The Establishment And Evaluation Of Models To Predict The Bone Metastases Of Patients With Newly Diagnosed Prostate Cancer In China

Part I The Logistic regression analysis of bone metastases of newly diagnosed prostate cancerPurpose:To identify the independent predictors of bone metastases of newly diagnosed prostate in China according to the Logistic regression analysis of the clinicopathological factors of newly diagnosed prostate cancer patients with bone metastases.Methods:There were a total of501patients with newly diagnosed prostate cancer hospitalized in our department from March2005to March2011. The clinicopathological variables of these patients, including year at diagnosis, age, with SRE or not, ALP, PSA, cT, cN and GS, were retrieved respectively from the medical records. The study sample was divided into two groups according to with bone metastases or not, and then the univariate analysis was utilized to explore the distributive differences of factors between the two groups and the Logistic regression analysis used to identify the independent predictors of bone metastases. Results:The incidence of bone metastases in newly diagnosed prostate cancer patients showed a downward trend, and the incidence of bone metastases was27.5%(138/501) in this study. The PPV and NPV of SRE used as a predictive criterion for the bone metastasis were100%and78.7%respectively, and the SRE occurrence rate was8.0%(40/51). When using ALP^143IU/L as a predictive criterion to predict the bone metastasis, we found that, the PPV and NPV were100%and80.3%respectively. The univariate analysis results showed that the distribution of each variable between the two groups wasn’t consistent (p<0.05). The Logistic regression model with a C-index of0.842contained four factors of which PSA and cT were independent prognostic factors for bone metastases of newly diagnosed prostate cancer (p<0.05). The ORs of PSA, cT4, GS=8-9and GS=10were1.02,14.28,1.68and3.46respectively.Conclusion:The factors of GS≥8, PSA and cT were independent predictors of bone metastases of newly diagnosed prostate cancer, and the highest OR were acquired by cT4factor. Part Ⅱ The establishment and evaluation of Fudan CART modelPurpose:To establish and validate a CART model which can be used to predict the bone metastasis risks of newly diagnosed prostate cancer in China, and compare the CART model with other models to find out the best model based on which to reduce the unnecessary bone scans.Methods:Fudan CART model was established using the CART statistical method to predict the risks of bone metastases based on the501cases in this study, and then validated internally by10-fold cross validation method to reduce the over-fit bias. Briganti’s CART model was verified in the501Chinese patients. Fudan CART model, Briganti’s CART model and only using SRE or ALP≥143IU/L (SRE model or ALP model) as the criterion to predict the bone metastases were compared on the predictive accuracy (by AUC) and clinical application values (by missed diagnosis and misdiagnosis rates).Results:The predictive risks of bone metastases in the5nodes of Fudan CART model, with a C-index of0.809after the internal validation showing a good predictive accuracy, were7.4%,21.8%,36.8%,58.1%and76.5%from low to high respectively. After being verified in the study sample, the lowest and highest predictive risks of Briganti’s CART model were3.9%and43.1%respectively. The AUC values of Fudan CART model, Briganti’s CART model, SRE model and ALP model were0.813,0.691,0.645and0.678respectively, and a significant difference was found between the former and the others on AUC values. When the Pt values ranged from24.2%to36.8%, Fudan CART model obtained lower missed diagnosis and misdiagnosis rates compared to Briganti’s CART model.Conclusion:Compared to other models, Fudan CART model had higher accuracy and clinical application value when using to predict the bone metastases of newly diagnosed prostate cancer. Part III The establishment and evaluation of a nomogramPurpose:To establish and validate a nomogram used to individually predict the risks of bone metastasis with newly diagnosed prostate cancer in China, and compare the nomogram with other models on the predictive accuracy and clinical application value to find out the best model based on which to reduce the unnecessary bone scans..Methods:The nomogram was established based on the four predictive factors and their coefficients in the Logistic regression model, validated internally by bootstrap resample validation method to reduce the over-fit bias and then calibrated by calibration plot. The nomogram, Fudan CART model and only using ALPP≥143IU/L (ALP model) as the criterion to predict the risks of bone metastasis were compared on the clinical application values by DCA method.Results:The C-index of the nomogram was0.833after internal validation, showing a higher predictive accuracy than Fudan CART model. The calibration plot showed that the nomogram had a good predictive calibration. Compared to other models, the nomogram gained a higher clinical application value when Pt ranged from3.4%to47.7%, except that Fudan CART model was rather better than the nomogram in the Pt range of10.1%to14.7%.Conclusion:The nomogram had the highest predictive accuracy compared to other models, and when the Pt ranged from3.4%to47.7%the nomogram and Fudan CART model both had good clinical application values.