Statistical controversies in clinical research: should schedules of tumor size assessments be changed?

BACKGROUND: Time to progression (TTP) is often used as a primary end point in phase II clinical trials. Since the actual date of nadir and progression is never known, most calculated TTP are overestimated. This study evaluates the imprecision on the estimate of TTP under two hypothetical tumor kinetic settings and various assessment schedules. DESIGN: A two-component tumor growth model was used to account for treatment effect assuming exponential decay for tumor shrinkage and linear growth for progression. Evolution of tumor burden (TB) was modelized according to two scenarios using either a cytotoxic or a cytostatic agent and several assessment schedules. TB, nadir, progression and TTP were simulated for each visit schedule. RESULTS: For cytotoxic agents, our model predicted response at 1.5 weeks, a TB at nadir of 40.2 mm (starting from 100 mm) occurring at 6.7 weeks and true progression at 11.2 weeks with a TB of 48.2 mm. For cytostatic agents, our model predicted no response, a TB at nadir of 77 mm occurring at 9.2 weeks and true progression at 19.4 weeks with a TB of 92 mm. Depending on the assessment schedule, estimated TTP was increased from 0.8 to 36.8 weeks and from 0.6 to 28.6 weeks when compared with the true TTP and varied from 5.2% to 298% and from 1.66 to 109.58% when compared with the true TB at progression for cytotoxic and cytostatic agents, respectively. Our model further shows that for cytotoxic agents, evaluation of TB every 6 weeks is optimal to capture the true nadir, the time to nadir, the true progression and the true TTP, whereas for cytostatic agents, this evaluation is optimal every 10 weeks. CONCLUSIONS: Our results emphasize the importance to estimate the effects of tested drugs on tumor shrinkage before design any phase II clinical trials to choose optimal TB evaluation's timing.

Guidelines for the definition of time-to-event end points in renal cell cancer clinical trials: results of the DATECAN project†.

BACKGROUND: In clinical trials, the use of intermediate time-to-event end points (TEEs) is increasingly common, yet their choice and definitions are not standardized. This limits the usefulness for comparing treatment effects between studies. The aim of the DATECAN Kidney project is to clarify and recommend definitions of TEE in renal cell cancer (RCC) through a formal consensus method for end point definitions. MATERIALS AND METHODS: A formal modified Delphi method was used for establishing consensus. From a 2006-2009 literature review, the Steering Committee (SC) selected 9 TEE and 15 events in the nonmetastatic (NM) and metastatic/advanced (MA) RCC disease settings. Events were scored on the range of 1 (totally disagree to include) to 9 (totally agree to include) in the definition of each end point. Rating Committee (RC) experts were contacted for the scoring rounds. From these results, final recommendations were established for selecting pertinent end points and the associated events. RESULTS: Thirty-four experts scored 121 events for 9 end points. Consensus was reached for 31%, 43% and 85% events during the first, second and third rounds, respectively. The expert recommend the use of three and two endpoints in NM and MA setting, respectively. In the NM setting: disease-free survival (contralateral RCC, appearance of metastases, local or regional recurrence, death from RCC or protocol treatment), metastasis-free survival (appearance of metastases, regional recurrence, death from RCC); and local-regional-free survival (local or regional recurrence, death from RCC). In the MA setting: kidney cancer-specific survival (death from RCC or protocol treatment) and progression-free survival (death from RCC, local, regional, or metastatic progression). CONCLUSIONS: The consensus method revealed that intermediate end points have not been well defined, because all of the selected end points had at least one event definition for which no consensus was obtained. These clarified definitions of TEE should become standard practice in all RCC clinical trials, thus facilitating reporting and increasing precision in between trial comparisons.

Assessment of health status over time by Prevalence and Weighted Prevalence functions: Interface in R.

The importance of evaluating complications and toxicity during and following treatment has been stressed in many publications. In most studies, these endpoints are presented descriptively and summarized by numbers and percentages but descriptive methods are rarely sufficient to evaluate treatment-related complications. Pepe and Lancar developed Prevalence and Weighted Prevalence functions which take into account the duration and the severity of complication unlike conventional methods of survival analysis or competing risks which are limited to the time to first event. The purpose of this paper is to describe features and use of two R functions, main.preval.func and main.wpreval.func, which were designed for the analysis of survival adjusted for quality of life. These functions compute descriptive statistics, survival and competing risks analysis and especially Prevalence and Weighted Prevalence estimations with confidence intervals and associated test statistics. The use of these functions is illustrated by several examples.

