Patients deemed to have a high suspicion of cervical cancer with strong evidence from PET/CT (PET/CT examinations were performed within one week before treatment) were retrospectively enrolled from January 1^st, 2013, to December 31^th, 2015 with the following criteria: (1) The patient underwent surgery (open surgery) and was confirmed as cervical squamous cell carcinoma by postoperative pathological results; (2) According to FIGO2018 pathological staging, the patient was confirmed to be stage IB1 or IB2; (3) Patients who had not undergone other treatments without surgery, and (4) Patients with complete follow-up records.

The Institutional Review Boards at the local institutions endorsed the study. We implemented it in accordance with the ethical standards of the 1964 Helsinki Declaration and subsequent amendments, and gave up informed consent of all participants.

Numerous studies have compared the effect of the Minimally Invasive and Abdominal Radical Hysterectomy for cervical cancer (20–22). The conclusion shows that for early cervical cancer the minimally invasive surgery (laparoscopic or robot-assisted radical hysterectomy) has a higher recurrence rate than open surgery. In recent years, patients of our hospital have also accepted open abdominal surgery as the main surgical method. Therefore, the patients included all received the open abdominal Type-C radical hysterectomy. The incision was located on the lower abdomen. The resection scope included the uterus, parauterine, upper vagina, and partial tissues around the vagina and pelvic lymph nodes. A sufficient length of adjacent connective tissues, including the front vesicocervix ligament (anterior and posterior lobes), the lateral main ligament, the posterior uterosacral ligament and rectovaginal ligament, also should be removed. The ovary was selectively removed according to the patient’s clinical conditions and willingness. The resection of lymph nodes involved the obturator lymph nodes and the internal, common, and external iliac vessel lymph nodes. Para-aortic nodes should be removed only in the following cases: (1) The preoperative PET/CT showed that there were suspicious metastatic lymph nodes in the para-aortic area [SUVmax of lymph nodes is higher than the background metabolism level (23, 24)]; and (2) the possibility of metastasis in the pelvic or para-aortic lymph nodes was suspected intra-operatively.

PFS was defined as the time interval between the date of surgery and the date at which the first recurrence of disease was confirmed. Starting from the completion of the surgery, follow-up examinations (chest CT, abdominal CT and pelvic MR) were performed approximately every 3 months for the first 2 years, and then every 6 months or 1 year for the next 3 years. Recurrent disease was defined as local recurrence, metastasis of pelvic or para-aortic lymph nodes, and metastases in distant organs following surgery. For patients with abnormal follow-up examinations results, PET/CT or biopsy was recommended, and the results used as the standard for recurrence. The end of the follow-up time is December 31^th, 2020.

A pathologist with more than 10 years of experience evaluated sections using HE staining and immunohistochemistry (IHC) of CD31 and D2-40. The following results based on postoperative pathology report were obtained: tumor differentiation grade, stage, lymph node metastasis, LVSI, cervical stromal invasion depth and histologic tumor type.

Tumor sections were obtained from our hospital pathology department and we purchased antibodies from Abcam (Shanghai, China). In addition, Cox-2, E-cadherin, Tenascin-C, Glut-1, Tigar, and Vegf-A protein expression were studied based on IHC, which was performed using the Leica BOND-MAX system (Leica Biosystems, Shanghai, China). Thorough mixing was done with the following polyclonal antibodies: Cox-2 at 1:100 dilution; E-cad at 1:700 dilution; Tenascin-C at 1:100 dilution; Glut-1 at 1:500 dilution; Tigar at 1:300 dilution; and Vegf-A at 1:50 dilution. Then the tissue section was sealed after the Imaging Mass Cytometry (IMC) was finished, the section was placed under the lens of a scanner (Pannoramic MIDI, 3D Histech), and moved gradually, imaging while moving and then scanning and imaging all the tissue information on the tissue section to form a file. After image scanning was completed, the DensitoQuant software application in the QuantCenter (QuantCenter is a piece of analysis software for Pannoramic viewer) automatically recognized and set all dark browns on the tissue section as strongly positive, brown-yellow as moderately positive, light yellow as weakly positive, and blue cell nuclei as negative. Furthermore, for each tissue, strongly positive, moderately positive, weakly positive and negative areas, and the percentage of positive areas, were identified. A negative expression was defined by a < 10% positive expression.

