A Comparison of Analysis of Covariate-Adjusted Residuals and Analysis of Covariance

Various methods to control the influence of a covariate on a response variable are compared. These methods are ANOVA with or without homogeneity of variances (HOV) of errors and Kruskal–Wallis (K–W) tests on (covariate-adjusted) residuals and analysis of covariance (ANCOVA). Covariate-adjusted residuals are obtained from the overall regression line fit to the entire data set ignoring the treatment levels or factors. It is demonstrated that the methods on covariate-adjusted residuals are only appropriate when the regression lines are parallel and covariate means are equal for all treatments. Empirical size and power performance of the methods are compared by extensive Monte Carlo simulations. We manipulated the conditions such as assumptions of normality and HOV, sample size, and clustering of the covariates. The parametric methods on residuals and ANCOVA exhibited similar size and power when error terms have symmetric distributions with variances having the same functional form for each treatment, and covariates have uniform distributions within the same interval for each treatment. In such cases, parametric tests have higher power compared to the K–W test on residuals. When error terms have asymmetric distributions or have variances that are heterogeneous with different functional forms for each treatment, the tests are liberal with K–W test having higher power than others. The methods on covariate-adjusted residuals are severely affected by the clustering of the covariates relative to the treatment factors when covariate means are very different for treatments. For data clusters, ANCOVA method exhibits the appropriate level. However, such a clustering might suggest dependence between the covariates and the treatment factors, so makes ANCOVA less reliable as well.     

A recommended analysis for 2 × 2 crossover trials with baseline measurements

In many two‐period, two‐treatment (2 × 2) crossover trials, for each subject, a continuous response of interest is measured before and after administration of the assigned treatment within each period. The resulting data are typically used to test a null hypothesis involving the true difference in treatment response means. We show that the power achieved by different statistical approaches is greatly influenced by (i) the ‘structure’ of the variance–covariance matrix of the vector of within‐subject responses and (ii) how the baseline (i.e., pre‐treatment) responses are accounted for in the analysis. For (ii), we compare different approaches including ignoring one or both period baselines, using a common change from baseline analysis (which we advise against), using functions of one or both baselines as period‐specific or period‐invariant covariates, and doing joint modeling of the post‐baseline and baseline responses with corresponding mean constraints for the latter. Based on theoretical arguments and simulation‐based type I error rate and power properties, we recommend an analysis of covariance approach that uses the within‐subject difference in treatment responses as the dependent variable and the corresponding difference in baseline responses as a covariate. Data from three clinical trials are used to illustrate the main points. Copyright © 2014 John Wiley & Sons, Ltd.     

Choice of Baselines in Clinical Trials: A Simulation Study from Statistical Power Perspective

Multiple assessments of an efficacy variable are often conducted prior to the initiation of randomized treatments in clinical trials as baseline information. Two goals are investigated in this article, where the first goal is to investigate the choice of these baselines in the analysis of covariance (ANCOVA) to increase the statistical power, and the second to investigate the magnitude of power loss when a continuous efficacy variable is dichotomized to categorical variable as commonly reported the biomedical literature. A statistical power analysis is developed with extensive simulations based on data from clinical trials in study participants with end stage renal disease (ESRD). It is found that the baseline choices primarily depend on the correlations among the baselines and the efficacy variable, with substantial gains for correlations greater than 0.6 and negligible for less than 0.2. Continuous efficacy variables always give higher statistical power in the ANCOVA modeling and dichotomizing the efficacy variable generally decreases the statistical power by 25%, which is an important practicum in designing clinical trials for study sample size and realistically budget. These findings can be easily applied in and extended to other clinical trials with similar design.     

Nearly optimal covariate designs—Part I

We propose to discuss at length several examples from standard text books. All of these examples deal with analysis of covariance (ANCOVA) models and related analyses of data. We intend to capitalize on our understanding of optimal covariate designs (OCDs) in different ANCOVA models and re-visit these examples with a view to suggest optimal/nearly optimal designs for estimation of the covariate parameter(s). As we will see, for some examples our task is very much routine but for others, it is indeed a highly non trivial exercise.

?We intent to cover a total of six examples—divided in two parts. This is Part I—dealing with two examples.     

