Kenneth J. Koehler

An Empirical Investigation of Goodness-of-Fit Statistics for Sparse Multinomials

Abstract Traditional discussions of goodness-of-fit tests for multinomial data consider asymptotic chi-squared properties under the assumption that all expected cell frequencies become large. This condition is not always satisfied, however, and another asymptotic theory must be considered. For testing a specified simple hypothesis, Morris (1975) and Hoist (1972) gave conditions for the asymptotic normality of the Pearson and likelihood ratio statistics when both the sample size and number of cells become large (even if the expected cell frequencies remain small). Monte Carlo techniques are used to examine the applicability of the normal approximations for moderate sample sizes with moderate numbers of cells.

Cell-Cycle Arrest at G2/M and Growth Inhibition by Apigenin in Human Colon Carcinoma Cell Lines

Apigenin, a common dietary flavonoid, has been shown to induce cell cycle arrest in both epidermal and fibroblast cells and inhibit skin tumorigenesis in murine models. The present study assessed the influence of apigenin on cell growth and the cell cycle in the human colon carcinoma cell lines SW480, HT‐29, and Caco‐2. Treatment of each cell line with apigenin (0–80 μM) resulted in a dose‐dependent reduction in both cell number and cellular protein content, compared with untreated control cultures. DNA flow cytometric analysis indicated that treatment with apigenin resulted in G2/M arrest in all three cell lines in a time‐ and dose‐dependent manner. Apigenin treatment (80 μM) for 48 h produced maximum G2/M arrest of 64%, 42%, and 26% in SW480 cells, HT‐29 cells, and Caco‐2 cells, respectively, in comparison with control cells (15%). The proportion of S‐phase cells was not altered by apigenin treatment in each of the three cell lines. The G2/M arrest was reversible after 48 h of apigenin treatment in the most sensitive cell line SW480. The degree of G2/M arrest by apigenin was inversely correlated with the corresponding inhibition of cell growth measurements in all three cell lines (r = −0.626 to −0.917, P≤0.005). Moreover, an immune complex kinase assay demonstrated an inhibition of p34cdc2 kinase activity, a critical enzyme in G2/M transition, in each cell line after treatment with apigenin (50–80 μM). Western blot analyses indicated that both p34cdc2 and cyclin B1 proteins were also decreased after apigenin treatment. These results indicate that apigenin inhibits colon carcinoma cell growth by inducing a reversible G2/M arrest and that this arrest is associated, at least in part, with inhibited activity of p34cdc2 kinase and reduced accumulation of p34cdc2 and cyclin B1 proteins. Differences in induction of G2/M arrest by apigenin in the three colon carcinoma cell lines suggest that dietary apigenin may be differentially effective against tumors with specific mutational spectra. Mol. Carcinog. 28:102–110, 2000.

The Soy Isoflavones for Reducing Bone Loss (SIRBL) Study: a 3-y randomized controlled trial in postmenopausal women

BACKGROUND Our previous study indicated that soy protein with isoflavones lessened lumbar spine bone loss in midlife women. OBJECTIVE We examined the efficacy of isoflavones (extracted from soy protein) on bone mineral density (BMD) in nonosteoporotic postmenopausal women. We hypothesized that isoflavone tablets would spare BMD, with biological (age, body weight, serum 25-hydroxyvitamin D) and lifestyle (physical activity, dietary intake) factors modulating BMD loss. DESIGN Our double-blind, randomized controlled trial (36 mo) included healthy postmenopausal women (aged 45.8-65.0 y) with intent-to-treat (n = 224) and compliant (n = 208) analyses. Treatment groups consisted of a placebo control group and 2 soy isoflavone groups (80 compared with 120 mg/d); women received 500 mg calcium and 600 IU vitamin D(3). Outcomes included lumbar spine, total proximal femur, femoral neck, and whole-body BMD. RESULTS Analysis of variance for intent-to-treat and compliant (> or =80%) models, respectively, showed no treatment effect for spine (P = 0.46, P = 0.21), femur (P = 0.86, P = 0.46), neck (P = 0.17, P = 0.14), or whole-body (P = 0.86, P = 0.78) BMD. From baseline to 36 mo, BMD declined regardless of treatment. In intent-to-treat and compliant models, respectively, BMD decreases were as follows: spine (-2.08%, -1.99%), femur (-1.43%, -1.38%), neck (-2.56%, -2.51%), and whole body (-1.66%, -1.62%). Regression analysis (compliant model) indicated that age, whole-body fat mass, and bone resorption were common predictors of BMD change. After adjustment for these factors, 120 mg (compared with placebo) was protective (P = 0.024) for neck BMD. We observed no treatment effect on adverse events, endometrial thickness, or bone markers. CONCLUSION Our results do not show a bone-sparing effect of extracted soy isoflavones, except for a modest effect at the femoral neck. This trial was registered at clinicaltrials.gov as NCT00043745.

