Issues for covariance analysis of dichotomous and ordered categorical data from randomized clinical trials and non-parametric strategies for addressing them.

Analysis of covariance is an effective method for addressing two considerations for randomized clinical trials. One is reduction of variance for estimates of treatment effects and thereby the production of narrower confidence intervals and more powerful statistical tests. The other is the clarification of the magnitude of treatment effects through adjustment of corresponding estimates for any random imbalances between the treatment groups with respect to the covariables. The statistical basis of covariance analysis can be either non-parametric, with reliance only on the randomization in the study design, or parametric through a statistical model for a postulated sampling process. For non-parametric methods, there are no formal assumptions for how a response variable is related to the covariables, but strong correlation between response and covariables is necessary for variance reduction. Computations for these methods are straightforward through the application of weighted least squares to fit linear models to the differences between treatment groups for the means of the response variable and the covariables jointly with a specification that has null values for the differences that correspond to the covariables. Moreover, such analysis is similarly applicable to dichotomous indicators, ranks or integers for ordered categories, and continuous measurements. Since non-parametric covariance analysis can have many forms, the ones which are planned for a clinical trial need careful specification in its protocol. A limitation of non-parametric analysis is that it does not directly address the magnitude of treatment effects within subgroups based on the covariables or the homogeneity of such effects. For this purpose, a statistical model is needed. When the response criterion is dichotomous or has ordered categories, such a model may have a non-linear nature which determines how covariance adjustment modifies results for treatment effects. Insight concerning such modifications can be gained through their evaluation relative to non-parametric counterparts. Such evaluation usually indicates that alternative ways to compare treatments for a response criterion with adjustment for a set of covariables mutually support the same conclusion about the strength of treatment effects. This robustness is noteworthy since the alternative methods for covariance analysis have substantially different rationales and assumptions. Since findings can differ in important ways across alternative choices for covariables (as opposed to methods for covariance adjustment), the critical consideration for studies with covariance analyses planned as the primary method for comparing treatments is the specification of the covariables in the protocol (or in an amendment or formal plan prior to any unmasking of the study.

Application of rank analysis of covariance methods to analysis of multiple anatomical regions with treatment for seborrheic dermatitis.

This paper presents the advantages of rank analysis of covariance in contrast to the Mantel-Haenszel procedure in the presence of a covariate. In this paper, data from a clinical trial with an indication for seborrheic dermatitis, which afflicts multiple anatomical regions, is presented. This paper presents analysis performed using both the Mantel-Haenszel procedure and rank analysis of covariance for separate anatomical regions, as well as for the combined anatomical regions. The analysis for the combined anatomical regions involves weighted sums over different strata.

Evaluation of alternative statistical models for crossover studies to demonstrate caffeine adjuvancy in the treatment of tension headache.

This paper discusses alternative statistical models for the analysis of six crossover studies to determine whether better relief of tension headache occurs from treatment with an analgesic plus caffeine (C) than with the analgesic alone (A) or with placebo (P). Each patient in these crossover studies randomly received a pair of distinct medications in such a way as to treat the first two of four headaches with the initial medication in the pair and to treat the third and fourth headaches with the last medication in the pair. In order to have greater power for the C versus A comparison, three times as many patients were randomly assigned to the A:C and C:A sequence groups as to the A:P, C:P, P:A, and P:C sequence groups. An issue of statistical interest for these crossover studies is the extent to which the possibility of unequal carryover effects of the three medications influences the roles of alternative models for data analysis and the interpretation of results. When carryover effects for all three medications are equal, univariate analysis of variance for the difference scores between the average response for the first two headaches and the average response for the third and fourth headaches for each patient provides nearly the same power for pairwise treatment comparisons as more comprehensive multivariate methods for all four headaches. However, for comparisons concerning carryover effects and for treatment comparisons with adjustment for carryover effects, multivariate methods encompassing all four headaches jointly can provide greater power than univariate analysis for difference scores, particularly when there is low intraclass correlation for responses within the same patient. Another noteworthy role for multivariate methods in situations with potentially unequal carryover effects is their capacity to clarify whether multiple types of carryover effects occur across the second, third, and fourth headaches in the respective sequence groups. Multivariate models with alternative specifications of carryover effects are fit to the data from the six crossover studies to compare C, A, and P by weighted least squares. The role of potential variation among centers is addressed in these analyses by the use of stratified proportional means over centers, means of center means, and means ignoring centers. The primary focus of attention in the respective analyses is the evaluation of treatment comparisons with and without adjustment for potential differences among carryover effects of the treatments.(ABSTRACT TRUNCATED AT 400 WORDS)

Evaluation of alternative statistical methods for linear model analysis to compare two treatments for 24-hour blood pressure response.

