Development and validation of nomogram based on a novel platelet index score to predict prognosis in patients with renal cell carcinoma

Purpose: This study aims to develop and validate a nomogram based on a novel platelet index score (PIS) to predict prognosis in patients with renal cell carcinoma (RCC). Patients and methods: We retrospectively analyzed the data of 759 consecutive patients with RCC. The Kaplan-Meier curves were performed to analyze the platelet parameters and PIS was established. The patients were randomly divided into training (N=456, 60%) and validation cohorts (N=303, 40%). The nomogram was created based on the factors determined by multivariable Cox proportional hazard regression of the training cohort. We assessed the discrimination and calibration of our nomogram in both training and validation cohorts. And then the nomogram was compared with other reported models. Results: High platelet count (PLT>285×109/L) and low platelet distribution width (PDW≤10.95fL) were associated with shorter progression-free survival (PFS). Thus, PLT and PDW were incorporated in a novel score system called PIS. On multivariable analysis of training cohort, PIS, American Joint Committee on Cancer (AJCC) stage, and sarcomatoid differentiation were independent prognostic factors, which were all selected into the nomogram. The nomogram exhibited good discrimination in both training (C-index: 0.835) and validation cohorts (C-index: 0.883). The calibration curves also showed good agreement between prediction and observation in both cohorts. The C-index of the nomogram (C-index: 0.810~0.902) for predicting 2-year, 3-year, and 4-year PFS were significantly higher than Leibovich (C-index: 0.772~0.813), SSIGN (C-index: 0.775~0.876), Cindolo (C-index: 0.642~0.798), Yaycioglu (C-index: 0.648~0.804), MSKCC (C-index: 0.761~0.862), Karakiewicz (C-index: 0.747~0.851), and AJCC stage models (C-index: 0.759~0.864). Conclusion: The nomogram based on a novel PIS could offer better risk stratification in patients with RCC.


Introduction
Renal cell carcinoma (RCC) is the third most common genitourinary worldwide, accounting for 3%~5% of the malignancies in adults [1]. Localize RCC can be cured by surgery, with a favorable 5-year overall survival of 71%~88% [2]. However, even with a relatively good prognosis, approximately 28% of patients will experience recurrence after curative surgery [3]. Nowadays several prognostic models like SSIGN, Leibovich, Cindolo, Yaycioglu, MSKCC, and Karakiewicz models have been used in predicting survival for patients with RCC. The reported concordance index (C-index) of the above models were range from 0.67 to 0.84 [3][4][5][6][7][8]. However, in a prospective external validation cohort, the predictive ability was decreased with a C-index range from 0.587 to 0.69 [9]. Thus, it is still worthy for us to further Ivyspring International Publisher explore the potential prognostic factors and develop more accurate predictive models for RCC patients.
Platelets have been recognized as the key factor in hemostasis and wound healing. Recent evidence demonstrated that cancer cells can activate platelets and induce aggregation, which in turn contribute to cancer genesis and metastasis [10]. Platelet count (PLT) is the most notable index that can be affected by the equilibrium between production and consumption. High PLT has been reported as an unfavorable prognostic factor in many malignancies including RCC [11][12][13][14]. Recently, mean platelet volume (MPV) and platelet distribution width (PDW), which reflect the size and uniformity of platelets, have shown their prognostic value in thyroid, lung, gastric, and breast cancers [15][16][17][18].
To our knowledge, there are few studies that explore the prognostic value of MPV and PDW in RCC populations. Besides, previous models contain similar variables like ECOG-PS, TNM stage, Fuhrman grade, symptom, necrosis, tumor size, et al. [3][4][5][6][7][8]. However, none of these prognostic models have included any platelet parameters, except the International Metastatic Renal Cell Carcinoma Database Consortium model (IMDC) which was included PLT to predict survival of metastatic RCC [19]. Thus, our study aims were as follows: (1) To explore the prognostic value of PLT, MPV, and PDW in RCC patients and develop a novel platelet index score (PIS); (2) To develop and validate nomogram based on the PIS in our training and validation cohorts; (3) To compare the predictive ability of our nomogram with other reported prognostic models.

Patients selection
The clinicopathologic data of renal mass admitted to Peking University Third Hospital from January 2015 to December 2017 were retrospectively analyzed. Inclusion criteria were patients who underwent nephrectomy in our department and pathologically confirmed RCC. Exclusion criteria were as follows: (1) recurrent or bilateral RCC; (2) combined with other malignancies; (3) those with the hematological disease, inflammatory disease, autoimmune disease; (4) those who underwent splenectomy in the past history; (5) those using antiplatelet drugs withing a week before blood collection; (6) missing platelet data. The flow chart was shown in Figure 1. Finally, the eligible study cohort consisted of 759 consecutive RCC patients were further analyzed.

