Many testicular cancers contain both seminoma and non-seminoma cells. These mixed germ cell tumors are treated as non-seminomas. In this stage, the cancer has not spread outside the testicle, and your Because seminoma cells are very sensitive to radiation, low doses can. Patients with Stage 1 testicular cancer of non-seminoma type have a primary cancer that is limited to the testes and is curable in more than 95% of cases.
|Published (Last):||19 January 2018|
|PDF File Size:||7.86 Mb|
|ePub File Size:||1.79 Mb|
|Price:||Free* [*Free Regsitration Required]|
Testicular germ cell tumours seminomatous and nonseminomatous are testicu,ar most common cancers among young adult men. Owing to the seminomxtoso overall cure rate, interest has shifted from increasing the overall cure rate to reducing treatment-related toxicity for patients with a good prognosis de Wit et al On the other hand, high-risk patients, eligible for more intensive treatment, for example, stem-cell support or high-dose chemotherapy, testiculsr be identified Bokemeyer et al, Several classifications have been proposed in the past to distinguish seminonatoso according to prognosis, by identifying and combining the main prognostic factors for progression-free survival PFS and overall survival Bajorin et al, ; Mead et al The coexistence of classifications differing in type, complexity and ability to separate good from poor prognosis complicated international collaboration in randomised trials and made comparison of nonrandomised studies impossible.
For the IGCC classification, readily available risk factors were selected from a wider set following Cox regression analyses, namely primary site, presence of nonpulmonary visceral metastases NPVM and elevation of carinoma tumour markers alpha-fetoprotein AFPhuman chorionic gonadotrophin HCG and lactic dehydrogenase LDH. All variables were categorical, since no major differences in performance were found compared to using continuous variables McCaffrey et al In Table 1how the risk factors were combined into three prognostic groups for patients with nonseminomatous germ cell tumours NSGCT with either good, intermediate or poor prognosis are shown.
The good prognosis group is characterised by the absence of adverse risk factors.
The intermediate prognosis group is defined by the presence of any intermediate tumour marker, that is, one or more intermediate tumour markers are present. The classification can be seen as a max function where the good, intermediate and poor prognosis groups carcioma a maximum score of zero, one or two, respectively. In the IGCC classification, all intermediate tumour markers and all poor risk factors were required only to be sufficiently bad to be classified as intermediate and poor prognosis, respectively, that is, differences in importance between intermediate tumour markers and differences in importance between poor risk factors are not taken into account.
Furthermore, no distinction is made between the number of intermediate tumour markers in the intermediate prognosis group and the number of poor risk factors in the poor prognosis group.
Better discrimination might be achieved by incorporating differences in predictive strength and testing specific interaction terms.
Furthermore, it is difficult to adjust the current classification for changes in treatment strategy. A more flexible scoring system could more easily identify subgroups for the identification of very high risk patients eligible for novel chemotherapy approaches such as high-dose chemotherapy or the use of novel cytotoxic agents Bokemeyer et al; Kollmannsberger et al We however note that an important consideration in developing the IGCC classification was that all the prognostic carconoma should be large srminomatoso to make randomised trials of new treatments for each prognostic group feasible IGCCCG, The aim of this study was to reconsider steps taken in the development of the IGCC classification, and to investigate alternative classifications based on Cox regression and recursive partitioning Breiman et althat ssminomatoso discriminate better and be more suitable to yesticular more subgroups.
All patients were treated between and with cisplatin-based chemotherapy. Data were collected on age, primary site, date of diagnosis, levels of serum AFP, HCG and LDH, nodal disease testiculag the abdomen, mediastinum, and neck, lung metastases, spread to other visceral sites like liver, bone and brain and on treatment details like previous therapy.
The outcome measures were PFS and overall survival from the start of the chemotherapy. Each tumour marker had three categories; good, intermediate and poor with specific cutoff points on the continuous tesitcular markers see Table 1 IGCCCG, The same risk factors and categories were used to construct the alternative classifications based on Cox regression and recursive partitioning.
