This retrospective clinical study of patients with nonmetastatic prostate cancer demonstrates that patients transfused at the time of initial diagnosis or operation have a higher frequency of recurrence (54 percent) and death due to cancer (19 percent) than patients not receiving blood transfusions (recurrence rate 31 percent, p = 0.005; death rate 10 percent, p = 0.08). This difference is not explained by the transfused patients being older, having a less favorable clinical stage of disease, or less differentiated tumor histology. A multivariate analysis confirmed that the additional risk of dying from prostate cancer was 2.82-fold higher in transfused patients than in those not transfused. As in previous studies, the risk of recurrence may be greater in those receiving whole blood transfusions. Prospective studies of the association between perioperative blood transfusion and cancer recurrence are needed. For the present, prudent clinical practice should include avoidance of whole blood, fresh frozen plasma, and platelet transfusions and greater reliance on autologous blood transfusions.

Chuang and Agresti (1986) propose a monotone-scores model to analyze an ordinal pain data set from a pharmaceutical study where 4 analgesics were compared for their pain-relieving capabilities. In this paper, we generalize the model of Chuang and Agresti to incorporate an additional pre-test categorical variable. This generalization allows an ordering of the drugs conditional on a subject's pre-test categorical presentation. It also allows the identification of subjects who can most benefit from different drugs under comparison. These extensions are applied to part of a CNS study where an active drug was compared to a placebo and a pre-medication response to a placebo was available.

For analysis of data from a titration study, this paper adopts the idea of a life table analysis and utilizes a logistic linear dose-response model. The procedure uses iterated reweighted least-squares to provide parameter estimates. Computation of the asymptotic covariance of the life-table estimates of the response rates involves Greenwood's formula and a conditioning argument. A special case consists of examination of the covariance structure when no dropouts took place. Data from a titration study that investigated the efficacy of an anti-hypertensive compound illustrate the procedure.

The row effects and column effects models for two-way contingency tables have parameters for the row and column categories pertaining to the association between the variables. For classifications having ordered categories, it is often reasonable to assume that the association parameters have a corresponding ordering. This article proposes order-restricted estimates of the association parameters in these models. The maximum likelihood solution can be determined by the solution of a simple isotonic regression of some of the model sufficient statistics. The models are primarily log-linear in form and can be expressed in terms of odds ratios for 2×2 subtables consisting of adjacent rows and adjacent columns. For the order-restricted solution, these local log-odds ratios have uniform sign. Goodness-of-fit statistics for this solution are related to corresponding statistics for collapsed tables and to statistics for testing equality of sets of the parameters. The row effects model discussed in this article has been proposed by Haberman (1974), Simon (1974), and Goodman (1979), among others. This model contains parameters for the rows in the contingency table that describe the structure of the association and can be used to describe dependence in corresponding logit models. This article deals with applications of the model in which there is a monotonic relationship between the variables, in the sense that the population values of the local log-odds ratios are uniformly nonnegative or uniformly nonpositive. For instance, one might expect a nonnegative relationship for the data analyzed by Haberman (1974) and by Goodman (1979) on mental health and socioeconomic status, and a nonpositive relationship for the data analyzed in Section 2 of this article on age and severity of disturbances in dreams. By using the methods described in this article, one can obtain monotone estimates of the association parameters, which imply a monotone relationship between the variables. With this approach, one obtains a simpler description of the relationship, and better estimates, when the parameter scores truly are ordered.

For the pain data analysed previously by Cox and Chuang, this paper proposes a new model that assumes monotone scores for ordered response categories. This proposed model possesses several attractive features and allows a stochastic ordering of the drugs under comparison. Such a model also provides insight regarding the ordinal scale used to classify response. Estimation of the parameters in the model is obtained by use of BMDP3R.

Assuming a first-order Markov chain, we propose a structural model for the transition probabilities in vote intention. The proposed model utilizes the ordering among the categories representing vote intentions and carries the flavor of distance models. It also allows a stochastic ordering among distributions reflecting the extent of change. The model is easy to fit and provides a nice interpretation of the data. The model is applied to a panel study of vote intention acquired through six successive interviews before the 1940 Presidential election in Erie County, Ohio.

A class of association models for contingency tables has parameters that are sometimes interpreted as category scores. For classifications having ordered categories, it is often reasonable to assume that the score parameters have a corresponding ordering. This article proposes order-restricted estimates of score parameters in these models. For these estimates, the local log odds ratios have uniform sign. For the Bayesian approach proposed here, prior distributions can induce the order restriction, and prior beliefs reflecting strong association have the effect of moving the estimates away from the boundary of the restricted parameter space. The orderrestricted maximum likelihood solution is obtained in the limit as the prior standard deviation for the strength of association parameter grows unboundedly.

