Testing for the presence of multiple sources of informative dropout in longitudinal data.

Longitudinal studies tracking the rate of change are subject to patient dropout. This dropout process might not only be informative but also heterogeneous in the sense that different causes might contribute to multiple patterns of informative dropout. We propose a random-effects approach to test for homogeneity of informative dropout that accommodates the realistic situation where reasons for dropout are not fully understood, or perhaps are even entirely unknown. The proposed score test is robust in that it does not depend on the underlying distribution of the informative dropout random effects. The test allows for an additional level of clustering among participating subjects, as might be found in a family study, provided the informative dropout random effects have a known correlation structure.

Endoscopic mapping and surrogate markers for better surveillance in Barrett esophagus. A study of 700 biopsy specimens.

Surveillance methods in Barrett esophagus (BE) using light microscopic examination of random biopsy specimens may miss focal dysplasia. In addition, dysplastic foci identified initially may not be relocated subsequently, making chemoprevention studies difficult. By using a special gastroscope, systematic mapping (4-quadrant biopsy specimens at 1-cm intervals) was performed in 22 patients (33 total mappings yielding 700 biopsy specimens). H&E, immunohistochemistry, and DNA ploidy analysis were performed. c-erbB-2 and positive Ki-67 were detected only in dysplastic sites; thus, their detection did not precede morphologically identifiable dysplasia. On the other hand, aneuploidy and p53 were detected in dysplastic and nondysplastic areas. p53 was correlated with dysplasia, and S-phase narrowly missed correlation, while aneuploidy was not correlated. PCNA and bcl-2 were ubiquitous, limiting their usefulness. On second maps, epithelial type was reidentified with 81% accuracy. A significant correlation was found between p53 and dysplasia. Sites of dysplasia and abnormal biomarkers could be relocated accurately by using endoscopic mapping. Therefore, mapping combined with biomarker studies may provide better surveillance and serve as a useful technique in chemoprevention studies.

A pilot trial assessing the efficacy of paroxetine hydrochloride (Paxil) in controlling hot flashes in breast cancer survivors.

BACKGROUND: Many breast cancer survivors suffer debilitating hot flashes. Estrogen, the drug of choice in perimenopausal women, is generally not recommenced to breast cancer survivors. Nonhormonal treatments are mostly disappointing. Anecdotal reports in our institution suggested that the selective serotonin-reuptake inhibitor, paroxetine hydrochloride, might be efficacious in alleviating hot flashes. PATIENTS AND METHODS: Thirty women with prior breast cancer who were suffering at least two hot flashes a day entered a single institution pilot trial to evaluate paroxetine's efficacy in reducing the frequency and severity of hot flashes. After completing daily diaries for one week on no therapy, the women received open-label paroxetine, 10 mg daily for one week, followed by four weeks of paroxetine, 20 mg daily. The women completed hot-flash daily diaries throughout the study period, and a health-related symptom-assessment questionnaire and a quality-of-life rating scale in the first and sixth week of the study. RESULTS: Twenty-seven women completed the six-week study period. The mean reduction of hot flash frequency was 67% (95% confidence interval (95% CI): 56%-79%). The mean reduction in hot flash severity score was 75% (95% CI: 66%-85%). There was a statistically significant improvement in depression, sleep, anxiety, and quality of life scores. Furthermore, 25 (83%) of the study participants chose to continue paroxetine therapy at the end of study. The most common adverse effect was somnolence, resulting in drug discontinuation in two women, and dose reduction in two women. One woman discontinued drug due to anxiety. CONCLUSIONS: Paroxetine hydrochloride is a promising new treatment for hot flashes in breast cancer survivors, and warrants further evaluation in a double-blind randomized placebo-controlled trial.

A modification of Simon's optimal design for phase II trials when the criterion is median sample size.

We present a modification to Simon's optimal design for phase II trials in which the objective is to minimize the median sample size rather than the expected sample size when the true response rate is poor (p = p0). We argue that the modified design may be preferred in smaller institutions when the focus is on a single or small number of phase II trials rather than a large program of phase II trials.

Optimal multi-stage designs for a phase II trial that permits one dose escalation.