For HPV detection we used an AB I2Prism7000 PCR detector, a Hybrid-Max medical nucleic acid molecular hybridizer (AB, USA), and a PCR kit (Qiagen, USA). HPV detection used an AB I2Prism7000 PCR detector, a Hybrid-Max medical nucleic acid molecular hybridizer (AB, USA), and a PCR kit (Qiagen, USA). Fifteen types of the HR-HPV types (HPV16, 18, 31, 33, 34, 39, 45, 51, 52, 53, 56, 58, 59, 66, and 68) were detected and diagnosed by a pathologist with more than 10 years of diagnostic experience.

Included in the novel systemic inflammation index were the following: NLR, defined as neutrophil-to-lymphocyte ratio; LMR, defined as mononuclear cell-to-lymphocyte ratio; and SII, defined as ratio of product of platelets and neutrophils to lymphocytes (25, 26). Evaluation was undertaken of the relationship between the novel systemic inflammation index and prognosis based on the ROC curve and calculation of the optimal cut-off value for meaningful indexes. Anemia was defined as a hemoglobin count below 11.0 g/dL. The optimal D-dimer, Fg and SCCA cut-off range for assessing prognosis was 0-0.5 mg/L, 2-4 g/L, and 0-2.5 ng/ml. These cut-off ranges were selected as they have been recognized as standard pathological definitions.

All patients were fasted for more than 6 hours before the 18F-FDG PET/CT examination, blood glucose levels were controlled below 7 mmol/L, and 18F-FDG was injected with the patient in a quiet state from 3.70 to 5.55 MBq/kg. After 45–60 minutes, 18F-FDG PET/CT was performed. The patient was placed in a GE Discovery Elite scanner (GE Healthcare, USA) with the scan ranging from the skull top to the middle of the thigh (120s/bed). The thickness of the CT scan layer was 3.75 mm, the tube voltage was 120–140 keV, and the tube current was 80mA. PET images were reconstructed using an Iterative adaptive algorithm.

Transmitting all ¹⁸F-FDG PET/CT images to GE AW4.6 workstation (GE Healthcare, USA), 2 radiologists with more than 5 years of radiodiagnostic experience used PET volume computer-assisted reading (PET VCAR) software to perform imaging analysis. After outline of a lesion region of interest (ROI) at the cross-sectional level of the largest area of the tumor, the fused-PET software automatically calculated SUV_mean, SUV_max, MTV and TLG of the entire tumor. MTV was calculated by a 40% SUV_max threshold (27, 28).

The patients were divided into a positive and a negative group according to prognosis. Differences in patient characteristics between the groups were compared with the t-test (Mann-Whitney U test, if not normal distribution) or chi-squared (Fisher’s exact test, if not the number of assumptions necessary). For quantitative variables, receiver operating characteristic (ROC) curves were generated to assess the area under the ROC curve (AUC) for differentiating prognosis. We used the cut-off threshold values for differentiating these meaningful quantitative variables. The Kaplan-Meier method was used to build survival functions and the log-rank test was used to compare survivals for all variables. Prognostic variables were evaluated by univariate and multivariate Cox proportional hazard model (variables with p<0.1 in the univariate Cox proportional hazard model were used in the multivariate analysis). In multiple testing, a correction was performed, based on the Benjamini-Hochberg procedure. Finally, these final prognostic variables were incorporated to construct the nomogram. The nomogram adopted the 1-, 2-, and 5-year PFS as primary endpoints. Validation of the nomogram was performed with 1000 bootstrapped samples, that is, we adopted the method of bootstrapped samples to set up a control group, randomly selected 1000 times from the existing samples, arbitrarily once, and then a final time to get the original sample number (95). The predictive capability was evaluated by ROC analysis. Generated calibration curves were used to visualize the difference between the predicted and actual 1-, 2-, and 5-year PFS. Decision curve analysis (DCA) was introduced to evaluate clinical utility of the nomogram (29). Based on the median of the total scores, a risk stratification system was developed, and the cervical cancer patients were divided into two risk subgroups, including high-risk and low-risk groups, and the Kaplan-Meier method and log rank test were used to compare the differences between the two subgroups. The statistical analysis was conducted by MedCalc (version 15.2.2), SPSS (version 22.0), R (version 3.5.3), and p<0.05 indicated a significant statistical difference (if not specified).