Trastuzumab duration effects within patient prognostic subgroups in the PHARE trial.

BACKGROUND: At 42.5 months of median follow-up, PHARE failed to show that 6 was non-inferior to 12 months of adjuvant trastuzumab. From the results of PHARE, questions remain regarding whether the magnitude of benefit derived from 1 year is sufficient to justify its systematic use for different patient subgroups. METHODS: Treatment effects were evaluated according to various tumour characteristics, and the multivariate Cox proportional hazards regression models were carried out on metastases-free survival (MFS) in the 12 months control arm. A prognostic score was defined providing the identification of patient categories with similar risks. The 6-month arm was used as a validation set in order to test for heterogeneity. This study is registered at clinicaltrials.gov, number NCT00381901. RESULTS: A total of 261 metastatic events were observed and four prognostic groups were defined: very low, low, intermediate and high risk in the 12-month arm. The corresponding 3-year MFS rates were 98.3%, 95.8%, 90.4% and 78.4% in the four prognostic groups, respectively. In the 6-month arm, the 3-year MFS rates were 98.3%, 94.2%, 85.7% and 74.8% in the four prognostic groups, respectively. CONCLUSION: In the very low-risk group, the potential absolute benefit of standard duration of trastuzumab was small enough to indicate that optimal standard treatment might be clinically questionable. On the other hand, the 3-year metastasis occurrence rates strongly support the need for a search of a more efficient treatment in the low-, intermediate- and high-risk groups.

Plasma and tissue proteomic prognostic factors of response in primary breast cancer patients receiving neoadjuvant chemotherapy.

A pathological complete response (pCR) after neoadjuvant chemotherapy is observed in approximately 20% of breast cancer patients. A proteomic analysis was performed on plasma and tumor tissue before treatment to evaluate its potential impact on the prediction of response. One hundred and forty-nine breast cancer patients eligible for neoadjuvant chemotherapy were included in the study between February 2004 and January 2009 at three centers. The proteomic analysis was performed using SELDI Technology (ProteinChip CM10 pH4, IMAC-Cu and H50). Three acquisition protocols were used according to the mass range. Plasma and tumor proteomic signatures were generated using generalized ROC criteria and cross-validation. Twenty-eight (18.8%) patients out of 149 experienced a pCR according to Sataloff criteria. In the cytosol analysis, respectively 4, 2 and 8 proteins had significantly different levels of expression in the responders and non-responders using IMAC-Cu, H50 and CM10 pH4. Among the 8 proteins of interest on CM10 pH4, 2 (C1 and C7) were selected and were validated in 95.0 and 85.6% of the models. In the plasma analysis, respectively 12, 6 and 2 proteins had different levels of expression using the same proteinchips. Among the 12 plasma proteins of interest on IMAC-Cu, 2 (P1 and P7) were selected and were validated in 94.8 and 97.6% of the models. A combined proteomic signature was generated, which remained statistically significant when adjusted for hormone receptor status and Ki-67. Our results show that proteomic analysis can differentiate complete pathological responders in breast cancer patients after neoadjuvant chemotherapy.

Tumor marker kinetics predict outcome in patients with relapsed disseminated non-seminomatous germ-cell tumors.