Simon's optimal two-stage design is widely used to conduct single-dose phase II clinical trials. We extend this basic methodology to the situation where the researcher desires to test an experimental drug for activity at a low dose level, but is willing to increase the dose part-way through the trial if the early results suggest that the low dose is ineffective. Interest is confined to at most one dose escalation, and no consideration is given to escalating the dose within a patient. Optimal multi-stage designs are presented that are more efficient than the naive approach of merely conducting two consecutive Simon optimal trials, one at the low dose and the second (if deemed necessary) at the high dose. As in Simon's original design, toxicity is not considered here as a primary endpoint. Hence, the designs presented in this paper are appropriate only when the toxicity of the drug is well understood at both dose levels.

Inference for odds ratio regression models with sparse dependent data.

Suppose the number of 2 x 2 tables is large relative to the average table size, and the observations within a given table are dependent, as occurs in longitudinal or family-based case-control studies. We consider fitting regression models to the odds ratios using table-level covariates. The focus is on methods to obtain valid inferences for the regression parameters beta when the dependence structure is unknown. In this setting, Liang (1985, Biometrika 72, 678-682) has shown that inference based on the noncentral hypergeometric likelihood is sensitive to misspecification of the dependence structure. In contrast, estimating functions based on the Mantel-Haenszel method yield consistent estimators of beta. We show here that, under the estimating function approach, Wald's confidence interval for beta performs well in multiplicative regression models but unfortunately has poor coverage probabilities when an additive regression model is adopted. As an alternative to Wald inference, we present a Mantel-Haenszel quasi-likelihood function based on integrating the Mantel-Haenszel estimating function. A simulation study demonstrates that, in medium-sized samples, the Mantel-Haenszel quasi-likelihood approach yields better inferences than other methods under an additive regression model and inferences comparable to Wald's method under a multiplicative model. We illustrate the use of this quasi-likelihood method in a study of the familial risk of schizophrenia.

Radiation therapy for rhabdomyosarcoma: local failure risk for Clinical Group III patients on Intergroup Rhabdomyosarcoma Study II.

PURPOSE: A subset of 362 pediatric patients with rhabdomyosarcoma was selected from a total of 532 eligible IRS-II patients in Clinical Group III to assess the local and regional failure rates following radiotherapy and to determine patient, tumor, and treatment factors contributing to the risk for local and regional failure. METHODS AND MATERIALS: The study population was selected from all eligible IRS-II Clinical Group III patients. Excluded patients were those with "special pelvic" primary sites whose protocol management restricted radiotherapy (n = 123), and those who were removed from the study before radiotherapy was to begin, or because it was omitted (n = 47). A binary recursive partitioning model was used to identify subgroups of the remaining 362 patients at risk of local or regional failure. RESULTS: The local (only) failure rate was 17% (95% confidence interval, 13-21%), and the local (all) failure rate was 20% (95% confidence interval, 16-24%). The 5-year actuarial risk of local (all) failure was 22% (95% confidence interval, 18-27%). The risk of regional (nodal) failure was between 2% and 23%. Increasing tumor size predicted an increased local failure risk. Primary tumors located above the clavicle had a reduced risk of local failure. The binary recursive partitioning model identified a subset of patients at high risk of local failure. Those patients had primary tumors in the chest, pelvic region, extremity, or trunk, or tumors > 10 cm in diameter. Their local failure rate was 35% (compared to 15% for the remaining patients). The subset of patients at high risk for regional (nodal) failure had node involvement at diagnosis and a primary tumor originating at a site other than orbit, parameningeal, or trunk. Compliance with radiation treatment guidelines approached but did not achieve statistical significance as a predictive factor for local failure. By univariate analysis, factors not influencing local failure risk were age, race, gender, adenopathy, and histology. CONCLUSION: Radiation therapy and chemotherapy administered to Clinical Group III patients entered into the IRS-II protocol produced sustained local control in most cases. Knowledge of the factors which predict an increased risk of local or regional failure will facilitate the design of new treatment strategies.

Statistical approaches to experimental design and data analysis of in vivo studies.

The objective of any experiment is to obtain an unbiased and precise estimate of a treatment effect in an efficient manner. Statistical aspects of the design, conduct, and analysis of the experiment play a major role in determining whether this goal is met. We highlight some of the more important statistical issues that pertain to in vivo studies. Particular emphasis is placed on the role of randomization, the number of animals, the utilization of repeated measures data, adjustments for missing data, and dealing with multiple causes of death or treatment failure. The discussion is not intended to be a comprehensive guide to all the statistical issues that can occur in animal experiments. Rather, the objective is to acquaint researchers with components of the experiment that will require careful statistical thought.