Establishment and verification of the nomogram that predicts the 3-year recurrence risk of epithelial ovarian carcinoma

The patients diagnosed as EOC were enrolled from Peking University First Hospital, Peking University Third Hospital and Beijing Obstetrics and Gynecology Hospital between January 2003 and December 2013. The inclusion criteria were EOC patients who reached CCR after initial or intermediate cytoreductive surgery and standard adjuvant platinum-based chemotherapy. Patients who received fertility sparing surgery or with history of other malignant tumors were excluded. CCR is defined as: (1) the level of serum CA125 is within the normal range; and (2) no residual lesions are found by imaging examination after primary treatments. General information, size of residual lesions, FIGO stage, histological grade, histological type, lymph node metastasis, expression of estrogen receptor (ER), progesterone receptor (PR) and Ki67, adjuvant therapy and serum CA125 level were collected from the original medical records. All the patients were followed up by telephone call and clinical visits. Follow-up was conducted every 2–4 months in the first and second year, every 3–6 months in the third, fourth and fifth year, and every year after 5 years. All patients were followed up until 30 June 2019. The end points of follow-up were recurrence, no recurrence but death, or no recurrence and no death at the end of observation. The definition of recurrence of EOC is that the serum CA125 level is higher than the normal value (35 U/mL) and/or the recurrence focus is found by imaging examination. Recurrence-free interval (RFI) is defined as the interval between the recurrence and the end of last chemotherapy of first line treatment. The EOC patients who fulfilled the criteria from Peking University First Hospital were enrolled into the training group to establish the nomogram. And the medical data of EOC patients with 5-year standard follow-up in Peking University Third Hospital and Beijing Obstetrics and Gynecology Hospital were used for external validation of the nomogram. The flow chart of the study was shown in Fig. 1. This study was approved by the Ethics Committee of the First Hospital of Peking University (Scientific Research No. 2018–109).

Kaplan-Meier univariate survival analysis, Log-rank test and Cox univariate and multivariate regression analysis were used to screen out the factors related to recurrence in patients with EOC. Least absolute shrinkage and selection operator regression was used to analyze the related factors. The 3-year recurrence rate nomogram was established with the risk factors selected by LASSO regression. Bootstrap resampling, AUC curve and Hosmer-Lemeshow good of fit test were used to evaluate the discrimination and calibration.

One hundred ninety-three EOC patients from Peking University First Hospital were enrolled into the training group. The characteristics of these patients, including age, FIGO stage, histological grade, histological type, lymph node metastasis, residual lesion size, serum CA125 level and molecular markers of tumor tissues were summarized in Table 1. One hundred six cases (54.9%) had recurrence. The RFI ranged from 1.8 months to 173.2 months, with a median of 46.7 months. Seventy-seven cases had no recurrence; 10 cases had censored data, 9 cases had lost follow-up, 1 case died of other disease, the rate of lost follow-up was 4.7%. The results of Kaplan Meier survival analysis and log rank test are summarized in Table 1.

Cox regression univariate analysis showed that FIGO staging, histological grade, histological type, size of residual lesions after surgery, lymph node metastasis, pre-treatment CA125 level, ER expression in tumor tissue had significant differences in the impact of internal stratification on recurrence. Cox regression multivariate analysis showed that advanced EOC, histological grade and histological type were independent risk factors for recurrence of epithelial ovarian cancer. The results of specific stratification factor were shown in Table 2.

LASSO regression was used to screen the best influencing factors for the establishment of the model. The optimal number of factors used to establish the contour map prediction model was 5. The final selected model included the following 5 variables: FIGO staging, histological grade, histological type, lymph node metastasis and serum CA125 level before treatment. Each stratification factor is assigned with a specific grading value (see Table 3 for details). When the grading values of the five influencing factors are determined, the total score can be obtained by adding them together. Figure 2 showed the nomogram for predicting 3-year recurrence risks of patients with EOC. The mathematical formulas between the total score and the recurrence rate for 3 years are as follows:

For example, a patient with EOC had a serum CA125 level of 600 U/ml (21 points) underwent the initial cytoreductive surgery. Pathology result showed that she was stage IIIC (65 points), serous carcinoma (26 points), grade G3 (100 points), lymph node metastasis (41 points) and she has reached CCR after 6 cycles of standardized chemotherapy. According to the above-mentioned nomogram, the total score of the patient was 253. The relatively overall 3-year predicted recurrence rate for this patient was 82.01%.

The ROC curve of the nomogram with internal validation was shown in Fig. 3. The AUC (C statistics) was 0.828 (95% CI, 0.764–0.884). When the threshold value was set at 198, the sensitivity, specificity, positive predictive value, negative predictive value and concordance index were 88.8, 67.0, 71.8, 86.3% and 0.558 respectively. Patients with total score higher than 198 were identified with high-risk recurrence and those with total score lower than 198 were identified with low-risk recurrence. Hosmer-Lemeshow test for evaluation of calibration showed that the Chi-square value is 3.6 (P = 0.731 > 0.05), As the calibration curve shown in Fig. 4, if the 3-year predicted recurrence rate calculated by the model is within the range of 15 to 30%, the predicted value is basically consistent with the actual recurrence rate; if the predicted value is below 15% or above 30%, the predicted value is less than the actual recurrence rate, indicating that the recurrence risk is underestimated.

The medical data of 187 EOC patients from in Peking University Third Hospital and Beijing Obstetrics and Gynecology Hospital were enrolled into the external validation group. The ROC curve of the nomogram with external validation was shown in Fig. 5. The AUC (C statistics) for the validation data group was 0.803 (95% CI, 0.738–0.867). When using the threshold value of 198, the sensitivity, specificity, positive predictive value, negative predictive value and concordance index were 75.7, 77.0, 83.2, 67.9%, and 0.52 respectively. Hosmer-Lemeshow test for evaluation of calibration showed that the Chi-square value is 11.074 (P = 0.135 > 0.05).