Quantification of Treatment Effect Modification on Both an Additive and Multiplicative Scale

BackgroundIn both observational and randomized studies, associations with overall survival are by and large assessed on a multiplicative scale using the Cox model. However, clinicians and clinical researchers have an ardent interest in assessing absolute benefit associated with treatments. In older patients, some studies have reported lower relative treatment effect, which might translate into similar or even greater absolute treatment effect given their high baseline hazard for clinical events.MethodsThe effect of treatment and the effect modification of treatment were respectively assessed using a multiplicative and an additive hazard model in an analysis adjusted for propensity score in the context of coronary surgery.ResultsThe multiplicative model yielded a lower relative hazard reduction with bilateral internal thoracic artery grafting in older patients (Hazard ratio for interaction/year = 1.03, 95%CI: 1.00 to 1.06, p = 0.05) whereas the additive model reported a similar absolute hazard reduction with increasing age (Delta for interaction/year = 0.10, 95%CI: -0.27 to 0.46, p = 0.61). The number needed to treat derived from the propensity score-adjusted multiplicative model was remarkably similar at the end of the follow-up in patients aged < = 60 and in patients >70.ConclusionsThe present example demonstrates that a lower treatment effect in older patients on a relative scale can conversely translate into a similar treatment effect on an additive scale due to large baseline hazard differences. Importantly, absolute risk reduction, either crude or adjusted, can be calculated from multiplicative survival models. We advocate for a wider use of the absolute scale, especially using additive hazard models, to assess treatment effect and treatment effect modification.     

Empirical best linear unbiased prediction in cultivar trials using factor-analytic variance-covariance structures

 Results of multi-environment trials to evaluate new plant cultivars may be displayed in a two-way table of genotypes by environments. Different estimators are available to fill the cells of such tables. It has been shown previously that the predictive accuracy of the simple genotype by environment mean is often lower than that of other estimators, e.g. least-squares estimators based on multiplicative models, such as the additive main effects multiplicative interaction (AMMI) model, or empirical best-linear unbiased predictors (BLUPs) based on a two-way analysis-of-variance (ANOVA) model. This paper proposes a method to obtain BLUPs based on models with multiplicative terms. It is shown by cross-validation using five real data sets (oilseed rape, Brassica napus L.) that the predictive accuracy of BLUPs based on models with multiplicative terms may be better than that of least-squares estimators based on the same models and also better than BLUPs based on ANOVA models. Received: 18 October 1997 / Accepted: 31 March 1998     

Multiplicative background correction for spotted microarrays to improve reproducibility

We propose a simple approach, the multiplicative background correction, to solve a perplexing problem in spotted microarray data analysis: correcting the foreground intensities for the background noise, especially for spots with genes that are weakly expressed or not at all. The conventional approach, the additive background correction, directly subtracts the background intensities from foreground intensities. When the foreground intensities marginally dominate the background intensities, the additive background correction provides unreliable estimates of the differential gene expression levels and usually presents M-A plots with fishtails or fans. Unreliable additive background correction makes it preferable to ignore the background noise, which may increase the number of false positives. Based on the more realistic multiplicative assumption instead of the conventional additive assumption, we propose to logarithmically transform the intensity readings before the background correction, with the logarithmic transformation symmetrizing the skewed intensity readings. This approach not only precludes the fishtails and fans in the M-A plots, but provides highly reproducible background-corrected intensities for both strongly and weakly expressed genes. The superiority of the multiplicative background correction to the additive one as well as the no background correction is justified by publicly available self-hybridization datasets.     

Analyses of two-way chronic studies

This paper proposes a method for analyzing tumor data from chronic studies when the experimental design includes combinations of two factors, for example, sex and dose. Both main effects and combined-effect (interaction) hypotheses are considered. A stratified log-rank statistic is presented for tests of no column or row (main) effects. The paper shows that when the numbers of animals in the cells are unequal and disproportional, the null distribution of the unstratified log-rank statistic does not have a chi-square distribution. Two simple models, additive and multiplicative, for representing the combined effect of row and column are considered under the proportional hazards model. A simple conservative statistic is proposed for testing the additivity of the row and column effects. A simulation experiment to examine the behavior of the null distribution of the combined-effect test statistic under the additive model and the power of the test against the multiplicative model is reported. The procedure is illustrated by analyzing mammary tumors induced by 7,12-dimethylbenz[a]anthracene (DMBA) in yellow and agouti F1 female mice from a laboratory experiment.     

