Statistical model to determine the relationship of response and survival in patients with advanced ovarian cancer treated with chemotherapy.

BACKGROUND: A statistically appropriate analysis of the association between survival and response measures in patients with ovarian cancer could help to define the role of response rate in planning, monitoring, and interpreting the results of clinical trials. PURPOSE: This study was designed to investigate the relationship between antitumor response determined by clinical or pathological means and survival in patients with advanced ovarian cancer with no previous treatment. We focused on avoiding the limitations of the usual approach of comparing durations of survival for patients responding to therapy with those for nonresponders. METHODS: A new meta-analytic statistical model we developed was used to analyze data from 26 randomized clinical trials published between 1975 and 1989. Our model incorporates intra-study and inter-study sources of variability in the estimates of response and survival. The study also addresses the methodological problems of evaluating response as a surrogate end point and the relevance of this association to clinical decision making and the design of clinical trials. RESULTS: For 13 studies in which response was pathologically assessed, an improvement in surgically documented complete response rate was associated with an increase in median survival. A similar but apparently smaller effect was found for the association between objective clinical response and median survival in the 25 studies reporting these data. CONCLUSIONS: These results suggest that therapeutic measures must produce large improvements in clinical response rates to achieve meaningful effects on median survival. Improvement in surgically documented complete response rate appears to be more strongly associated with increased median survival and, hence, might be used for interim monitoring in clinical trials, but the role of second-look procedures in clinical management is controversial.

Effect of different types and amounts of fat on the development of mammary tumors in rodents: a review.

We performed a meta-analysis on data extracted from 97 reports of experiments, involving a total of 12,803 mice or rats, studying the effect on mammary tumor incidence of different types of dietary fatty acids. Fatty acids were categorized into saturated, monounsaturated, n-6 polyunsaturated, and n-3 polyunsaturated. We modeled the relation between tumor incidence and percentage of total calories from these fatty acids using conditional logistic regression and allowing for varying effects between experiments, and for each fatty acid we estimated the effect of substituting the fatty acid calories for nonfat calories. Our results show that n-6 polyunsaturated fatty acids (PUFAs) have a strong tumor-enhancing effect and that saturated fats have a weaker tumor-enhancing effect. The n-3 PUFAs have a small protective effect that is not statistically significant. There is no significant effect of monounsaturated fats. n-6 PUFAs have a stronger tumor-enhancing effect at levels under 4% of total calories, but an effect is still present at intake levels greater than 4% of calories. In addition, when the intake of n-6 PUFAs is at least 4% of calories, the n-6 PUFA effect remains stronger than the saturated fat effect.

Evaluation of 2 brief instruments and a food-frequency questionnaire to estimate daily number of servings of fruit and vegetables.

BACKGROUND: Measurement of fruit and vegetable intake is important in the surveillance of populations and in epidemiologic studies that examine the relations between diet and disease. Some situations require the use of brief dietary assessment tools. OBJECTIVE: Our objective was to evaluate the performance of 2 brief dietary assessment instruments, a 7-item standard screener and a new 16-item screener, and a complete food-frequency questionnaire (FFQ) in measuring total fruit and vegetable consumption. DESIGN: About 800 men and women from the National Institutes of Health-AARP Diet and Health Study completed an FFQ, 1 of the 2 screeners, and two 24-h dietary recalls. Fruit and vegetable intakes as measured by each screener and the FFQ were compared with estimated true usual intake by using a measurement-error model. RESULTS: Median daily servings of fruit and vegetables were underestimated by both screeners. The estimated agreement between true intake and the screener was higher for the new screener than for the standard screener and was higher for women than for men. The estimated agreement between true intake and the FFQ was higher than that for both screeners. Attenuation coefficients for the FFQ and screeners were comparable. CONCLUSIONS: For estimating median intakes of fruit and vegetables and the prevalence of recommended intakes being met, the use of screeners without appropriate adjustment is suboptimal. For estimating relative risks in the relations between fruit and vegetable intake and disease, screeners and this FFQ are similar in performance.

