). Different histology comes with a different prognosis and because of this difference, it's important to put histology variable into univariate analysis to see how they contribute.
There are many changes in the latest 8 th edition of the AJCC cancer staging manual: For PTC, FTC and ATC, T3a is a new category and refers to a tumor >4 cm in greatest dimension limited to the thyroid gland (this number is ≥4cm for MTC), T3b is a new category and is de ned as a tumor of any size with gross extrathyroidal extension invading only strap muscles (sternohyoid, sternothyroid, thyrohyoid, or omohyoid muscles), as well as level VII lymph nodes were added to N1a and MTC has been removed from above becoming a new chapter (15). Because of these changes and the latest version of SEER program didn't provide with details of 8 th edition, we converted all the patients selected from 6 th edition and 7 th edition to 8 th edition using IBM SPSS for further analysis.
There are mainly ve strategies for DTC (Differentiated thyroid carcinoma) patients treatment including: TSH-suppressive therapy, 131 I therapy, locoregional and adjuvant/adjunctive treatments (like surgery, radiotherapy, thermal/ethanol or cryoablation or embolization), targeted treatment, redifferentiation and other novel therapeutic approaches (16). All ATC patients fail to uptake iodine and are usually resistant to chemotherapy and the preferred strategy is surgery according to the American Thyroid Association (ATA) guidelines (17). As for patients with unresectable primary tumors, the role of surgery is to establish advantageous conditions to further perform palliative protocols (18). Different strategy should produce different prognosis, so we selected three factors including chemotherapy, 131 I therapy, and surgical method to explore whether these treatment factors can be used as predictive variables.
All those factors above are associated with prognosis of thyroid carcinoma, so we evaluated in uences of these factors by putting them into univariate COX regression model and Kaplan-Meier model.

Multivariate analysis and variable screening
To nd out whether a certain variable still shows signi cantly statistical difference when other variables exist at the same time, we had to put all these variables into a COX regression model for multivariate analysis. COX model, also known as proportional hazards model, is widely used in medical researches to analyze the in uences of multiple risk factors (19). In this step, we discarded those variables which may show signi cantly statistical difference in univariate analysis but may not in multivariate COX analysis.
This COX model could produce several coe cients which later was used to develop a nomogram model.

Test of clinical use
Conventionally, there are mainly several diagnostic test indicators such as sensitivity, speci city and AUC as demonstrated below and these indicators only measure the diagnostic accuracy of the prediction model, but fail to consider the clinical availability of it. DCA (Decision Curve Analysis) is such a novel tool which can be used to evaluate whether a prediction model has clinical usage by calculating the value of net bene t within certain range of threshold probabilities (20,21). This net bene t is produced by comparing the difference between expected bene t and expected harm related to each proposed testing and treatment method (22). We used this tool to analyze the clinical availability of the nal model.

Design and validation of predictive nomogram model
Based on cox model nal results (coe cients of all variables), we then used an R package called RMS to plot a nomogram to estimate 1-year, 3-year and 5-year survival probability with a line segment (23). In order to test the accuracy of this model, we divided all patients into two groups randomly-The rst dataset, training set, was used to build the nomogram model accounting for 80% (94,474 cases) and the second dataset, validation set, was used for external validation accounting for 20% (40,488 cases). The accuracy of this nomogram model can be evaluated by AUC, C-index (Harrell's Concordance Index), and calibration plot (24,25). We used this model to predict patients' survival probability of 1-year, 3-year and 5year time point and calculated the AUCs, C-indexes as well as calibration performances of each time point of each dataset.

Univariate analysis
In univariate COX regression and Kaplan-Meier analysis ( Test of clinical use Figure 1 is a comparison between model 3 and model 3 without 131 I therapy. It showed that using either one of these two models to predict patient's prognosis would obtain more net bene t compared to treatall-patients group or treat-none-patients group. However, within most of the threshold probability range, DCA indicated that model 3 without 131 I therapy would de nitely add even a lot more net bene t when comparing model 3 in terms of predicting patients' prognosis. Both prediction models were absolutely clinically useful, but model 3 without 131 I therapy was the best ideal one to maximize patients' net bene t.  (Table 4, Figure 3 ). Figure 4 showed the calibration curve of each time point, X-axis stood for predicted survival probability and Y-axis stood for observed survival probability. The gray line was the ideal calibration segment. In Figure 4, all calibration curves twisted around the gray ideal one.