BACKGROUND: An early serum tumor marker (TM) decline during chemotherapy was shown to independently predict survival in patients with poor-prognosis disseminated non-seminomatous germ-cell tumors (NSGCTs). The aim of this study was to assess whether a TM decline (TMD) also correlates with the outcome in the salvage setting. PATIENTS AND METHODS: Data regarding 400 patients with progressive or relapsed disseminated NSGCTs after first-line chemotherapy prospectively accrued onto two phase III clinical trials were obtained. Serum alpha-fetoprotein (AFP) and/or human chorionic gonadotropin (hCG) were assessed at baseline and after 6 weeks of chemotherapy. A total of 297 patients, 185 and 112 in the training and validation sets, with initially abnormal TMs for whom a change from baseline could be established were used for this analysis. RESULTS: An unfavorable decline in either AFP or hCG was predictive of progression-free survival (PFS) [hazard ratio, HR = 2.15, (95% CI 1.48-3.11); P < 0.001; 2-year PFS rate: 50% versus 26%] as was the Lorch prognostic score (LPS). In the multivariate analysis, an unfavorable TMD, stratified based on the LPS, was an independent adverse prognostic factor for PFS and OS. CONCLUSION: An unfavorable TMD during the first 6 weeks after chemotherapy is associated with a poorer outcome in patients with relapsed disseminated NSGCTs.

Prediction of early death among patients enrolled in phase I trials: development and validation of a new model based on platelet count and albumin.

BACKGROUND: Selecting patients with 'sufficient life expectancy' for Phase I oncology trials remains challenging. The Royal Marsden Hospital Score (RMS) previously identified high-risk patients as those with ≥ 2 of the following: albumin <35 g l(-1); LDH > upper limit of normal; >2 metastatic sites. This study developed an alternative prognostic model, and compared its performance with that of the RMS. METHODS: The primary end point was the 90-day mortality rate. The new model was developed from the same database as RMS, but it used Chi-squared Automatic Interaction Detection (CHAID). The ROC characteristics of both methods were then validated in an independent database of 324 patients enrolled in European Organization on Research and Treatment of Cancer Phase I trials of cytotoxic agents between 2000 and 2009. RESULTS: The CHAID method identified high-risk patients as those with albumin <33 g l(-1) or ≥ 33 g l(-1), but platelet counts ≥ 400.000 mm(-3). In the validation data set, the rates of correctly classified patients were 0.79 vs 0.67 for the CHAID model and RMS, respectively. The negative predictive values (NPV) were similar for the CHAID model and RMS. CONCLUSION: The CHAID model and RMS provided a similarly high level of NPV, but the CHAID model gave a better accuracy in the validation set. Both CHAID model and RMS may improve the screening process in phase I trials.

[Stereotactic body radiation therapy for liver primary and metastases: the Lille experience]. / Radiothérapie stéréotaxique hépatique par CyberKnife(®) : l'expérience lilloise.

PURPOSE: The CyberKnife(®) system is a recent radiation therapy technique that allows treatment of liver lesions with real-time tracking. Because of its high precision, the dose administered to the tumor can be increased. We report Oscar-Lambret Cancer Centre experience in the treatment of primary and secondary liver lesions. PATIENTS AND METHODS: It is a retrospective study analyzing all the patients who have been treated for their liver lesions since July 2007. A hundred and twenty patients have been treated: 42 for hepatocellular carcinoma, 72 for liver metastases and six for cholangiocarcinoma. Gold seeds need to be implanted before the treatment and are used as markers to follow the movement of the lesion due to respiration. On average, the treatment is administered in three to four sessions over 12 days. A total dose of 40 to 45 Gy at the 80% isodose is delivered. Local control and overall survival analysis with Log-rank is performed for each type of lesion. RESULTS: Treatment tolerance is good. The most common toxicities are of digestive type, pain and asthenia. Six gastro-duodenal ulcers and two radiation-induced liver disease (RILD) were observed. At a median follow-up of 15 months, the local control rate is respectively of 80.4% and 72.5% at 1 and 2 years. Overall survival is 84.6 and 58.3% at 1 and 2 years. The local control is significantly better for the hepatocellular carcinoma and overall survival is significantly better for liver metastases (P<0.05). The local control rate and overall survival at 1 year for cholangiocarcinoma is 100%. CONCLUSION: CyberKnife(®) is a promising technique, well tolerated, with tumoral local control rates comparable to other techniques. Its advantage is that it is very minimally invasive delivered as an outpatient procedure in a frail population of patient (disease, age).

SAE: an R package for early stopping rules in clinical trials.