The IGCC classification makes no clear distinction between the intermediate tumour markers and between the poor risk factors and is represented by a max score.
One way to assess this assumption is by evaluating whether the weights in the IGCC classification were optimally allocated to the risk factors. We used the Cox aeminomatoso to study the univariable and multivariable effects of the IGCC risk factors on the overall survival, expressed as Hazard ratios and testixular coefficients.
We multiplied the multivariate regression coefficients by 10 and rounded them to obtain weights. A sum score was calculated by multiplying the weights with individual patient characteristics and adding the resulting individual scores Assmann et al We calculated the estimated 5-year survival rate for each score.
Tesficular IGCC classification can be viewed as implying that the risk factors are strongly dependent, that is, that there are interactions between risk factors.
There is, for example, no distinction made between patients with one poor risk factor or three poor risk factors. To test whether and which interactions were present, we added all two-way interactions between the IGCC risk factors in a Cox regression model. Since interactions based on small number of patients give unreliable regression coefficients, the interaction terms were defined as linear.
A sum score based on a regression model with interactions is, however, more difficult to calculate and interpret.
Therefore, a table was constructed with 5-year survival estimates for all possible combinations of the IGCC risk factors based on the Cox regression model with linear interactions. The number of patients on which each survival estimate was based is given to indicate the reliability of the survival semihomatoso. A binary tree is built by the following process: Splitting continues until the on reach a minimum size or until no improvement can be made stopping rule.
Testjcular full tree, which is often too complex and overfit, is pruned using crossvalidation.
Stage I Non-Seminoma Testicular Cancer | Texas Oncology
All trees within one standard error of the lowest crossvalidated prediction error are considered as equivalent. From these equivalent seminomatosk, the simplest is chosen as final tree Breiman et alsemonomatoso As a splitting method, the exponential scaling method was used Therneau et al; LeBlanc and Crowley, The splitting process stopped when a minimum of five patients per groups was reached or when there was no further decrease in prediction error.
We used fold crossvalidation to determine the optimal tree size. Modelling was performed with S-plus version using the RPART library that contains a recursive partitioning method for survival data. In all classifications, three prognostic groups were identified using the estimated 5-year survival by sum score caricnoma 5Rcombination of risk factors 5Ri or binary tree 5T.
Furthermore for each classification, we explored the possibility of identifying more subgroups. For the IGCC classification, this was carried out by allowing weights to vary from zero to four instead of zero to twoand comparing all tesgicular combinations on performance.
Prediction of metastatic status in non-seminomatous testicular cancer.
For classifications 5R, 5Ri and 5T, we changed the cutoff points on estimated 5-year survival. The classifications were evaluated by their ability to distinguish between patients seminomatiso in survival. An indication of the discriminative ability is the difference in 5-year survival rates between the good, intermediate and poor prognosis groups.
A c -statistic was also calculated for both the three and five group classifications. For binary outcomes, the c -statistic is similar to the area under the ROC curve Harrell et al The c -statistic for survival data indicates the probability that for a randomly chosen pair czrcinoma patients, the one having the higher predicted survival is the one who survives longer Harrell et al When a model is developed and evaluated on the same data, the performance of the model is usually too optimistic.
The optimism can be quantified with statistical methods, known as internal validation techniques Steyerberg et al To estimate and correct for the optimism in discriminative ability, the steps taken in the Cox regression and recursive partitioning were internally validated by taking random bootstrap samples Efron and Tibshirani, ; Harrell et al The median follow-up time of surviving patients was 50 months.
Disease progression occurred in patients, and patients died. All risk factors were predictors of survival as indicated by the Hazard ratios ranging from 2. The regression-based weights of the risk factors in classification 5R, and the cutoff points on the semiomatoso sum score are presented in Table 3 testiculat, with the weights and cutoff points of the IGCC classification.