Data collection
All patients underwent a blood tests and image examinations. Blood collection was performed within a week before surgery. The whole blood sample was collected in EDTA-containing tubes and processed within an hour after collection. All patients underwent chest X-ray, B-mode ultrasonography, abdominal computed tomography (CT), and/or magnetic resonance imaging (MRI) before surgery. Position emission CT, chest CT, cranial MRI, and bone scans were performed when suspicious distant metastasis. Surgical type and approach were determined by the surgeon' preference based on the current guidelines. All tumor specimen was evaluated by two experienced pathologists. Patients with distant metastasis were treated using tyrosine kinase (TKI) inhibitors as the first-line adjuvant target therapy.
Clinicopathologic variables were collected, including age, sex, body mass index (BMI), symptom at presentation, comorbidities, surgical type, surgical approach, surgical time, interoperative blood loss, tumor size, tumor side, tumor stage, histologic subtype, nuclear grade, necrosis, sarcomatoid and rheumatoid differentiation as well as platelet parameters including PLT, MPV, and PDW through electronic medical records of Peking University Third Hospital.
Surgical type, approach, surgical time, and interoperative blood loss were determined based on surgical records. Tumor size, tumor side were evaluated according to the preoperative CT/MRI image report. The tumor stage was evaluated according to both preoperative image and postoperative pathology report, and defined based on the 2016 WHO TNM classification [20]. Nuclear grades were defined according to the 2016 WHO/ISUP grading system [20].

Follow up
Patients were recommended to follow up regularly at our institution for every 3 months during the first year, every 6 months for the following year, and then annually. Blood tests and chest x-ray were performed at each follow-up while abdominal CT scan was done annually. Follow-up data were collected through medical records or telephone calls. Disease progression was defined as any evidence of recurrence or metastasis, or tumor progress in an already existed metastatic site. Progression-free survival (PFS) was defined as the time from the date of surgery to disease progression.

Statistical analysis
Normally distributed continuous variable was reported as mean ± standard deviations. Non-normally distributed continuous variables were reported as medians and interquartile range (IQR). Categorical variables using the frequency count and percentage. The analysis of variance (ANOVA) test or Kruskal-Wallis test were applied to compare continuous variables. The Chi-square test or Fisher exact test was applied to compare categorical variables. The receiver operating characteristic curve (ROC) was used to determine the optimal cut-off value of PLT, MPV, PDW according to survival outcome. The Kaplan-Meier method was performed to analyze the platelet parameters, and the difference between groups was compared using the log-rank test.
To develop a reliable and well-calibrated nomogram, patients were randomly divided into training cohorts (N=456, 60%) and validation cohorts (N=303, 40%). Univariable and multivariable Cox proportional hazard regression were performed to identify the independent prognostic factors in the training cohort. Selected independent factors were incorporated in the nomogram to predict the probability of 2-year, 3-year, and 4-year PFS. The discrimination of the nomogram was measured by C-index, which ranges from 0.5 (no predictive power) to 1 (perfect prediction) [21]. Calibration was evaluated using calibration curves, which assessed between the observed outcome probabilities and the nomogram predicted probabilities with a bootstrapped resample of 1000 times. The total points of each patient in the validation cohort were calculated according to the nomogram, then Cox regression was performed using the total points as a factor, and finally, the C-index and calibration curve was derived based on the regression analysis. Time-dependent ROC curves and areas under the curves (AUCs) at 2-year, 3-year, and 4-year were generated to assess prognostic accuracy of the nomogram in training and validation cohorts [22]. Improvement in the predictive accuracy was assessed by calculating the integrated discrimination improvement (IDI) and the net reclassification improvement (NRI) [23]. Higher IDI and NRI indicated greater risk discrimination and improved classification. Furthermore, the linear predictor was calculated as the sum of the product of regression coefficients and variables reported by other models, and C-index was calculated according to the linear predictor and then compared with our nomogram. Nomogram establishment and calibration were conducted by R software (Version 4.0.3) using the "rms" package, and other statistical analysis was performed with SPSS (Version 26, IBM, Armonk, NY) and GraphPad Prism (Version 8, GraphPad Software Inc., San Diego, CA, USA). All tests were on two sided, and P<0.05 was considered statistically significant.

Nomogram development and validation
On univariable and multivariable analysis in the training cohort, higher PIS (HR: 2.098 for score 1, 2.775 for score 2, P<0.001), higher AJCC stage (HR: 2.746 for II, 5.352 for III, 13.662 for IV, P<0.001), and sarcomatoid differentiation (HR: 2.032, P=0.030) were independent factors predict shorter PFS (Table 3). Thus, we established the nomogram by incorporating all the independent factors to predict 2-year, 3-year, and 4-year PFS (Fig. 3). The C-index was 0.835 for the training cohort and the calibration curves for the probability of 2-year, 3-year, and 4-year PFS showed an optimal agreement with the prediction by nomogram and actual observation (Fig. 4). In the validation cohort, the C-index was 0.883 and calibration curves also showed good agreement between prediction and observation (Fig. 4). To further compare the discrimination of the nomogram in training and validation cohorts, the 2-year, 3-year, and 4-year of time-dependent ROC curves were showed in Figure 5, which suggested good discrimination in both training and validation cohorts.