Biological dosimetry of chernobyl cleanup workers: inclusion of data on age and smoking provides improved radiation dose estimates

We report the results of a study of chromosome translocations in 126 Russian subjects who participated in the cleanup activities at Chernobyl and another 53 subjects, from other places in Russia, who were not exposed at Chernobyl. In agreement with our earlier study, we find increased translocation frequencies among the exposed compared to Russian controls. We describe statistical methods for estimating the dose of ionizing radiation determined by scoring chromosome translocations found in circulating lymphocytes sampled several years after exposure. Two statistical models were fitted to the data. One model assumed that translocation frequencies followed an overdispersed Poisson distribution. The second model assumed that translocation frequencies followed a negative binomial distribution. In addition, the effects of radiation exposure were modeled as additive or as multiplicative to the effects of age and smoking history. We found that the negative binomial model fit the data better than the overdispersed Poisson model. We could not distinguish between the additive and the multiplicative model with our data. Individual dose estimates ranged from 0 (for 43 subjects) to 0.56 Gy (mean 0.14 Gy) under the multiplicative model and from 0 to 0.95 Gy (mean 0.15 Gy) under the additive model. Dose estimates were similar under the two models when the number of translocations was less than 4 per 100 cells. The additive model tended to estimate larger doses when the number of translocations was greater than 4 per 100 cells. We also describe a method for estimating upper 95% tolerance bounds for numbers of translocations in unexposed individuals. We found that inclusion of data on age and smoking history was important for dose estimation. Ignoring these factors could result in gross overestimation of exposures, particularly in older subjects who smoke.     

Stickbreaking: a novel fitness landscape model that harbors epistasis and is consistent with commonly observed patterns of adaptive evolution

In relating genotypes to fitness, models of adaptation need to both be computationally tractable and qualitatively match observed data. One reason that tractability is not a trivial problem comes from a combinatoric problem whereby no matter in what order a set of mutations occurs, it must yield the same fitness. We refer to this as the bookkeeping problem. Because of their commutative property, the simple additive and multiplicative models naturally solve the bookkeeping problem. However, the fitness trajectories and epistatic patterns they predict are inconsistent with the patterns commonly observed in experimental evolution. This motivates us to propose a new and equally simple model that we call stickbreaking. Under the stickbreaking model, the intrinsic fitness effects of mutations scale by the distance of the current background to a hypothesized boundary. We use simulations and theoretical analyses to explore the basic properties of the stickbreaking model such as fitness trajectories, the distribution of fitness achieved, and epistasis. Stickbreaking is compared to the additive and multiplicative models. We conclude that the stickbreaking model is qualitatively consistent with several commonly observed patterns of adaptive evolution.     

Linkage strategies for genetically complex traits. I. Multilocus models

In order to investigate linkage detection strategies for genetically complex traits, multilocus models of inheritance need to be specified. Here, two types of multilocus model are described: (1) a multiplicative model, representing epistasis (interaction) among loci, and (2) an additive model, which is shown to closely approximate genetic heterogeneity, which is characterized by no interlocus interaction. A ratio lambda R of risk for type R relatives that is compared with population prevalence is defined. For a single-locus model, lambda R - 1 decreases by a factor of two with each degree of relationship. The same holds true for an additive multilocus model. For a multiplicative (epistasis) model, lambda R - 1 decreases more rapidly than by a factor of two with degree of relationship. Examination of lambda R values for various classes of relatives can potentially suggest the presence of multiple loci and epistasis. For example, data for schizophrenia suggest multiple loci in interaction. It is shown in the second paper of this series that lambda R is the critical parameter in determining power to detect linkage by using affected relative pairs.     

A semiparametric response surface model for assessing drug interaction

Summary .   When multiple drugs are administered simultaneously, investigators are often interested in assessing whether the drug combinations are synergistic, additive, or antagonistic. Existing response surface models are not adequate to capture the complex patterns of drug interactions. We propose a two-component semiparametric response surface model with a parametric function to describe the additive effect of a combination dose and a nonparametric function to capture the departure from the additive effect. The nonparametric function is estimated using the technique developed in thin plate splines, and the pointwise bootstrap confidence interval for this function is constructed. The proposed semiparametric model offers an effective way of formulating the additive effect while allowing the flexibility of modeling a departure from additivity. Example and simulations are given to illustrate that the proposed model provides an excellent estimation for different patterns of interactions between two drugs.     