Goodness-of-Fit Tests for Log-Linear Models in Sparse Contingency Tables

Abstract The asymptotic normality of the likelihood ratio goodness-of-fit statistic is demonstrated for testing the fit of log-linear models with closed form maximum likelihood estimates in sparse contingency tables. Unlike the traditional chi-squared theory, the number of categories in the table increases as the sample size increases, but not all of the expected frequencies are required to become large. Some results of a small Monte Carlo study are presented. The traditional chi-squared approximation is reasonably accurate for the Pearson statistic for many sparse tables, but cases are presented for which it fails. The normal approximation can be much more accurate than the chi-squared approximation for the likelihood ratio statistic, but the bias of estimated moments is a potential problem for very sparse tables.

Strain specificity of the immune response of pigs following vaccination with various strains of porcine reproductive and respiratory syndrome virus.

The primary objective of the study was to determine strain specificity of the immune response of pigs following vaccination with selected strains of porcine reproductive and respiratory syndrome virus (PRRSV). The experimental design included five groups (I through V, six pigs per group) free of antibody for PRRSV at the beginning of the experiment (day 0). On day 0, groups III, IV, and V were vaccinated with attenuated versions of PRRSV strains 8, 9, and 14, respectively. On day 21, the immunity of group II (non-vaccinated/challenged controls) and groups III, IV, and V was challenged by exposing each pig to a composite of the virulent versions of these same three strains. On day 35, all pigs, including non-vaccinated/non-challenged pigs of group I, were necropsied. Lungs and selected lymph nodes were examined for lesions. Serum samples obtained at weekly intervals throughout the study and lung lavage fluids obtained at necropsy were tested for the presence of PRRSV and its strain identity. Before challenge the strain of PRRSV identified in the sera of vaccinated pigs was always that with which the particular pig had been vaccinated (i.e. homologous strain), whereas, with one exception, only heterologous strains were identified after challenge. This apparent strain exclusion as a result of vaccination was statistically significant (P = 0.004). The tendency for heterologous strains to predominate after challenge suggests that a pigs immune response to PRRSV has some degree of strain specificity. Whether this finding has any clinical relevance in regard to immunoprophylaxis remains to be determined.

