Testing for treatment and interaction effects in semi-parametric analysis of covariance.

Semi-parametric regression models assume that the effects of covariates on the mean response are additive. We propose a test of additivity when there is one continuous covariate and a group indicator. At p fixed points, the differences of the within-group kernel estimates of the means are calculated, and the likelihood ratio test that the p differences have a constant mean is formed. The kernel bandwidth and the location of the p fixed points are chosen to give the test good power. Performance of the proposed test is compared with parametric and non-parametric tests of additivity. Published in 2001 by John Wiley & Sons, Ltd.

Modeling repeated count data subject to informative dropout.

In certain diseases, outcome is the number of morbid events over the course of follow-up. In epilepsy, e.g., daily seizure counts are often used to reflect disease severity. Follow-up of patients in clinical trials of such diseases is often subject to censoring due to patients dying or dropping out. If the sicker patients tend to be censored in such trials, estimates of the treatment effect that do not incorporate the censoring process may be misleading. We extend the shared random effects approach of Wu and Carroll (1988, Biometrics 44, 175-188) to the setting of repeated counts of events. Three strategies are developed. The first is a likelihood-based approach for jointly modeling the count and censoring processes. A shared random effect is incorporated to introduce dependence between the two processes. The second is a likelihood-based approach that conditions on the dropout times in adjusting for informative dropout. The third is a generalized estimating equations (GEE) approach, which also conditions on the dropout times but makes fewer assumptions about the distribution of the count process. Estimation procedures for each of the approaches are discussed, and the approaches are applied to data from an epilepsy clinical trial. A simulation study is also conducted to compare the various approaches. Through analyses and simulations, we demonstrate the flexibility of the likelihood-based conditional model for analyzing data from the epilepsy trial.

Quantitative insulin sensitivity check index: a simple, accurate method for assessing insulin sensitivity in humans.

Insulin resistance plays an important role in the pathophysiology of diabetes and is associated with obesity and other cardiovascular risk factors. The "gold standard" glucose clamp and minimal model analysis are two established methods for determining insulin sensitivity in vivo, but neither is easily implemented in large studies. Thus, it is of interest to develop a simple, accurate method for assessing insulin sensitivity that is useful for clinical investigations. We performed both hyperinsulinemic isoglycemic glucose clamp and insulin-modified frequently sampled iv glucose tolerance tests on 28 nonobese, 13 obese, and 15 type 2 diabetic subjects. We obtained correlations between indexes of insulin sensitivity from glucose clamp studies (SI(Clamp)) and minimal model analysis (SI(MM)) that were comparable to previous reports (r = 0.57). We performed a sensitivity analysis on our data and discovered that physiological steady state values [i.e. fasting insulin (I(0)) and glucose (G(0))] contain critical information about insulin sensitivity. We defined a quantitative insulin sensitivity check index (QUICKI = 1/[log(I(0)) + log(G(0))]) that has substantially better correlation with SI(Clamp) (r = 0.78) than the correlation we observed between SI(MM) and SI(Clamp). Moreover, we observed a comparable overall correlation between QUICKI and SI(Clamp) in a totally independent group of 21 obese and 14 nonobese subjects from another institution. We conclude that QUICKI is an index of insulin sensitivity obtained from a fasting blood sample that may be useful for clinical research.

Valid inference in random effects meta-analysis.

The standard approach to inference for random effects meta-analysis relies on approximating the null distribution of a test statistic by a standard normal distribution. This approximation is asymptotic on k, the number of studies, and can be substantially in error in medical meta-analyses, which often have only a few studies. This paper proposes permutation and ad hoc methods for testing with the random effects model. Under the group permutation method, we randomly switch the treatment and control group labels in each trial. This idea is similar to using a permutation distribution for a community intervention trial where communities are randomized in pairs. The permutation method theoretically controls the type I error rate for typical meta-analyses scenarios. We also suggest two ad hoc procedures. Our first suggestion is to use a t-reference distribution with k-1 degrees of freedom rather than a standard normal distribution for the usual random effects test statistic. We also investigate the use of a simple t-statistic on the reported treatment effects.

A simple permutation-type method for testing circular uniformity with correlated angular measurements.

A simple method is provided for testing uniformity on the circle that allows dependence among repeated angular measurements on the same subject. Our null hypothesis is that the distribution of repeated angles is unaffected by rotation. This null can be evaluated with any test of uniformity by using a null reference distribution obtained by simulation, where each subject's vector of angles is rotated by a random amount. A new weighted version of the univariate Rayleigh test of circular uniformity is proposed.

A multivariate test of interaction for use in clinical trials.

An important issue in clinical trials is whether the effect of treatment is essentially homogeneous as a function of baseline covariates. Covariates that have the potential for an interaction with treatment may be suspected on the basis of treatment mechanism or may be known risk factors, as it is often thought that the sickest patients may benefit most from treatment. If disease severity is more accurately determined by a collection of baseline covariates rather than a single risk factor, methods that examine each covariate in turn for interaction may be inadequate. We propose a procedure whereby treatment interaction is examined along a single severity index that is a linear combination of baseline covariates. Formally, we derive a likelihood ratio test based on the null beta0 = beta1 versus the alternative abeta0 = beta1, where X'beta(k) (k = 0, 1) corresponds to the severity index in arm k and X is a vector of baseline covariates. While our explicit test requires a Gaussian response, it can be readily implemented whenever the estimates of beta0,beta1 are approximately multivariate normal. For example, it is appropriate for large clinical trials where beta(k) is based on a logisitic or Cox regression of response on X.