Analyzing Variety by Environment Data Using Multiplicative Mixed Models and Adjustments for Spatial Field Trend

The recommendation of new plant varieties for commercial use requires reliable and accurate predictions of the average yield of each variety across a range of target environments and knowledge of important interactions with the environment. This information is obtained from series of plant variety trials, also known as multi-environment trials (MET). Cullis, Gogel, Verbyla, and Thompson (1998) presented a spatial mixed model approach for the analysis of MET data. In this paper we extend the analysis to include multiplicative models for the variety effects in each environment. The multiplicative model corresponds to that used in the multivariate technique of factor analysis. It allows a separate genetic variance for each environment and provides a parsimonious and interpretable model for the genetic covariances between environments. The model can be regarded as a random effects analogue of AMMI (additive main effects and multiplicative interactions). We illustrate the method using a large set of MET data from a South Australian barley breeding program.     

On the Additive and Multiplicative Models of Relative Risk

Two popular models of absence of synergism in epidemiologic cohort studies are analyzed and compared. It is shown that the statistical concept of the union of independent events that traditionally has given rise to the “additive” model of relative risk can also generate the “multiplicative” model of relative risk. In fact, the same set of approximating conditions can be used to generate both models, which suggests a lack of identifiability under the traditional approach. An alternate approach is proposed in this paper. The new approach does not require the assumption that background risk factors are independent from causal agents of interest. The concept of “dose additivity” is discussed.     

Statistical Methods for Linkage Analysis of Complex Traits from High-Resolution Maps of Identity by Descent

A multilocus model for complex traits is described that generalizes the additive and multiplicative models and hence allows simultaneously for both heterogeneity and gene interaction (epistasis). Statistical methods of linkage analysis are discussed under the assumption that identity by descent data from a dense set of polymorphic markers are available. Three methods, single locus search, simultaneous search and conditional search, are described and compared.     

Sperm competition games: a comparison of loaded raffle models and their biological implications

Our main aim is to compare the additive model, due to Mesterton-Gibbons, and the multiplicative model, due to Parker, of sperm allocation under sperm competition, when other influences are treated in the same way. We first review these (and other) models and their foundations, leading to a generalization of the multiplicative model. Sperm is assumed to cost energy, and this constraint is incorporated differently in the two models. These give the same results in the random-roles situation when the males occupy roles (of first and second to mate) randomly: the number of sperm ejaculated in the favoured role is greater than that in the disfavoured role by an amount that depends on the effect of sperm limitation (i.e. the probability that there is insufficient sperm to ensure full fertility). If the latter is negligible, or the fertilization raffle fair, this difference is zero, as Parker found originally. In the constant roles situation (where males of a particular type always occupy the same role) the predictions differ: the additive model has the same predictions as in the random roles case, but the multiplicative model predicts that males of the type occupying the favoured role ejaculate less than males of the type occupying the disfavoured role, in accord with Parker's original conclusion. The fitnesses of the two types of male can be calculated in the multiplicative model: the fitness of the favoured male is usually higher, even if he has to expend more energy in "finding" a female, e.g. through fighting, etc. These conclusions relate to inter-male behaviour (i.e. of different male types), as distinct from intra-male behaviour (i.e. of a given male when in different roles). We analyse situations in which one male type has some probability of acting in its less usual role: calculations with varying amounts of sperm limitation are presented. It is found that the presence of a male of a different type has an effect on intra-male ejaculate behaviour, which also depends critically on the role usually occupied. We conclude that the multiplicative model is the more accurate model and provides more information. Some experimental data on sperm numbers are used to find the effects of sperm limitation. For species which conform to the loaded raffle model, sperm limitation typically has small or negligible effects: in this case, we argue that empiricists should look for equal ejaculates in the two roles when studying random role situations; when roles are occupied non-randomly average sperm expenditure should be greater by male types typically occupying the disfavoured role, but within a male type, expenditure should be greater in the role it typically occupies.     

The generalized multiplicative model for viability selection at multiple loci

Selection due to differential viability is studied in an n-locus two-allele model using a set indexation that allows the simplicity of the one-locus two-allele model to be carried to multi-locus models. The existence condition is analyzed for polymorphic equilibria with linkage equilibrium: Robbins' equilibria. The local stability condition is given for the Robbins' equilibria on the boundaries in the generalized non-epistatic selection regimes of Karlin and Liberman (1979). These generalized non-epistatic regimes include the additive selection model, the multiplicative selection model and the multiplicative interaction model, and their symmetric versions cover all the symmetric viability models.Research supported by grant no. 11-7805 from the Danish Natural Science Research Council, by NIH grant GM 28016, by a fellowship from the Research Foundation of Aarhus University, and by a visiting fellowship from the University of New England, N.S.W.     