An expanded class of multiplicative-interaction (M-I) models is proposed for two-way contingency tables. These models a generalization of Goodman's association models, fill in the gap between the independence and the saturated models. Diagnostic rules based on a transformation of the data are proposed for the detection of such models. These rules, utilizing the singular value decomposition of the transformed data, are very easy to use. Maximum likelihood estimation is considered and the computational algorithms discussed. A data set from Goodman (1981) and another from Gabriel and Zamir (1979) are used to demostrate the diagnostic rules. Keywords: attached scorediagnostic ruleestimated basic termfixed basis termmatrix approximationmaximum likelihoodsingular value decompositionunattached score

Pseudo maximum likelihood estimation (PML) for the Dirich-let-multinomial distribution is proposed and examined in this pa-per. The procedure is compared to that based on moments (MM) for its asymptotic relative efficiency (ARE) relative to the maximum likelihood estimate (ML). It is found that PML, requiring much less computational effort than ML and possessing considerably higher ARE than MM, constitutes a good compromise between ML and MM. PML is also found to have very high ARE when an estimate for the scale parameter in the Dirichlet-multinomial distribution is all that is needed.

When samples are taken independently from I populations and the subjects classified into J categories, can the Pearson's chisquare statistic X2 testing the homogeneity model on the resulting I×J two-way table be decomposed into components familiar in the analysis of variance? Will the X2 testing the homogeneity model on tables derived by collapsing columns in the spirit of orthogonal comparisons in factorial experiments be asymptotically independent? The answers to both questions are generally negative. This paper gives a theoretical justification.

We study three approaches for the analysis of ordinal pain data. The first is essentially non-parametric, based on partitioning the goodness-of-fit statistic for testing the homogeneity model. The other two involve the modelling of quite different logit functions of the multinomial probabilities. We discuss and compare the characteristics of these three approaches using actual data from an analgesic trial.

Multiplicative-interaction (M-I) logit models are proposed for three-way IxJx2 contingency tables where the third variable constitutes a binary response. Models are derived by assigning unknown scores to the categories and forming product interactions from them. Asymptotic results under special sampling constraints are derived for maximum likelihood estimates and the goodness-of-fit statistics. The class of models proposed in this paper are found to be useful when no obvious scores are available. An example is included.

2-amino-5-bromo-6-phenyl-4-(3H)-pyrimidinone (ABPP) is one of the pyrimidinone compounds which induce interferon (IFN) in several animal species and have potent antivirus activities. ABPP has also been shown to exert growth inhibition of mouse B-16 melanoma and mammary carcinoma. We investigated the effectiveness of ABPP in the treatment of 7,12-dimethylbenz (α) anthracene (DMBA)-induced rat mammary cancer alone or in combination with tamoxifen. Fifty-two 50-day-old female Sprague Dawley rats were randomly divided into four groups after each rat received 25 mg of DMBA. The following treatments were started 90 days later for four weeks: Group I, 0.5 cc of 1% carboxymethyl cellulose (CMC) by intraperitoneal (ip) injection three times a week; Group II, 200 mg/kg ABPP in 0.5 cc of 1% CMC ip three times a week; Group III, 0.2 mg/kg tamoxifen in 0.5 cc peanut oil ip daily five days a week; Group IV the combination treatment of Groups II and III. Tumors were measured once every week. There were 1, 5, 6, and 12 tumors in Groups I, II, III, and IV rats, respectively, showing greater than 50% reduction in size. Numbers of new tumors developed during treatment were 10, 8, 10, and 2 (p < 0.01) in Groups I, II, III, and IV, respectively. There was no correlation between tumor response and the amount of interferon induced by ABPP. There was also no definite pattern of change in estrogen receptor content before and after ABPP treatment. Results from a logit analysis suggest that ABPP has antitumor activity against DMBAinduced rat mammary carcinoma about the same magnitude as tamoxifen. Also, the antitumor effect of the simultaneous application of ABPP and tamoxifen is additive and is probably better then either one alone.

Empirical Bayes methods are used to estimate cell probabi-lities under a multiplicative-Interaction model for a two-way contingency table. The methods assign uniform and normal priors with unknown variances to the main effects and the separable scores. A priori the analysis assumes exchangeability of sets of parameters. The unknown variance components are estimated empirically from the data via the EM algorithm as discussed by Laird (1978)and Dempster, Laird and Rubin (1977). An example Is Included.