SC < C!

c, < SC < c2

C2 <SC

PLTC = 0




PLTC = 1




No references were found that address this methodology, and the corresponding p-value adjustment formulae require further investigation.


4.1. Background and rationale for categorization

The majority of patients with the testicular cancer called seminoma who are treated initially with chemotherapy are found to have a residual tumour mass. The management of these patients is controversial, and may involve surgery, radiotherapy, or close observation. If surgery is performed, the nature of the residual mass is established, and if found to be cancerous, it is removed, rendering the patient disease-free. However, 80-85 per cent of residual masses contain only non-cancerous tissue which do not require further therapy.18 Therefore surgery is most justified for the subset of patients who are at greatest risk of having a cancerous residual mass. Identification of these high-risk patients using prognostic variables is essential to the treatment decisionmaking process.

The size of the residual mass on a post-chemotherapy computer tomography (CT) scan, measured as the largest diameter, is reported in the literature to predict for poor prognosis.19,20 Puc et al. found this to be the only significant variable predictive of outcome (logrank p-value = 0 03).21 They sought a CT diameter cutpoint to aid in identifying high-risk patients who were candidates for surgery. Their data set is used here to demonstrate how to find a cutpoint for the continuous CT diameter variable when the outcome variable is censored.

4.2. Patients

A total of 104 patients with advanced seminoma who were treated with various chemotherapy regimens at Memorial Sloan-Kettering Cancer Center from 1979 to 1992 were considered for this retrospective analysis; 55 patients had post-chemotherapy surgery, and 49 patients were closely observed using CT scans following their chemotherapy. The median follow-up time was 4 years.

Patients were designated as site failures or non-site failures. Site failure was defined as either the presence of cancerous tumour found at post-chemotherapy surgery, or clinical relapse at the assessed site during follow-up evaluation. Non-site failures had no cancerous tumour found at post-chemotherapy surgery, and no clinical relapse at the assessed site during follow-up evaluation. Failure-free survival time (time to site failure), calculated from the first day of chemotherapy treatment to the date of last follow-up evaluation, date of surgery, or date of relapse, was the

The frequency distribution of CT diameter is shown in Table I; the median CT diameter is 1-5 and ranges from 0 0 to 15 0. Figure 5 shows a scatter plot of the raw data, with time to site failure and non-site failure described by dots and stars, respectively. Most observed failure times occurred below a CT diameter of 5-0 cm. For patients with site failures, the range of CT diameters was 0 0-5-5, and for non-site failures the range was 0-0-15-0 (although only 4 per cent of the patients with non-site failures had CT diameters greater than 5-5). No CT diameter cut-off is suggested by these overlapping ranges. In fact, the general nature of the relationship between time to site-failure and CT diameter is not apparent from this scatter plot.

It was not possible to smooth these data by computing Kaplan-Meier median failure times for CT diameter groups, since an insufficient number of patients in the data set failed (10 out of 104). Therefore, a predictive failure time plot based on a Cox regression model was used to further investigate the functional relationship between CT diameter and failure-free survival time. The last column in Table I, labelled Ptime, lists the predicted failure-free survival times, which were

The time at which 90 per cent of the patients remained failure free was selected for prediction due to the high proportion of censoring in the data. A line showing the predicted failure-free survival times (Ptimes) for each CT diameter value is superimposed on the scatter plot in Figure 5. This clearly shows that a higher CT diameter is associated with a greater risk of failure. The plateau at 41-9 months occurs simply because that is the maximum observed failure-free survival time, which is the maximum predicted under a Cox regression model. The curve also shows a steady decline in predicted failure-free survival times over a CT diameter interval of 2-5-5-0.

Further examination of the predicted failure-free survival times (Ptimes) in Table I reveals a separation of outcomes around CT diameters of 2-6 and 3-0. Based on empirical evidence, Motzer et al. selected a 3-0 cutpoint, recommending that patients with tumours below that size be closely observed without surgery.20 As in case study 1, since the table and graphs show that monotonicity holds, a systematic minimum p-value analysis was appropriate, and was used to

Table I. Frequency distribution and predicted failure times for CT diameter
0 0

Post a comment