Optimal estimation in functional linear regression for sparse noise‐contaminated data

In this article, we propose a novel approach to fit a functional linear regression in which both the response and the predictor are functions. We consider the case where the response and the predictor processes are both sparsely sampled at random time points and are contaminated with random errors. In addition, the random times are allowed to be different for the measurements of the predictor and the response functions. The aforementioned situation often occurs in longitudinal data settings. To estimate the covariance and the cross‐covariance functions, we use a regularization method over a reproducing kernel Hilbert space. The estimate of the cross‐covariance function is used to obtain estimates of the regression coefficient function and of the functional singular components. We derive the convergence rates of the proposed cross‐covariance, the regression coefficient, and the singular component function estimators. Furthermore, we show that, under some regularity conditions, the estimator of the coefficient function has a minimax optimal rate. We conduct a simulation study and demonstrate merits of the proposed method by comparing it to some other existing methods in the literature. We illustrate the method by an example of an application to a real‐world air quality dataset. The Canadian Journal of Statistics 47: 524–559; 2019 © 2019 Statistical Society of Canada     

Power of analysis of covariance tests under conditional specification

The procedure-wise power functions of two strategies for balanced single-factor analysis of covariance in the presence of possibly unequal regression slopes are evaluated and illustrated. The strategies differ in the action to be taken following a re-^ jection by the preliminary test for equal slopes. The first strategy simply discards the covariate and respecifies the model as the one-way ANOVA model for testing factor effects. The second leaves the unequal slopes covariance model intact, but respecifies the factor effects hypothesis to address the factor level means adjusted to the sample average of the covariate. One additional strategy, that of testing factor effects only if the preliminary slopes test does not reject, is included for comparison purposes. Computation of the power functions requires extensive use of the results obtained in Hawkins and Han (1986) concerning the bivariate distributions of certain ratios of independent noncentral chi-square random variables.     

An alternative to the analysis of covariance for comparing randomized treatments

The problem of comparing mean responses for several treatments applied to a common population is considered. The analysis of co-variance ‘ANCOVA’ is frequently used to take advantage of covariate information in this setting, but in many cases ANCOVA's assumption of parallel regression functions precludes the use of ANCOVA. In this paper, an alternative method is developed which does not make this assumption     

Issues for stratified randomization based on a factor derived from a continuous baseline variable

Stratified randomization based on the baseline value of the primary analysis variable is common in clinical trial design. We illustrate from a theoretical viewpoint the advantage of such a stratified randomization to achieve balance of the baseline covariate. We also conclude that the estimator for the treatment effect is consistent when including both the continuous baseline covariate and the stratification factor derived from the baseline covariate. In addition, the analysis of covariance model including both the continuous covariate and the stratification factor is asymptotically no less efficient than including either only the continuous baseline value or only the stratification factor. We recommend that the continuous baseline covariate should generally be included in the analysis model. The corresponding stratification factor may also be included in the analysis model if one is not confident that the relationship between the baseline covariate and the response variable is linear. In spite of the above recommendation, one should always carefully examine relevant historical data to pre-specify the most appropriate analysis model for a perspective study.     

Comparison of Statistical Methods for Pretest–Posttest Designs in Terms of Type I Error Probability and Statistical Power

The pretest–posttest design is widely used to investigate the effect of an experimental treatment in biomedical research. The treatment effect may be assessed using analysis of variance (ANOVA) or analysis of covariance (ANCOVA). The normality assumption for parametric ANOVA and ANCOVA may be violated due to outliers and skewness of data. Nonparametric methods, robust statistics, and data transformation may be used to address the nonnormality issue. However, there is no simultaneous comparison for the four statistical approaches in terms of empirical type I error probability and statistical power. We studied 13 ANOVA and ANCOVA models based on parametric approach, rank and normal score-based nonparametric approach, Huber M-estimation, and Box–Cox transformation using normal data with and without outliers and lognormal data. We found that ANCOVA models preserve the nominal significance level better and are more powerful than their ANOVA counterparts when the dependent variable and covariate are correlated. Huber M-estimation is the most liberal method. Nonparametric ANCOVA, especially ANCOVA based on normal score transformation, preserves the nominal significance level, has good statistical power, and is robust for data distribution.     