Checking for completeness of 24-h urine collection using para-amino benzoic acid not necessary in the Observing Protein and Energy Nutrition study.

BACKGROUND/OBJECTIVES: The orally administered para-amino benzoic acid (PABA) is known to have near 100% excretion in urine and is used as a measure of 24-h urine collection completeness (referred to as PABAcheck). The purpose was to examine the effect of including urine collections deemed incomplete based on PABAcheck in a dietary measurement error study. SUBJECTS/METHODS: The Observing Protein and Energy Nutrition (OPEN) study was conducted in 1999-2000 and included 484 men and women aged 40-69 years. A food frequency questionnaire and 24-h dietary recalls were evaluated using recovery biomarkers that included urinary nitrogen and potassium from two 24-h urine collections. Statistical modeling determined the measurement error properties of dietary assessment instruments. In the original analyses, PABAcheck was used as a measure of complete urine collection; incomplete collections were either excluded or adjusted to acceptable levels. The OPEN data were reanalyzed including all urine collections and by using criteria based on self-reported missing voids to assess the differences. RESULTS: Means and coefficients of variation for biomarker-based protein and potassium intakes, and measurement error model-based correlations and attenuation factors were similar regardless of whether PABAcheck or missed voids were considered. CONCLUSION: PABAcheck may not be required in large population-based biomarker studies. However, until there are more analyses evaluating the necessity of a PABAcheck, it is recommended that PABA be given to all participants, but not necessarily analyzed. Then, PABAcheck could be used selectively as a marker of completeness among the collections in which low levels of biomarker are detected or for which noncompliance is suspected.

Associations between food patterns defined by cluster analysis and colorectal cancer incidence in the NIH-AARP diet and health study.

BACKGROUND/OBJECTIVES: To examine associations between food patterns, constructed with cluster analysis, and colorectal cancer incidence within the National Institutes of Health-AARP Diet and Health Study. SUBJECTS/METHODS: A prospective cohort, aged 50-71 years at baseline in 1995-1996, followed until the end of 2000. Food patterns were constructed, separately in men (n=293,576) and women (n=198,730), with 181 food variables (daily intake frequency per 1000 kcal) from a food frequency questionnaire. Four large clusters were identified in men and three in women. Cox proportional hazards regression examined associations between patterns and cancer incidence. RESULTS: In men, a vegetable and fruit pattern was associated with reduced colorectal cancer incidence (multivariate hazard ratio, HR: 0.85; 95% confidence interval, CI: 0.76, 0.94), when compared to less salutary food choices. Both the vegetable and fruit pattern and a fat-reduced foods pattern were associated with reduced rectal cancer incidence in men. In women, a similar vegetable and fruit pattern was associated with colorectal cancer protection (age-adjusted HR: 0.82; 95% CI: 0.70, 0.95), but the association was not statistically significant in multivariate analysis. CONCLUSIONS: These results, together with findings from previous studies support the hypothesis that micronutrient dense, low-fat, high-fiber food patterns protect against colorectal cancer.

OPEN about obesity: recovery biomarkers, dietary reporting errors and BMI.