Discussion
The ability to predict oncologic outcomes in RCC patients is essentially significant to clinicians. Platelets parameters have been explored as a promising biomarker in predicting oncologic outcomes in cancers. In this study we analyzed a consecutive 759 RCC patients and reported several noteworthy findings. Firstly, high PLT and low PDW were associated with shorter PFS in our study group. Then we established a novel score system named PIS by combining PLT and PDW, which have shown their significant correlation with disease progression. Secondly, we created a nomogram based on the PIS and other independent factors to predict PFS, which exhibited good discrimination and calibration in the training and validation cohort. Thirdly, our nomogram displayed better accuracy in predicting 2-year, 3-year, and 4-year PFS than other reported models and conventional AJCC staging system.
Notably, several studies have explored that high PLT was related to worse prognosis in RCC populations, which consistent with our findings [14,24]. Recent evidence also showed increased PLT had a lower response rate to tyrosine kinase inhibitors therapy and shorter overall survival in metastatic RCC [25]. Thus, PLT was included in the IMDC model to predict prognosis for metastatic RCC patients [19]. However, the conflicting results were also reported that PLT was not an independent factor predicting prognosis after adjusting some known pathologic factors, which suggested we should focus more on other platelet parameters that can better reflect the function rather than platelet count alone [26,27].
Preclinical studies have found cancer cells can activate platelet through a wide variety of crosstalk, eventually caused tumor genesis and metastasis [8]. MPV and PDW can serve as indicators reflecting the activation of platelets [28,29]. Clinical studies have found a strong correlation with diagnosis and prognosis in several malignancies [15][16][17][18]. Seles M et al. [30] have found MPV represented a highly significant predictor of recurrence and cancer-specific death in patients with RCC. Although our study failed to show the prognostic value of MPV, the PDW was another novel biomarker in predicting prognosis for RCC patients, which has not been investigated before. Several studies have shown high PDW was associated with poor survival in hepatocellular, breast cancers as well as skull base chordoma [31][32][33]. However, low PDW has also been reported that correlated with worse survival in endometrial, esophageal, gastric cancers [34][35][36], which consistent with our study result that low PDW was related to shorter PFS in RCC patients. The reason for the conflicting results was unclear. We speculate that within the tumor microenvironment, tumor cells can trigger activated platelet release granules that contain platelet-derived growth factor (PDGF), vascular endothelial growth factor (VEGF), and transforming growth factor β (TGF-β), which lead to decreased in size and increased heterogeneity of platelet at an early stage of tumor [10]. But as tumor growth and tumor burden increased, more and more platelets were affected by cancer cells, and eventually, the heterogeneity was decreased. However, the underline mechanism remains to be elucidated.
Furthermore, we created a novel score system called PIS by combining PLT and PDW and found that high PIS correlated with shorter PFS in RCC. Chen H et al. [36] was the first to established a similar platelet index score by combining PLT, MPV, and PDW, which showed its strong correlation with recurrent-free survival and overall survival in endometrial cancer. It is worth noting that in RCC patients, conventional prognostic models, except the IMDC model, were not included any platelet parameters [3][4][5][6][7][8]19]. Since our study result has shown the prognostic value of PIS, we created a nomogram based on PIS to predict PFS in patients with RCC. The nomogram performed well in the prediction of RCC patients by its reported C-index (0.835 and 0.883 for the training and validation cohort) and calibration curve. Although the external validation of other models has also shown its good discrimination with C-index range from 0.642 to 0.895, the nomogram is more accurate to predict PFS than the models like Leibovich, SSIGN, Cindolo, MSKCC, Karakiewicz, and Yaycioglu models as well as AJCC staging system. Besides, our nomogram only included three features, which was simpler than other models. More importantly, our findings not only raise significant concerns regarding the prognostic value of preoperative platelet parameters in patients with RCC but also predict prognosis for patients with RCC in a more accurate way by combining platelet parameters with conventional prognostic factors.
Several limitations may apply to our findings. We cannot eliminate the inherent bias owing to the retrospective nature of our study. As we know the preoperative drugs, inflammation, and hematologic disease may affect platelet function and parameters. But we could only achieve this information from the electronic medical records, which may not reflect the real state of an individual. Besides, although our study demonstrated that the PDW and PLT were associated with prognosis, the cut-off value of these parameters varied between different studies. Finally, the nomogram was only developed and validated in a retrospective single-center population. Thus, future studies are needed to externally validate the proposed nomogram in a multicenter, prospective, and large cohort.

Conclusion
In conclusion, we found high PLT, low PDW, and high PIS were associated with shorter PFS in patients with RCC. Then we established a nomogram based on the PIS for predicting PFS in patients with RCC. The nomogram could offer better risk stratification than other reported prognostic models in patients with RCC.

Ethical approval
The study was conducted in accordance with the Declaration of Helsinki and was approved by the Peking University Third Hospital Medical Science Research Ethics Committee. Because of the retrospective nature of the study, patient consent for inclusion was waived.