This paper evaluates alternative statistical methods for the analysis of 24-hour blood pressure data from a clinical trial which compares two treatments for hypertension. The primary objective of the study discussed here was to determine the time course for the blood pressure lowering effects of a test drug given once daily in the treatment of mild to moderate hypertension when compared with placebo. Thirty-five patients (24 on drug and 11 on placebo) were monitored for 24 hours at baseline and at two weeks post treatment, with diastolic blood pressure (DBP) measurements recorded at 22 time points within each 24-hour visit. The changes in DBP from baseline across the 22 time points are the response variables of interest. Various statistical methods for the assessment of treatment effects over the entire 24-hour dosing interval in a setting with small sample size are discussed and illustrated. The results from a special application of weighted least squares analysis of covariance, which employs a smoothed covariance matrix, support the hypothesis that a once daily dose of the drug significantly reduces DBP over the entire 24-hour dosing interval when compared with placebo. This method has the distinct advantage of enabling evaluation of treatment differences for the change in DBP from baseline at the 22 time points with the corresponding 22 baseline DBP as covariates simultaneously in a situation where the treatment group sample sizes are small.

Five-year follow-up of men with androgenetic alopecia treated with topical minoxidil.

Thirty-one men with androgenetic alopecia completed 4 1/2 to 5 years of therapy with 2% and 3% topical minoxidil. Hair regrowth with topical minoxidil tended to peak at 1 year with a slow decline in regrowth over subsequent years. However, at 4 1/2 to 5 years, maintenance of nonvellus hairs beyond that seen at baseline was still evident. Topical minoxidil appears to be effective in helping to maintain nonvellus hair growth in men with androgenetic alopecia.

A two-period crossover design for the comparison of two active treatments and placebo.

This paper discusses a two-period crossover design for the comparison of two active treatments (A and B) and placebo (P) for relief of recurrent symptoms of a chronic health disorder. It is based on blocks of ten patients for which the treatment sequences A:B and B:A are each assigned to three patients and the sequences A:P, P:A, B:P, and P:B are each assigned to one patient; thus, treatment periods have a 2:2:1 allocation for A, B, and P. The principal model for analysis of this design involves additive subject effects, period effects, and treatment effects. Analysis of within-patient differences provides an estimate of the comparison between active treatments with variance (2vw/7r) and an estimate of the comparison between an active treatment and placebo with variance (4vw/7r); here vw is the within-patient variance and r is the number of blocks of ten patients. Analyses which address carryover effects and treatment effects adjusted for carryover effects are also described. An example using simulated data on relief of recurrent gastrointestinal pain illustrates the methodology.

A review of some statistical methods for covariance analysis of categorical data.

Three general methods for covariance analysis of categorical data are reviewed and applied to an example from a clinical trial in rheumatoid arthritis. The three methods considered are randomization-model nonparametric procedures, maximum likelihood logistic regression, and weighted least squares analysis of correlated marginal functions. A fourth heuristic approach, the unweighted linear model analysis, is an approximate procedure but it is easy to implement. The assumptions and statistical issues for each method are discussed so as to emphasize philosophical differences between their rationales. Attention is given to computational differences, but it is shown that the methods lead to similar results for analogous problems. It is argued that the essential differences between the methods lie in their underlying assumptions and the generality of the conclusions which may be drawn.

Acetaminophen overdose. 662 cases with evaluation of oral acetylcysteine treatment.

Six hundred sixty-two consecutive patients with acetaminophen overdoses were evaluated. Those at risk on the basis of their acetaminophen blood levels, as plotted on the study nomogram, were treated with oral acetylcysteine. Statistically significant differences in severity of hepatic toxicity were observed between patients treated within 16 hours after ingestion and those treated between 16 and 24 hours after ingestion. No deaths occurred among patients treated within 24 hours of ingestion, except for one patient who was an alleged gunshot homicide. Seven percent of patients with plasma acetaminophen levels in the potentially toxic range and treated with acetylcysteine within ten hours of ingestion showed transient SGOT level elevations, whereas 29% of those treated between ten and 16 hours after ingestion and 62% of those treated between 16 and 24 hours after ingestion showed such transient toxicity. No consistent difference in hepatotoxicity could be demonstrated between those patients with a history of chronic alcohol use and those patients with no history of chronic alcohol use. Acute alcohol use resulted in less severe toxic reactions than in those patients without acute alcohol use.