OBJECTIVE: Obesity-related under-reporting of usual dietary intake is one of the most persistent sources of bias in nutrition research. The aim of this paper is to characterize obese and non-obese individuals with respect to reporting errors observed with two common dietary instruments, using energy and protein recovery biomarkers as reference measures. POPULATION AND METHODS: This report employs data from the Observing Protein and Energy Nutrition (OPEN) study. Analyses are based on stratified samples of 211 (57 obese) men and 179 (50 obese) women who completed 24-h recalls (24HR), food frequency questionnaires (FFQ), doubly labelled water (DLW) and urinary nitrogen (UN) assessments. RESULTS: In obese and non-obese subgroups, FFQ yielded lower energy and protein intake estimates than 24HR, although biomarker-based information indicated under-reporting with both dietary instruments. Gender differences in obesity-related bias were noted. Among women, the DLW-based energy requirement was 378 kcal greater in obese than in non-obese groups; the FFQ was able to detect a statistically significant portion of this extra energy, while the 24HR was not. Among men, the DLW-based energy requirement was 485 kcal greater in the obese group; however, neither FFQ nor 24HR detected this difference in energy requirement. Combining protein and energy estimates, obese men significantly over-reported the proportion of energy from protein using the 24HR, but not with the FFQ. In obese women, no significant reporting error for energy percent protein was observed by either method. At the individual level, correlations between energy expenditure and reported energy intake tended to be weaker in obese than non-obese groups, particularly with the 24HR. Correlations between true and reported protein density were consistently higher than for protein or energy alone, and did not vary significantly with obesity. CONCLUSION: This work adds to existing evidence that neither of these commonly used dietary reporting methods adequately measures energy or protein intake in obese groups. The 24HR, while capturing more realistic energy distributions for usual intake, may be particularly problematic in the obese.

Time-to-event analysis of longitudinal follow-up of a survey: choice of the time-scale.

Following individuals sampled in a large-scale health survey for the development of diseases and/or death offers the opportunity to assess the prognostic significance of various risk factors. The proportional hazards regression model, which allows for the control of covariates, is frequently used for the analysis of such data. The authors discuss the appropriate time-scale for such regression models, and they recommend that age rather than time since the baseline survey (time-on-study) be used. Additionally, with age as the time-scale, control for calendar-period and/or birth cohort effects can be achieved by stratifying the model on birth cohort. Because, as discussed by the authors, many published analyses have used regression models with time-on-study as the time-scale, it is important to assess the magnitude of the error incurred from this type of incorrect modeling. The authors provide simple conditions for when incorrect use of time-on-study as the time-scale will nevertheless yield approximately unbiased proportional hazards regression coefficients. Examples are given using data from the first National Health and Nutrition Examination Survey (NHANES I) Epidemiologic Followup Study. Additional issues concerning the analysis of longitudinal follow-up of survey data are briefly discussed.

Two approaches for estimating disease prevalence from population-based registries of incidence and total mortality.

Two approaches are described for estimating the prevalence of a disease that may have developed in a previous restricted age interval among persons of a given age at a particular calendar time. The prevalence for all those who ever developed disease is treated as a special case. The counting method (CM) obtains estimates of prevalence by dividing the estimated number of diseased persons by the total population size, taking loss to follow-up into account. The transition rate method (TRM) uses estimates of transition rates and competing risk calculations to estimate prevalence. Variance calculations are described for CM and TRM as well as for a variant of CM, called counting method times 10 (CM10), that is designed to yield more precise estimates than CM. We compare these three estimators in terms of precision and in terms of the underlying assumptions required to justify the methods. CM makes fewer assumptions but is typically ess precise than TRM or CM10. For common diseases such as breast cancer, CM may be preferred because its precision is excellent even though not as high as for TRM or CM10. For less common diseases, such as brain cancer, however, TRM or CM10 and other methods that make stabilizing assumptions may be preferred to CM.

Statistical analysis of animal cancer chemoprevention experiments.

We explore the use of a statistical model proposed by Kokoska (1987, Biometrics 43, 525-534) for the analysis of animal cancer chemoprevention experiments. We show, using an example, that the results derived from the method can be sensitive to the parametric forms of the distributions that are assumed, particularly to the distribution of the number of tumors per animal. We propose goodness-of-fit tests to aid in the choice of the distributions.

Permutation tests for joinpoint regression with applications to cancer rates.