Impact of missing data on type 1 error rates in non‐inferiority trials

In this paper, a simulation study is conducted to systematically investigate the impact of different types of missing data on six different statistical analyses: four different likelihood‐based linear mixed effects models and analysis of covariance (ANCOVA) using two different data sets, in non‐inferiority trial settings for the analysis of longitudinal continuous data. ANCOVA is valid when the missing data are completely at random. Likelihood‐based linear mixed effects model approaches are valid when the missing data are at random. Pattern‐mixture model (PMM) was developed to incorporate non‐random missing mechanism. Our simulations suggest that two linear mixed effects models using unstructured covariance matrix for within‐subject correlation with no random effects or first‐order autoregressive covariance matrix for within‐subject correlation with random coefficient effects provide well control of type 1 error (T1E) rate when the missing data are completely at random or at random. ANCOVA using last observation carried forward imputed data set is the worst method in terms of bias and T1E rate. PMM does not show much improvement on controlling T1E rate compared with other linear mixed effects models when the missing data are not at random but is markedly inferior when the missing data are at random. Copyright © 2009 John Wiley & Sons, Ltd.     

Analysis of covariance under inverse Gaussian model

This paper considers the problem of analysis of covariance (ANCOVA) under the assumption of inverse Gaussian distribution for response variable. We develop the essential methodology for estimating the model parameters via maximum likelihood method. The general form of the maximum likelihood estimator is obtained in color closed form. Adjusted treatment effects and adjusted covariate effects are given, too. We also provide the asymptotic distribution of the proposed estimators. A simulation study and a real world application are also performed to illustrate and evaluate the proposed methodology.     

The two-period crossover design with baseline measurements

We consider a two-period crossover study in which each patients measured on the response variable at the start as well as at the end of both periods. We examine models in which the carryover effect at the start of the second period may be different from the carryover effect at the end, and in which the correlations between observations decrease as a function of the time between them.

In trials with a relatively short washout period, we recommend that the second baseline measurement not be incorporated into the analysis and that the data be evaluated by analysis of covariance, with the difference between the post-treatment values as the response variable and the first period's baseline value as the covariate. The absence of carryover effects must be assumed.

When the washout period is moderately long (comparable in length to either treatment period), the preferred analysis for a difference between direct treatment effects will again generally be based on the differences between post-treatment values. An analysis based on changes from baseline would, under certain assumptions about the form of the variance-covariance matrix, be preferred only for quite long washout periods and large correlations between observations. Even then, the efficiency of the test for equality of direct effects is improved if the difference between the baseline values is used as the covariate.     

A Bayesian conditional model for bivariate mixed ordinal and skew continuous longitudinal responses using quantile regression

In this paper, we develop a conditional model for analyzing mixed bivariate continuous and ordinal longitudinal responses. We propose a quantile regression model with random effects for analyzing continuous responses. For this purpose, an Asymmetric Laplace Distribution (ALD) is allocated for continuous response given random effects. For modeling ordinal responses, a cumulative logit model is used, via specifying a latent variable model, with considering other random effects. Therefore, the intra-association between continuous and ordinal responses is taken into account using their own exclusive random effects. But, the inter-association between two mixed responses is taken into account by adding a continuous response term in the ordinal model. We use a Bayesian approach via Markov chain Monte Carlo method for analyzing the proposed conditional model and to estimate unknown parameters, a Gibbs sampler algorithm is used. Moreover, we illustrate an application of the proposed model using a part of the British Household Panel Survey data set. The results of data analysis show that gender, age, marital status, educational level and the amount of money spent on leisure have significant effects on annual income. Also, the associated parameter is significant in using the best fitting proposed conditional model, thus it should be employed rather than analyzing separate models.     

Patient-specific dose finding in phase I clinical trials

This article proposes an extension of the continual reassessment method to determine the maximum tolerated dose (MTD) in the presence of patients' heterogeneity in phase I clinical trials. To start with a simple case, we consider the covariate as a binary variable representing two groups of patients. A logistic regression model is used to establish the dose–response relationship and the design is based on the Bayesian framework. Simulation studies for six plausible dose–response scenarios show that the proposed design is likely to determine the MTD more accurately than the design that does not take covariate into consideration.     