In the case of an unexpected high frequency of serious adverse events (SAE), statistical methods are needed to help in the decision making process as to continuation of accrual to the trial. This paper describes an R package, named SAE that implements a method recently developed by defining stopping rules after each observed SAE. The package function control for excessive toxicity either during the trial at the observation of each SAE (function SAE) or during the planning phase of a clinical trial (function DESIGN). This description and the package documentation are complementary to help the users to apply the method. The main difficulty in the implementation of the method is the choice of a priori parameters. Data from an ongoing clinical trial are presented as an example to improve the understanding and the use of the package.

Efficacy of irinotecan in combination with 5-fluorouracil (FOLFIRI) for metastatic gastric or gastroesophageal junction adenocarcinomas (MGA) treatment.

OBJECTIVES: The most commonly used schedules are 5-FU in combination with CDDP with or without epirubicin (ECF) or docetaxel (TCF) in treatment of MGA patients (pts), independently of HER status. We evaluated the efficacy of FOLFIRI regimen in a large retrospective series of MGA pts. METHODS: Two hundred and twelve pts from 13 French centers were treated with at least one cycle of FOLFIRI (irinotecan 180 mg/m2 intravenous (i.v.) over 90 minutes on day 1 with folinic acid (FA) 400mg/m2 i.v. over two hours followed by 5-FU 400mg/m2 i.v. bolus then 5- FU 2400 mg/m2 continuous infusion over 46 hours on day 1, repeated every 14 days). Primary tumour sites were 120 (58%) stomach and 92 (42%) gastroesophageal junction. FOLFIRI was administered as first-line in 137 (65%) pts and as later-line in 75 (35%) pts for MGA. RESULTS: There was no difference between chemonaive and not chemonaive pts treated as firstline in terms of response rate 37% (95% CI: 25-50) vs 44% (95% CI: 21-69), median PFS, 6.7 (95% CI: 5.5-9.9) vs 5.3 months (95% CI: 3.6-6.9) (P = 0.25), and OS, 13.1 (95% CI: 11.7-18.7) vs 8.8 months (95% CI: 7.3­15.6) (P = 0.19), respectively. There was no difference between pts treated as second or later-line in terms of response rate 20% (95% CI: 8-39) vs 22% (95% CI: 6-48), median PFS, four months (95% CI: 2.8-5.4) vs 3.5 months (95% CI: 2.3-4.5) (P = 0.56), and OS, 10.4 months (95% CI: 5.4-14.4) vs 5.3 months (95% CI: 3.5-11.3) (P = 0.58), respectively. The global grade 3-4 toxicities were: diarrhea 11%, vomiting 9%, neutropenia 18%, febril neutropenia 4% (one toxic death). CONCLUSIONS: This retrospective study confirms the activity and good tolerance of FOLFIRI regimen in MGA as first-line as well as later-line.

A comparison of model choices for the Continual Reassessment Method in phase I cancer trials.

Determination of the maximum tolerated dose (MTD) is the main objective of phase I trials. Trials are typically carried out with restricted sample sizes. Model-based approaches proposed to identify the MTD (including the Continual Reassessment Method or CRM) suppose a simple model for the dose-toxicity relation. At this early stage of clinical development, the true family of models is not known and several proposals have been done. Asymptotic convergence of the recommendation to the true MTD can be obtained with a one-parameter model even in case of model misspecification. Nevertheless, operating characteristics with finite sample sizes can be largely affected by the choice of the model. In this paper, we evaluate and compare several models in a simulation framework. This framework includes a large class of dose-toxicity relations against which to test the competing models, an 'optimal' method that provides efficient non-parametric estimates of the probability of dose limiting toxicity to serve as a benchmark and as a graphic representation. In particular we explore the use of a one-parameter versus a two-parameter model, we compare the power and the logistic models and finally we investigate the impact of dose recoding on the operating characteristics. Comparisons are carried out with both a likelihood approach and a Bayesian approach for model estimations. We show that average performances of a one-parameter model are superior and that the power model has good operating characteristics. Some models can speed up dose escalation and lead to more aggressive designs. We derive some behavior related to the choice of model and insist on the use of simulations under several scenarios before the initiation of each new trial in order to determine the best model to be used.

A phase III randomised trial of LV5FU2 + irinotecan versus LV5FU2 alone in adjuvant high-risk colon cancer (FNCLCC Accord02/FFCD9802).