The identification of changes in the recent trend is an important issue in the analysis of cancer mortality and incidence data. We apply a joinpoint regression model to describe such continuous changes and use the grid-search method to fit the regression function with unknown joinpoints assuming constant variance and uncorrelated errors. We find the number of significant joinpoints by performing several permutation tests, each of which has a correct significance level asymptotically. Each p-value is found using Monte Carlo methods, and the overall asymptotic significance level is maintained through a Bonferroni correction. These tests are extended to the situation with non-constant variance to handle rates with Poisson variation and possibly autocorrelated errors. The performance of these tests are studied via simulations and the tests are applied to U.S. prostate cancer incidence and mortality rates.

Conditional logistic regression with sandwich estimators: application to a meta-analysis.

Motivated by a meta-analysis of animal experiments on the effect of dietary fat and total caloric intake on mammary tumorigenesis, we explore the use of sandwich estimators of variance with conditional logistic regression. Classical conditional logistic regression assumes that the parameters are fixed effects across all clusters, while the sandwich estimator gives appropriate inferences for either fixed effects or random effects. However, inference using the standard Wald test with the sandwich estimator requires that each parameter is estimated using information from a large number of clusters. Since our example violates this condition, we introduce two modifications to the standard Wald test. First, we reduce the bias of the empirical variance estimator (the middle of the sandwich) by using standardized residuals. Second, we approximately account for the variance of these estimators by using the t-distribution instead of the normal distribution, where the degrees of freedom are estimated using Satterthwaite's approximation. Through simulations, we show that these sandwich estimators perform almost as well as classical estimators when the true effects are fixed and much better than the classical estimators when the true effects are random. We achieve simulated nominal coverage for these sandwich estimators even when some parameters are estimated from a small number of clusters.

Evaluation of alternative approaches to assign nutrient values to food groups in food frequency questionnaires.

Although every food frequency questionnaire (FFQ) requires a nutrient database to produce nutrient intake estimates, it is often unclear how a particular database has been generated. Moreover, alternative methods for constructing a database have not been rigorously evaluated. Using 24-hour recalls from the 1994-1996 Continuing Survey of Food Intake by Individuals, the authors categorized 5,261 individual foods reported by 10,019 adults into 170 food groups consistent with line items on an FFQ. These food groups were used to generate 10 potential nutrient databases for a FFQ that varied by whether the authors 1) used means or medians, 2) did or did not consider age, 3) incorporated collapsing strategies for small age-gender-portion size cells, 4) excluded outliers in a regression, and 5) used weighted median nutrient density x age-gender-portion size-specific median gram weights (Block method). Mean error, mean squared error, and mean absolute error were calculated and compared across methods, with error being the difference in total observed (from recalls for each individual) and total estimated intake (from each of the 10 methods) for seven nutrients. Mean methods for assigning nutrients to food groups were superior to median approaches for all measurements. Among the mean methods, no single variation was consistently better.

A comparison of two phase I trial designs.

Phase I cancer chemotherapy trials are designed to determine rapidly the maximum tolerated dose of a new agent for further study. A recently proposed Bayesian method, the continual reassessment method, has been suggested to offer an improvement over the standard design of such trials. We find the previous comparisons did not completely address the relative performance of the designs as they would be used in practice. Our results indicate that with the continual reassessment method, more patients will be treated at very high doses and the trials will take longer to complete. We offer some suggested improvements to both the standard design and the Bayesian method.

Covariate measurement error adjustment for matched case-control studies.

We propose a conditional scores procedure for obtaining bias-corrected estimates of log odds ratios from matched case-control data in which one or more covariates are subject to measurement error. The approach involves conditioning on sufficient statistics for the unobservable true covariates that are treated as fixed unknown parameters. For the case of Gaussian nondifferential measurement error, we derive a set of unbiased score equations that can then be solved to estimate the log odds ratio parameters of interest. The procedure successfully removes the bias in naive estimates, and standard error estimates are obtained by resampling methods. We present an example of the procedure applied to data from a matched case-control study of prostate cancer and serum hormone levels, and we compare its performance to that of regression calibration procedures.