An efficient analysis of covariance model for crossover thorough QT studies with period‐specific pre‐dose baselines

Baseline adjustment is an important consideration in thorough QT studies for non‐antiarrhythmic drugs. For crossover studies with period‐specific pre‐dose baselines, we propose a by‐time‐point analysis of covariance model with change from pre‐dose baseline as response, treatment as a fixed effect, pre‐dose baseline for current treatment and pre‐dose baseline averaged across treatments as covariates, and subject as a random effect. Additional factors such as period and sex should be included in the model as appropriate. Multiple pre‐dose measurements can be averaged to obtain a pre‐dose‐averaged baseline and used in the model. We provide conditions under which the proposed model is more efficient than other models. We demonstrate the efficiency and robustness of the proposed model both analytically and through simulation studies. The advantage of the proposed model is also illustrated using the data from a real clinical trial. Copyright © 2014 John Wiley & Sons, Ltd.     

Cox Regression with Incomplete Covariate Measurements using the EM-algorithm

Ibrahim (1990) used the EM-algorithm to obtain maximum likelihood estimates of the regression parameters in generalized linear models with partially missing covariates. The technique was termed EM by the method of weights. In this paper, we generalize this technique to Cox regression analysis with missing values in the covariates. We specify a full model letting the unobserved covariate values be random and then maximize the observed likelihood. The asymptotic covariance matrix is estimated by the inverse information matrix. The missing data are allowed to be missing at random but also the non-ignorable non-response situation may in principle be considered. Simulation studies indicate that the proposed method is more efficient than the method suggested by Paik & Tsai (1997). We apply the procedure to a clinical trials example with six covariates with three of them having missing values.     

Empirical evaluation of the implementation of the EMA guideline on missing data in confirmatory clinical trials: Specification of mixed models for longitudinal data in study protocols

In confirmatory clinical trials, the prespecification of the primary analysis model is a universally accepted scientific principle to allow strict control of the type I error. Consequently, both the ICH E9 guideline and the European Medicines Agency (EMA) guideline on missing data in confirmatory clinical trials require that the primary analysis model is defined unambiguously. This requirement applies to mixed models for longitudinal data handling missing data implicitly. To evaluate the compliance with the EMA guideline, we evaluated the model specifications in those clinical study protocols from development phases II and III submitted between 2015 and 2018 to the Ethics Committee at Hannover Medical School under the German Medicinal Products Act, which planned to use a mixed model for longitudinal data in the confirmatory testing strategy. Overall, 39 trials from different types of sponsors and a wide range of therapeutic areas were evaluated. While nearly all protocols specify the fixed and random effects of the analysis model (95%), only 77% give the structure of the covariance matrix used for modeling the repeated measurements. Moreover, the testing method (36%), the estimation method (28%), the computation method (3%), and the fallback strategy (18%) are given by less than half the study protocols. Subgroup analyses indicate that these findings are universal and not specific to clinical trial phases or size of company. Altogether, our results show that guideline compliance is to various degrees poor and consequently, strict type I error rate control at the intended level is not guaranteed.     

A computationally simple bivariate survival estimator for efficacy and safety

Both treatment efficacy and safety are typically the primary endpoints in Phase II, and even in some Phase III, clinical trials. Efficacy is frequently measured by time to response, death, or some other milestone event and thus is a continuous, possibly censored, outcome. Safety, however, is frequently measured on a discrete scale; in Eastern Cooperative Oncology Group clinical trial E2290, it was measured as the number of weekly rounds of chemotherapy that were tolerable to colorectal cancer patients. For the joint analysis of efficacy and safety, we propose a non-parametric, computationally simple estimator for the bivariate survival function when one time-to-event is continuous, one is discrete, and both are subject to right-censoring. The bivariate censoring times may depend on each other, but they are assumed to be independent of both event times. We derive a closed-form covariance estimator for the survivor function which allows for inference to be based on any of several possible statistics of interest. In addition, we derive its covariance with respect to calendar time of analysis, allowing for its use in sequential studies.     

Robust Estimation and Hypothesis Testing of Linear Contrasts in Analysis of Covariance with Stochastic Covariates

Estimators of parameters are derived by using the method of modified maximum likelihood (MML) estimation when the distribution of covariate X and the error e are both non-normal in a simple analysis of covariance (ANCOVA) model. We show that our estimators are efficient. We also develop a test statistic for testing a linear contrast and show that it is robust. We give a real life example.