Discussion
In this study, we developed a nomogram model using univariate and multivariate analysis method based on 134,962 thyroid carcinoma patients' clinical data. The nal nomogram model consisted of 6 variables including gender, age, histology type, T-stage, N-stage as well as M-stage. Its accuracy had been demonstrated by C-index, AUC as well as calibration plot and its clinical availability had been demonstrated by DCA. Our study showed that this nomogram model could be used to predict patients' survival probability.
The nal nomogram was developed after going through 4 procedures. Firstly, we selected several positive risk factors which can worsen the nal outcome such as age, gender and TNM classi cation and several negative treatment factors such as 131 I therapy and chemotherapy which can cure or alleviate the nal outcome. We could then nd out whether these negative treatment factors can be used to predict patients' survival probability. Secondly, we screened out variables by putting them into univariate and multivariate analysis. In univariate analysis, all variables showed signi cantly statistical differences, so we then put them into multivariate analysis to see how they perform. In model 1, surgical method showed no statistical difference (P value S/N T: 0.24, TT: 0.89), which means surgical method had no effect on prognosis generally, so we removed this variable to obtain model 2. Without surgical method, chemotherapy showed no statistical difference (P value: 0.23). Likewise, we excluded chemotherapy to obtain model 3. In model 3, all variables showed signi cantly statistical differences. Thirdly, we used DCA to test its clinical usage and we found out that although model 3 and model 3 without 131 I therapy were both clinically useful, DCA suggested that model 3 without 131 I therapy was the best ideal one to maximize patients' net bene t in terms of predicting prognosis alone. This result also suggested that using negative treatment factors to predict patients' prognosis is inappropriate. Though 131 I therapy is considered to be an effective strategy, it's often used to treat DTC and is not suitable to treat other types of thyroid carcinoma (26,27). Because of this, the nal nomogram model did not contain 131 I therapy. Lastly, we used several indicators including C-index, AUC as well as calibration plot to evaluate its accuracy.
In univariate analysis, the HR value of female group was 0.38 compared to male group and this gure increased to 0.77 in multivariate analysis but still showed signi cantly statistical difference. This result indicated that gender is an applicable factor to develop a predictive model. Hwang, S.H., et al.'s study also manifested that male gender is indeed a signi cant independent risk factor (28). Only few studies elaborated the disparity caused by gender. Zhang, L.J., et al.'s work suggested using castrated mice model, that this difference is likely to be caused by the role of testosterone which reduces the tumorsuppressive effects of Glipr1 and Sfrp1 by restraining the secretion of CCL5 during cancer progression, a chemokine which activates and reinforces the antineoplastic immunologic function (29). Consequently, male patients often present advanced thyroid carcinoma (30). Several reports also manifested the higher rate of malignant thyroid nodules among male patients (31)(32)(33).
Our study showed that age variable presented a pretty higher coe cient and HR value whether in univariate analysis or multivariate analysis and the gure was even higher than T3a stage in multivariate analysis. This suggested that age is an important prognostic indicator for thyroid carcinoma. Younger patients often achieve a much better prognosis compared to older ones even they have the same degree of disease. The mechanism behind this is still unclear. One hypothesis is that older patients have a much higher level of thyroid stimulating hormone which may stimulate the mutations of TSHR (Thyroid stimulating hormone receptor) causing high rate of malignant character through two pathways including the cAMP pathway via G αs and the Ras dependent MAPK pathway via G βγ and PI3K γ (34)(35)(36). Another hypothesis is that older age damages an individual's immune system like lymphocyte which can limit the invasion of malignant nodules (37).
The 8 th edition of AJCC cancer staging manual subdivided N0 stage into two stages--N0a and N0b. The de nition of N0a is one or more cytologically or histologically con rmed benign lymph nodes and the de nition of N0b is no radiologic or clinical evidence of locoregional lymph node metastasis (38). In our analysis, our result suggested that N0b patients indeed have a higher HR value compared to N0a (2.06, 95% CI: 1.81, 2.35 in univariate analysis, 1.53, 95% CI: 1.26, 1.85 in multivariate analysis) ( Table2 and   Table3). This also demonstrated the advantage of current 8 th staging method which can provide clinicians with the most speci c detail to treat patients with different stages. There are two limitations to our study. First, serological indexes were not considered currently because it's not available in the SEER database. For example, serum thyroglobulin is a necessary index which can monitor recurrence or progression of DTC and further guide the adjustment of follow-up plan and treatment strategy (40). Second, T0 stage was excluded due to the lack of cases and this made contrast less accurate compared to N-stages or M-stages. Further study will be required.

Conclusion
Generally, our study presented a nomogram model which was not only accurate but also clinically useful than conventional predictive model in terms of predicting the outcome of thyroid carcinoma. This nomogram model is also suitable to apply in other elds especially in oncology and is worth promoting vigorously. Author's contributions RZ, ZM and KX contributed to the conception and design of the study. RZ wrote the manuscript. XL, MW, QJ, SW, XZ, XH, CH, YF and HW revised the manuscript. All authors contributed to manuscript revision, read, approved the submitted version and agreed to be accountable for all aspects of the research in ensuring the accuracy of this study. All authors have given consent to the publication of this manuscript.