Empirical evidence of correlated biases in dietary assessment instruments and its implications.

Multiple-day food records or 24-hour recalls are currently used as "reference" instruments to calibrate food frequency questionnaires (FFQs) and to adjust findings from nutritional epidemiologic studies for measurement error. The common adjustment is based on the critical requirements that errors in the reference instrument be independent of those in the FFQ and of true intake. When data on urinary nitrogen level, a valid reference biomarker for nitrogen intake, are used, evidence suggests that a dietary report reference instrument does not meet these requirements. In this paper, the authors introduce a new model that includes, for both the FFQ and the dietary report reference instrument, group-specific biases related to true intake and correlated person-specific biases. Data were obtained from a dietary assessment validation study carried out among 160 women at the Dunn Clinical Nutrition Center, Cambridge, United Kingdom, in 1988-1990. Using the biomarker measurements and dietary report measurements from this study, the authors compare the new model with alternative measurement error models proposed in the literature and demonstrate that it provides the best fit to the data. The new model suggests that, for these data, measurement error in the FFQ could lead to a 51% greater attenuation of true nutrient effect and the need for a 2.3 times larger study than would be estimated by the standard approach. The implications of the results for the ability of FFQ-based epidemiologic studies to detect important diet-disease associations are discussed.

Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires : the Eating at America's Table Study.

Researchers at the National Cancer Institute developed a new cognitively based food frequency questionnaire (FFQ), the Diet History Questionnaire (DHQ). The Eating at America's Table Study sought to validate and compare the DHQ with the Block and Willett FFQs. Of 1,640 men and women recruited to participate from a nationally representative sample in 1997, 1,301 completed four telephone 24-hour recalls, one in each season. Participants were randomized to receive either a DHQ and Block FFQ or a DHQ and Willett FFQ. With a standard measurement error model, correlations for energy between estimated truth and the DHQ, Block FFQ, and Willett FFQ, respectively, were 0.48, 0.45, and 0.18 for women and 0.49, 0.45, and 0.21 for men. For 26 nutrients, correlations and attenuation coefficients were somewhat higher for the DHQ versus the Block FFQ, and both were better than the Willett FFQ in models unadjusted for energy. Energy adjustment increased correlations and attenuation coefficients for the Willett FFQ dramatically and for the DHQ and Block FFQ instruments modestly. The DHQ performed best overall. These data show that the DHQ and the Block FFQ are better at estimating absolute intakes than is the Willett FFQ but that, after energy adjustment, all three are more comparable for purposes of assessing diet-disease risk.

Design and serendipity in establishing a large cohort with wide dietary intake distributions : the National Institutes of Health-American Association of Retired Persons Diet and Health Study.

In 1995-1996, the authors mailed a food frequency questionnaire to 3.5 million American Association of Retired Persons members who were aged 50-69 years and who resided in one of six states or two metropolitan areas with high-quality cancer registries. In establishing a cohort of 567,169 persons (340,148 men and 227,021 women), the authors were fortunate in that a less-than-anticipated baseline response rate (threatening inadequate numbers of respondents in the intake extremes) was offset by both a shifting and a widening of the intake distributions among those who provided satisfactory data. Reported median intakes for the first and fifth intake quintiles, respectively, were 20.4 and 40.1 (men) and 20.1 and 40.0 (women) percent calories from fat, 10.3 and 32.0 (men) and 8.7 and 28.7 (women) g per day of dietary fiber, 3.1 and 11.6 (men) and 2.8 and 11.3 (women) servings per day of fruits and vegetables, and 20.7 and 156.8 (men) and 10.5 and 97.0 (women) g per day of red meat. After 5 years of follow-up, the cohort is expected to yield nearly 4,000 breast cancers, more than 10,000 prostate cancers, more than 4,000 colorectal cancers, and more than 900 pancreatic cancers. The large size and wide intake range of the cohort will provide ample power for examining a number of important diet and cancer hypotheses.