The median follow-up was 62.8 months (range, 2-96). A total of 95 patients were enrolled: 24 (25.3%) patients manifested progressive disease and 71 (74.7%) patients presented no evidence of disease ( Figure 1 ).

The clinicopathological characteristics of the patients are summarized in Table 1A . Cervical stromal invasion depth (chi-squared test, p=0.046; log-rank test, p=0.032), LVSI (chi-squared test, p=0.032; log-rank test, p=0.008) and HPV-16 (chi-squared test, p=0.001; log-rank test, p<0.001) showed differences between patients with and without disease progression. The remaining indicators had no significant correlation.

Chi-squared test and Kaplan-Meier method showed that patients with Glut1-negative had a statistically significant better clinical outcome than patients with Glut-1-positive (chi-squared test, p=0.039; log-rank test, p=0.039) ( Table 1B ). The remaining protein indicators were not statistically significant.

Chi-squared test and Kaplan-Meier method showed only D-dimer was related to disease progression and PFS ( Table 1C ). The ROC curve showed that no novel systemic inflammation index was statistically significant for disease progression.

SUV_max (U test, p=0.014) and SUV_peak (U test, p=0.002) showed significant between-group differences ( Table 2 ). ROC analysis showed that SUV_max (AUC 0.668, p=0.006), SUV_peak (AUC 0.716, p<0.001) had a positive effect on predicting disease progression ( Figure 2 ). The optimal cut-off threshold values for SUV_max, SUV_peak were 9.12 g/cm³ (sensitivity 91.67, specificity 39.44) and 7.63 g/cm³ (sensitivity 91.67, specificity 47.89), respectively. Kaplan-Meier method showed the DFS rates of patients exhibiting high SUV_max and SUV_peak of the primary tumor were significantly lower than those of patients exhibiting low SUV_max and SUV_peak of the primary tumor (p=0.006 and p=0.001, respectively) ( Figure 2 ).

Univariate Cox proportional hazard model showed that cervical stromal invasion depth (p=0.041), LVSI (p=0.018), HPV-16 (p=0.001), Glut1 (p=0.048), D-dimer (p=0.040), SUV_max (p=0.016) and SUV_peak (p=0.005) were associated with PFS. Multivariate Cox proportional hazard model showed that SUV_peak (p=0.012), and HPV-16 (p=0.007) were independent risk factors for PFS ( Table 3 ).

SUV_peak and HPV-16 were incorporated to develop the nomogram for predicting 1-, 2-, and 5-year PFS. The nomogram showed that SUV_peak made the largest contribution to the prognosis, followed by HPV-16 which showed a certain amount of impact on the PFS ( Figure 3 ). ROC analysis showed the AUCs at 1-, 2-, and 5-year PFS reached 0.828, 0.808 and 0.814 in the original model, and 0.814, 0.835 and 0.841 in the validation model, respectively ( Figures 4B, D ). Whether in the original model or in the validation model, the calibration curves demonstrated considerable agreement between the nomogram and predicted survival ( Figures 4A, C ). Clinical utility of the nomogram was evaluated by DCA. The nomogram showed enormous positive net benefits across wide ranges of mortality risk in both models, demonstrating its predominant clinical utility in predicting PFS ( Figure 5 ). In particular, it has the greatest clinical utility for 5-year PFS in both models. In addition, we calculated the total scores for the original model to build a risk stratification system based on our nomogram, and then distinguished the patients according to the median quantile of total scores into high-risk subgroups and low-risk subgroups. The PFS in the two subgroups was exactly separated by this system ( Figures 6A, B ).