| [Objective]To assess the affection of the sample-size，censored value proportion to thepredictive model,and provide the theory basis and the practical operation method forclinicians to determine possible poor prognosis of patients with nasopharyngealcarcinoma (NPC), to guide the implementation of the individual comprehensivetreatment.[Methods]Patients who were primarily diagnosed as nasopharyngeal carcinoma wereenrolled in this study in the Fujian Provincial Cancer Hospital from June，2005to July，2007. According to the patient’s medical records and the investigation, the baselinedata, disease and treatment information was collected, and then all patients werefollowed up. The Kaplan-meier method was used to calculate the survival rates ofpatients. The Cox proportional risk model was applied for screening on independentfactors which had impact on prognosis. According to the selected factors andregression coefficients, the individual predictive index was calculated. The method ofBreslow was used for patient’s1-year,3-years and5-years survival rate. Thecomputer simulation of database with different sample-size and censored valueproportion was applied for assessing the affection of the sample size，censored valueproportion to the predictive model.[Results]1. A total of816patients meet the criterions, and the patient’s average survival timewas72.7months,5-years overall survival was76.7%.2. Gender, age, T stage, N stage, radiation technology are the independent prognosis factors of patients with nasopharyngeal carcinoma.3.1-year survival function of NPC patients wasS (12,X)0.997exp(PI),3-yearsurvival function of NPC patients wasS (36,X)0.980exp(PI),5-year survivalfunction of NPC patients wasS (60,X)0.966exp(PI).4. Three groups: low-risk group(PI<1.1), intermediate-risk group (1.1≤PI<1.8) andhigh-risk group(PI≥1.8).The averages of survival time were respectively81.28,76.27,67.27months,5years survival rates were respectively93.2%,84.3%,60.6%.5. Keeping the censored value proportion unchanged, different size samples wererandomly selected from all patients. The results showed that when sample size was atleast fifty times of the number of covariates to choose, changes of the number ofvariables and the regression coefficients contained in predictive index tended to bestable, and the predictive effect of COX model was relatively stable.6. Keeping the sample size changeless, the censored value proportion was changed,and the EPV was calculated. The results showed that when the censored valueproportion was90%, EPV <10, at the same time,the predicted survival and actualsurvival rate in patients were different (P <0.05). When the censored valueproportion was70%or80%, EPV>10, the predicted survival and actual survival ratein patients had no statistical significance (P>0.05).[Conclusions]1．The calculation formula of predictive index of patients with nasopharyngealcarcinoma is:PI0.517gender0.535age0.491Tstage0.400Nstage0.167radiationtechnology2．1-year,3-year,5-year survival predictive model based on the predictive index hadgood accuracy and applicability, realizing the survival prediction.3．The prognosis group based on the predictive index had higher accuracy inprognosis evaluation. In this study,the PI threshold value of best prognosis groupswas1.1and1.8.4．When sample size was at least fifty times of the number of covariates to choose, the predictive model had high stability and accuracy in predict the prognosis ofnasopharyngeal carcinoma (NPC).5．When the censored value proportion was not more than80%, the Cox model hadhigh stability and accuracy in predict the prognosis of nasopharyngeal carcinoma(NPC). |
PDF Full Text Request