Coherent modeling of longitudinal causal effects on binary outcomes

Analyses of biomedical studies often necessitate modeling longitudinal causal effects. The current focus on personalized medicine and effect heterogeneity makes this task even more challenging. Toward this end, structural nested mean models (SNMMs) are fundamental tools for studying heterogeneous treatment effects in longitudinal studies. However, when outcomes are binary, current methods for estimating multiplicative and additive SNMM parameters suffer from variation dependence between the causal parameters and the noncausal nuisance parameters. This leads to a series of difficulties in interpretation, estimation, and computation. These difficulties have hindered the uptake of SNMMs in biomedical practice, where binary outcomes are very common. We solve the variation dependence problem for the binary multiplicative SNMM via a reparameterization of the noncausal nuisance parameters. Our novel nuisance parameters are variation independent of the causal parameters, and hence allow for coherent modeling of heterogeneous effects from longitudinal studies with binary outcomes. Our parameterization also provides a key building block for flexible doubly robust estimation of the causal parameters. Along the way, we prove that an additive SNMM with binary outcomes does not admit a variation independent parameterization, thereby justifying the restriction to multiplicative SNMMs.     

Simple parametric and nonparametric models for excess and relative mortality

This paper studies two classes of hazard-rate-based models for the mortality in a group of individuals taking normal life expectancy into account. In a multiplicative hazard model, the estimate for the relative mortality generalises the standardised mortality ratio, and the adequacy of a model with constant relative mortality can be tested using a type of total time on test statistic. In an additive hazard model, continuous-time generalisations of a "corrected" survival curve and a "normal" survival curve are obtained, and the adequacy of a model with constant excess mortality can again be tested using a type of total time on test statistic. A model including both the multiplicative hazard model and the additive hazard model is briefly considered. The use of the models is illustrated on a set of data concerning survival after operation for malignant melanoma.     

Estimation and interpretation of incidence rate difference for recurrent events when the estimation model is misspecified

Recurrent events data are common in experimental and observational studies. It is often of interest to estimate the effect of an intervention on the incidence rate of the recurrent events. The incidence rate difference is a useful measure of intervention effect. A weighted least squares estimator of the incidence rate difference for recurrent events was recently proposed for an additive rate model in which both the baseline incidence rate and the covariate effects were constant over time. In this article, we relax this model assumption and examine the properties of the estimator under the additive and multiplicative rate models assumption in which the baseline incidence rate and covariate effects may vary over time. We show analytically and numerically that the estimator gives an appropriate summary measure of the time‐varying covariate effects. In particular, when the underlying covariate effects are additive and time‐varying, the estimator consistently estimates the weighted average of the covariate effects over time. When the underlying covariate effects are multiplicative and time‐varying, and if there is only one binary covariate indicating the intervention status, the estimator consistently estimates the weighted average of the underlying incidence rate difference between the intervention and control groups over time. We illustrate the method with data from a randomized vaccine trial.     

An improved null model for assessing the net effects of multiple stressors on communities

Ecological stressors (i.e., environmental factors outside their normal range of variation) can mediate each other through their interactions, leading to unexpected combined effects on communities. Determining whether the net effect of stressors is ecologically surprising requires comparing their cumulative impact to a null model that represents the linear combination of their individual effects (i.e., an additive expectation). However, we show that standard additive and multiplicative null models that base their predictions on the effects of single stressors on community properties (e.g., species richness or biomass) do not provide this linear expectation, leading to incorrect interpretations of antagonistic and synergistic responses by communities. We present an alternative, the compositional null model, which instead bases its predictions on the effects of stressors on individual species, and then aggregates them to the community level. Simulations demonstrate the improved ability of the compositional null model to accurately provide a linear expectation of the net effect of stressors. We simulate the response of communities to paired stressors that affect species in a purely additive fashion and compare the relative abilities of the compositional null model and two standard community property null models (additive and multiplicative) to predict these linear changes in species richness and community biomass across different combinations (both positive, negative, or opposite) and intensities of stressors. The compositional model predicts the linear effects of multiple stressors under almost all scenarios, allowing for proper classification of net effects, whereas the standard null models do not. Our findings suggest that current estimates of the prevalence of ecological surprises on communities based on community property null models are unreliable, and should be improved by integrating the responses of individual species to the community level as does our compositional null model.