Performance and prospects of Rag genes for management of soybean aphid

The soybean aphid, Aphis glycines Matsumura (Hemiptera: Aphididae), is an invasive insect pest of soybean [Glycine max (L.) Merr. (Fabaceae)] in North America, and it has led to extensive insecticide use in northern soybean‐growing regions there. Host plant resistance is one potential alternative strategy for managing soybean aphid. Several Rag genes that show antibiosis and antixenosis to soybean aphid have been recently identified in soybean, and field‐testing and commercial release of resistant soybean lines have followed. In this article, we review results of field tests with soybean lines containing Rag genes in North America, then present results from a coordinated regional test across several field sites in the north‐central USA, and finally discuss prospects for use of Rag genes to manage soybean aphids. Field tests conducted independently at multiple sites showed that soybean aphid populations peaked in late summer on lines with Rag1 or Rag2 and reached economically injurious levels on susceptible lines, whereas lines with a pyramid of Rag1 + Rag2 held soybean aphid populations below economic levels. In the regional test, aphid populations were generally suppressed by lines containing one of the Rag genes. Aphids reached putative economic levels on Rag1 lines for some site years, but yield loss was moderated, indicating that Rag1 may confer tolerance to soybean aphid in addition to antibiosis and antixenosis. Moreover, no yield penalty has been found for lines with Rag1, Rag2, or pyramids. Results suggest that use of aphid‐resistant soybean lines with Rag genes may be viable for managing soybean aphids. However, virulent biotypes of soybean aphid were identified before release of aphid‐resistant soybean, and thus a strategy for optimal deployment of aphid‐resistant soybean is needed to ensure sustainability of this technology.

Experimental evaluation of realized niche models for predicting responses of plant species to a change in environmental conditions

Abstract. We estimated, using logistic regression techniques, the realized niches of the four dominant species in an experimental marsh complex located in the Delta Marsh, Manitoba, Canada. These models were then used to predict the probability of occurrence of these species in selected elevation ranges when water levels were raised in 1985 either 0, 30 or 60 cm above the long-term normal water level. These realized-niche models were calculated using elevation and species data collected in 1980. After having been eliminated by two years of deep flooding, the emergent vegetation in this complex had been re-established during a drawdown beginning in either 1983 or 1984. Our hypothesis was that from 1985 to 1989 the frequencies of occurrence of species in selected elevation ranges would converge to their probabilities predicted from the 1980 logistic models. This was not borne out by our results. Actual frequencies and predicted probabilities of occurrence of a species were similar at best less than 40% and then mostly in the control (0 cm) treatment. The realized-niche models were not adequate to predict the distribution of emergents after an increase in water level in the short term because the emergent species did not migrate upslope. Emergent species in the medium and high treatments either (1) died out - Scolochloa festucacea and Scirpus lacustris - after 3 yr because they could not survive permanent flooding, (2) stayed where they were - Phragmites australis - because they were unable to move upslope through clonal growth, or (3) became more widespread - Typha glauca only because of the expansion of small local populations already established in 1985 in areas dominated formerly by other species.

Probability Plots and Distribution Curves for Assessing the Fit of Probability Models

Abstract Properties of P–P and Q–Q probability plots for visually assessing the fit of probability models are reviewed. It is shown how to construct families of curves for standardized P–P plots, which are useful for identifying viable alternatives to a proposed probability model. The effectiveness of such curves is illustrated with plots of real and simulated data. Percentiles of a related correlation test of goodness of fit are presented for testing the fit of either the normal, exponential, or Gumbel distribution with unknown location and scale parameters.

Chi-square tests for comparing vectors of proportions for several cluster samples

Test statistics are developed for comparing vectors of proportions obtained from several independent two–stage cluster samples. It is assumed that clusters are selected with probability proportional to size for each sample. Walds general method of constructing quadratic forms is used to obtain a large sample chi–square test. More easily evaluted chi–square tests are derived from the Dirichlet–multinnomial model. Corresponding goodness–of–fit test for the Dirichlet–multinomial model are also derived.

Modification of the Greenwood formula for correlated response times.

Life-table methodology for interval-censored survival times is used to estimate marginal survival probabilities from data consisting of independent cohorts of correlated responses. We restrict our attention to situations where response times within cohorts are exchangeable and the marginal survival distributions are the same for all individuals within any cohort. A modification of the Greenwood formula is derived for estimating variances and covariances of the estimates of the marginal survival probabilities. Simulation results show that this formula provides accurate adjustments for within-cohort correlations in finite samples. Comparisons with a nonparametric maximum likelihood estimator indicate negligible loss in efficiency from using simple life-table estimators of marginal survival probabilities that ignore the within-cohort correlations. Data from an angioplasty study where more than one procedure is performed on some patients are used to illustrate this methodology.