BACKGROUND: This multicenter adjuvant phase III trial evaluated the addition of irinotecan to LV5FU2 in colon cancer patients at high risk of relapse. PATIENTS AND METHODS: A total of 400 patients with histologically proven primary colon cancer with postoperative N1 detected by occlusion/perforation or N2 were randomised to: A-LV5FU2 [leucovorin 200 mg/m(2), 2-h infusion, 5-fluorouracil (5-FU) 400 mg/m(2) bolus, 600 mg/m(2) 22-h continuous infusion, days 1 and 2] or B-LV5FU2 + IRI (irinotecan 180 mg/m(2) 90-min infusion day 1 + LV5FU2) fortnightly for 12 cycles. Primary end point was disease-free survival (DFS). RESULTS: Median follow-up was 63 months. Significantly more T4 tumours and 15 or more positive lymph nodes were observed in arm B. 5-FU relative dose intensity (RDI) was >0.80 for 94% and 77% in arms A and B, respectively (P < 0.001). Irinotecan RDI was >0.80 for 70% patients. There were more grades 3 and 4 neutropenia in arm B (4% versus 28%, P < 0.001). The 3-year DFS was 60% [95% confidence interval (CI) 53% to 66%] and 51% (95% CI 44% to 58) in arms A and B, respectively. No difference was observed [hazard ratio (HR) = 1.12, 95% CI 0.85-1.47, P = 0.42] even when adjusted for prognostic factors (adjusted HR = 0.98, 95% CI 0.74-1.31, P = 0.92). The 5-year overall survival (OS) was 67% (95% CI 59% to 73%) and 61% (95% CI 53% to 67%) in arms A and B, respectively. CONCLUSION: Adjuvant LV5FU2 + IRI compared with LV5FU2 alone in patients at high risk of relapse showed no improvement in DFS and OS.

A randomized, prospective study using the LPG technique in treating radiation-induced skin fibrosis: clinical and profilometric analysis.

BACKGROUND/PURPOSE: Cutaneous fibrosis is the quite mandatory sequela after a breast cancer treated by radiotherapy and it induces more or less important functional troubles. The LPG technique is a technique of mechanical massage that allows skin mobilization by folding/unfolding. The aim of this study was to evaluate the changes on irradiated skin before and after LPG treatment by clinical and skin replica analysis. METHODS: Twenty women, 43-55 years old, who had been treated from 6 to 16 months before, for breast cancer with a conservative surgery and radiotherapy, had been enrolled in the trial. They were divided into two groups after randomization: the first group was LPG treated three times a week for 1 month; the second group was only placed under medical supervision. The clinical criteria studied were systematically studied before (T0), at the end of treatment (T1) and 1 month after the end of treatment (T2): pain, itching, skin dryness, erythema, skin infiltration, feeling of tightness and of induration of the skin. Softening of the skin was assessed at T1 and T2. Cutaneous replica was performed on the internal upper 1/4 of each breast with silicone material before, after and 1 month later after the end of the treatment. After polymerization, the replica was stored and then blindly analyzed by image analysis software. The following parameters were systematically measured: average skin roughness, average of wrinkles' depth and residual length, wrinkle number and the space between them. RESULTS: Clinically, the LPG treatment induced a decrease of erythema (10% of the patients vs. 40% before treatment), a decrease of pain and pruritus (10% vs. 20% and 40%, respectively) and a decrease of the feeling of induration of the skin (10% of the patients vs. 70% before treatment). Furthermore, a skin-softening sensation was noted by seven patients vs. one in the control group. Replica shows an increase of roughness and of furrow depth without any change in the residual length and an increase in the space between the wrinkles, whose number decreases. CONCLUSION: This study confirms the impact of the clinical sequelae induced on skin after radiotherapy and shows improvement of the clinical signs after treatment by the LPG technique. The latter induces changes of micro relief, suggesting a softening effect on the skin. These preliminary results have to be confirmed on a more important group of patients.

Tritherapy with fluorouracil/leucovorin, irinotecan and oxaliplatin (FOLFIRINOX): a phase II study in colorectal cancer patients with non-resectable liver metastases.