Weights, coding of variables, and cutoff on the max function of the IGCC classification and the sum score of the regression-based classification 5R. The weights suggest that differences between risk factors were present. As a result, a poor AFP level score 3 is not sufficient to be classified as poor prognosis np classification 5R.
Also, the combination of two or three intermediate tumour markers, which would lead to an intermediate prognosis in the IGCC classification, results in a score of over 10 and thus in classification in the poor prognosis group in classification 5R.
The regression coefficients all had negative signs, indicating that the effect of the risk factors together was smaller than the sum of their separate effects. For all combinations of the IGCC risk factors, we present 5-year survival estimates from the Cox regression model with interactions Appendix.
Since the number of patients with more than one poor risk factor was limited, the survival estimates for these patients were less reliable. The final tree fitted by recursive partitioning, using the exponential scaling method. The 5-year survival rates, events and total number of observations per subgroup are given.
The resulting subgroups are displayed in rectangulars and determine classification 5T. The 5-year survival rates for the good, intermediate and poor prognosis groups were comparable for the IGCC classification and classifications 5R, 5Ri and 5T Table 4. The c -statistic of the IGCC classification was 0.
Prediction of metastatic status in non-seminomatous testicular cancer.
The apparent c -statistics of classifications 5R, 5Ri and 5T were 0. Validation showed minor optimism in the c -statistic in the Cox regression models 0. More optimism was present in the classification 5T, with the c -statistic decreasing from 0. Survival of the IGCC classification, the regression-based classifications 5R and 5Ri and classification 5T based on recursive partitioning. The cutoff points on the sum score for the five groups of classification 5R are also given in Table 5.
The difference in survival between the prognostic groups for each classification is illustrated in Figure 2. The c -statistic for the five groups of the IGCC classification and classifications 5R and 5Ri was slightly higher than for the three group classifications 0. The increase of the c -statistic for the five groups of classification 5T was very limited 0.
Survival of subgroups within the IGCC classification, the regression-based classifications 5R and 5Ri and classification 5T based on recursive partitioning. The discriminative ability of classifications derived through Cox regression and recursive partitioning was in concordance with the IGCC classification and therefore supports the validity of the IGCC classification.
We did, however, find that not all intermediate tumour markers and poor risk factors were equally important, and that taking these differences into account does affect the classification of patients. That AFP is of less importance than the other risk factors is confirmed by recursive partitioning where AFP was not selected in the final tree.
Furthermore, not all risk factors had statistical interactions. In classifications 5Ri and 5T, only a limited number of interactions were included.
Combining several risk factors led to differences in 5-year survival, that is, patients with one poor risk factor had a better chance of survival than patients with three risk factors. These deviations from the weights used by the IGCC classification did, however, not lead to improvements in discriminative ability, in contrast with what we expected.
It appears that the maximum discriminative ability might have been reached with the current IGCC risk factors and coding, making further improvement in discriminative ability difficult. The risk factors selected for the IGCC classification are in agreement with risk factors used in other studies on identifying good and poor prognosis patients with NSGCT Bajorin et al; Mead et al Some other potentially useful risk factors include age, lung metastases and abdominal mass size.
However, adding these three risk factors to the Cox model had no substantial effect on discriminative ability c increased from 0. One could also consider using continuous codings of tumour markers, but this would lead to an undesirable increase in complexity and decrease in applicability. The division into more prognostic groups is similar to another division by recursive partitioning of poor prognosis patients Kollmannsberger et al Kollmannsberger et al identified three prognosis groups: These survival rates are higher than the survival rates of the good-poor, intermediate-poor and poor-poor risk groups identified in the IGCC dataset.
The data in Kollmannsberger et al are more recent and improvements in treatment may have led to the difference in survival. The lack of improvement in discriminative ability in both the classifications with three and five groups might also be explained by the dominance of the good prognosis group, which has a similar survival for all classifications and contains more than half of all patients.