PURPOSE: To assess the rate of R(0) resection of liver metastases achieved after chemotherapy with FOLFIRINOX. PATIENTS AND METHODS: Patients with histologically proven primary colorectal cancer and bidimensionally measurable liver metastasis, not fully resectable based on technical inability to achieve R(0) resection, but potentially resectable after tumor reduction, were given FOLFIRINOX: oxaliplatin 85 mg/m(2), irinotecan 180 mg/m(2), leucovorin 400 mg/m(2), bolus fluorouracil 400 mg/m(2) and fluorouracil 46-h continuous IV infusion 2,400 mg/m(2), every 2 weeks for a maximum of 12 cycles. RESULTS: Thirty-four patients were enrolled. Response rate before surgery was 70.6% (95%CI: 52.5-84.9). Twenty-eight patients (82.4%) underwent hepatic resection and nine achieved R(0) resection [26.5% (95% CI: 12.9-44.4%)]. The rate of clinical complete remission after surgery was 79.4%. Two-year overall survival was 83%. The most frequent grade 3 or 4 toxicities were neutropenia (64.8%), diarrhea (29.4%), fatigue (23.5%), abdominal cramps (14.7%), neuropathy and nausea (11.8% each), and AST/ALT elevation (14.7/11.8%). Only one patient experienced febrile neutropenia, four patients withdrew due to toxicity and no toxic death was observed. CONCLUSION: FOLFIRINOX, with an acceptable toxicity profile, shows a high response rate in liver metastases from colorectal cancer. The rate of hepatic resection in patients initially not resectable, is attractive and warrants further assessment of this regimen in randomized studies compared to standard regimens.

Peritoneal fluid cytokines and the differential diagnosis of benign and malignant ovarian tumors and residual/recurrent disease examination.

This study aimed to assess the potential value of peritoneal fluid cytokine examination for the differential diagnosis of ovarian tumors and for evaluating residual or recurrent disease after treatment. The cytokines that are commonly elevated in ovarian cancer, VEGF, IL-6, bFGF, IL-8 and M-CSF, and a reference ovarian tumor marker, CA 125, were measured in peritoneal fluids of 53 previously untreated patients with epithelial ovarian cancer, 18 ovarian cancer patients after surgical treatment and chemotherapy, and 17 patients with benign epithelial ovarian tumors. Non-parametric statistical analysis of data was performed. Ovarian cancer peritoneal fluids, as compared to peritoneal fluids of patients with benign ovarian tumors, contained significantly higher concentrations of IL-6, VEGF and CA 125, and significantly lower concentrations of bFGF and M-CSF, but only the levels of IL-6 and VEGF were significantly higher in peritoneal fluids of stage I and II ovarian cancer patients than of patients with benign ovarian conditions. IL-6 at the cutoff level of 400 pg/mL discriminated benign and malignant ovarian tumors with 92% sensitivity and 60% specificity, while VEGF at the cutoff of 400 pg/mL had 90% sensitivity and 80% specificity. At the cutoff level of 1200 pg/mL, IL-6 had 84% sensitivity and 87% specificity. A radical decrease in local cytokine and CA 125 levels in patients after treatment was independent of therapy outcome. IL-6 and VEGF measurements in peritoneal fluids might be useful for the differential diagnosis of malignant and benign ovarian conditions, but not for residual or recurrent disease examination.

Refining the optimal chemotherapy regimen for good-risk metastatic nonseminomatous germ-cell tumors: a randomized trial of the Genito-Urinary Group of the French Federation of Cancer Centers (GETUG T93BP).

BACKGROUND: High cure rates are expected in good-risk metastatic nonseminomatous germ-cell tumor (NSGCT) patients with bleomycin, etoposide and cisplatin. PATIENTS AND METHODS: Patients received either three cycles of BE500P or four cycles of E500P every 3 weeks. Disease was defined according to the Institut Gustave Roussy prognostic model. Patients were retrospectively assigned into the International Germ Cell Cancer Collaborative Group (IGCCCG) classification. A sample size of 250 patients was necessary for an expected favorable response rate (primary end point) of 90% and not more than a 10% difference between the two arms. RESULTS: Among 257 assessable patients, 124 and 122 patients achieved a favorable response in the 3BE500P and 4E500P arms, respectively (P = 0.34). Median follow-up was 53 months. The 4-year event-free survival rates were 91% and 86%, respectively (P = 0.135). The 4-year overall survival rates were not significantly different [five deaths versus 12 deaths, respectively (P = 0.096)]. Similar nonsignificant trends were observed in good IGCCCG prognosis patients. CONCLUSIONS: Both regimens produced similar results in terms of favorable response rates. As the trial was underpowered for survival analyses, conclusive data would require a larger randomized trial. Unless such a study is done, 3BE500P is the treatment of choice for metastatic NSGCT patients.

Measuring agreement for ordered ratings in 3 x 3 tables.

OBJECTIVES: When two raters consider a qualitative variable ordered according to three categories, the qualitative agreement is commonly assessed with a symmetrically weighted kappa statistic. However, these statistics can present paradoxes, since they may be insensitive to variations of either complete agreements or disagreements. METHODS: Agreement may be summarized by the relative amounts of complete agreements, partial and maximal disagreements beyond chance. Fixing the marginal totals and the trace, we computed symmetrically weighted kappa statistics and we developed a new statistic for qualitative agreements. Data sets from the literature were used to illustrate the methods. RESULTS: We show that agreement may be better assessed with the unweighted kappa index, kappa(c), and a new statistic zeta, which assesses the excess of maximal disagreements with respect to the partial ones, and does not depend on a particular weighting system. When zeta is equal to zero, maximal and partial disagreements beyond chance are equal. With its estimated large sample variance, we compared the values of two contingency tables. CONCLUSIONS: The (kappa(c), zeta) pair is sensitive to variations in agreements and/or disagreements and enables locating the difference between two qualitative agreements. The qualitative agreement is better with increasing values of kappa(c) and zeta.

A randomised trial of high-dose chemotherapy in the salvage treatment of patients failing first-line platinum chemotherapy for advanced germ cell tumours.

BACKGROUND: Incomplete remission or relapse from first-line chemotherapy has poor prognosis in male germ cell tumour patients. This phase III randomised trial compares conventional salvage to high-dose-intensification chemotherapy. PATIENTS AND METHODS: Between February 1994 and September 2001, 280 patients from 43 institutions in 11 countries, were randomly assigned to receive either four cycles of cisplatin, ifosfamide and etoposide (or vinblastine) (arm A), or three such cycles followed by high-dose carboplatin, etoposide and cyclophosphamide (CarboPEC) with haematopoietic stem cell support (arm B). RESULTS: Similar complete and partial response rates were observed in both treatment arms (56%; 95% CI 50% to 62%). There were 3% and 7% toxic deaths in arms A and B, respectively. No significant improvements with CarboPEC were observed in either 3-year event-free survival (35% versus 42%, P=0.16) or overall survival (53%; 95% CI 46% to 59%). Complete responders with CarboPEC had a significant improvement in disease-free survival (55% versus 75% at 3 years, P <0.04). CONCLUSIONS: The single cycle of high-dose salvage chemotherapy after three cycles of standard dose chemotherapy had no effect on treatment outcomes. These results suggest that data from uncontrolled studies should not be used to justify routine use of a toxic and expensive treatment without confirmation in a randomised trial.

Two-part statistics with paired data.

In epidemiology, we often study data from a mixed distribution, i.e. with a clump of observations at zero and positive continuous data. Lachenbruch developed statistics for this kind of data for independent samples. These tests are the sum of one test for equality of proportions of zero values and one conditional test for the continuous distribution. This paper concerns the adaptation of these tests to paired samples. Like Lachenbruch, we developed two statistics, which tend to a two-degree-of-freedom chi2 distribution. These two-part statistics are the sum of McNemar's test for testing the equality of proportions of zero values, and the Wilcoxon signed-rank test or the paired Student's test for testing the equality of the distribution of positive values. We studied the behaviour of these tests for various proportions of zeros, and mean values of the continuous distribution. All tests are efficient when the smaller proportion of zero values corresponds to the population with the larger mean. In all other situations, the two-part statistics are superior to the others. These methods are applied to a matched case-control study of lower limb venous insufficiency.