Use of summary measures to adjust for informative missingness in repeated measures data with random effects.

We discuss how to apply the conditional informative missing model of Wu and Bailey (1989, Biometrics 45, 939-955) to the setting where the probability of missing a visit depends on the random effects of the primary response in a time-dependent fashion. This includes the case where the probability of missing a visit depends on the true value of the primary response. Summary measures for missingness that are weighted sums of the indicators of missed visits are derived for these situations. These summary measures are then incorporated as covariates in a random effects model for the primary response. This approach is illustrated by analyzing data collected from a trial of heroin addicts where missed visits are informative about drug test results. Simulations of realistic experiments indicate that these time-dependent summary measures also work well under a variety of informative censoring models. These summary measures can achieve large reductions in estimation bias and mean squared errors relative to those obtained by using other summary measures.

Repeated probit regression when covariates are measured with error.

This paper develops a model for repeated binary regression when a covariate is measured with error. The model allows for estimating the effect of the true value of the covariate on a repeated binary response. The choice of a probit link for the effect of the error-free covariate, coupled with normal measurement error for the error-free covariate, results in a probit model after integrating over the measurement error distribution. We propose a two-stage estimation procedure where, in the first stage, a linear mixed model is used to fit the repeated covariate. In the second stage, a model for the correlated binary responses conditional on the linear mixed model estimates is fit to the repeated binary data using generalized estimating equations. The approach is demonstrated using nutrient safety data from the Diet Intervention of School Age Children (DISC) study.

Bayesian monitoring of event rates with censored data.

A Bayesian approach to monitoring event rates with censored data is proposed. A Dirichlet prior for discrete time event probabilities is blended with discrete survival times to provide a posterior distribution that is a mixture of Dirichlets. Approximation of the posterior distribution via data augmentation is discussed. Practical issues involved in implementing the procedure are discussed and illustrated with a simulation of the single arm Cord Blood Transplantation Study where 6-month survival is monitored.

A generalized estimating equation approach for modeling random length binary vector data.

A common measure in clinical trials and epidemiologic studies is the number of events such as seizures, hospitalizations, or bouts of disease. Frequently, a binary measure of severity for each event is available but is not incorporated in the analysis. This paper proposes methodology for jointly modeling the number of events and the vector of correlated binary severity measures. Our formulation exploits the notion that a given covariate may affect both outcomes in a similar way. We functionally link the regression parameters for the counts and binary means and discuss a generalized estimating equation (GEE) approach for parameter estimation. We discuss conditions under which the proposed joint modeling approach provides marked gains in efficiency relative to the common procedure of simply modeling the counts, and we illustrate the methodology with epilepsy clinical trial data.

Effect of zidovudine and didanosine treatment on heart function in children infected with human immunodeficiency virus.

OBJECTIVES: Human immunodeficiency virus (HIV) infection in children can be complicated by the development of cardiac disease. Decreased left ventricular function has been temporally associated with the use of zidovudine (azidothymidine; AZT) in adults with HIV and has been associated with changes in cardiac muscle mitochondria in animal models. This study was done in an attempt to determine whether the cardiac disease is related to the antiretroviral therapy or to progressive HIV infection. METHODS: We retrospectively reviewed echocardiograms, clinical records, and laboratory data from 137 HIV-infected children who were being treated by the Pediatric Branch, National Cancer Institute, and who were receiving AZT or didanosine, both drugs, or no antiretroviral therapy. RESULTS: Despite correction of the echocardiographic results for HIV disease severity with markers such as CD4+ lymphocyte count, time since infection, mode of acquisition of HIV, and age, children who were treated with AZT had a lower average fractional shortening than those who were not treated with AZT (p < 0.00001). There was a nonlinear relation between days of AZT use and this There was a nonlinear relation between days of AZT use and this decrease in fractional shortening. The odds that a cardiomyopathy would develop was 8.4 times greater in children who had previously used AZT than in those who had never taken AZT (95% confidence interval, 1.7 to 42.0). Didanosine was not associated with the development of a cardiomyopathy. CONCLUSIONS: Treatment of HIV-infected children with AZT may be associated with the development of a cardiomyopathy; didanosine does not appear to increase the risk of cardiomyopathy. The continued use of AZT in a child in whom a cardiomyopathy develops should be carefully assessed, and all children receiving AZT should be followed by serial cardiac examination and echocardiograms.

Use of preventive interventions by persons infected with type-1 human immunodeficiency virus (HIV-1). The Pulmonary Complications of HIV Study Group.

Measures aimed at preventing complications and slowing progression of type-1 human immunodeficiency virus (HIV-1) can potentially reduce morbidity. Although little is known about the use of such measures, such data are critical for program planning. This study was performed to quantify the frequency and patterns of use for such interventions. We enrolled 1,171 persons infected with HIV, but without an acquired immunodeficiency syndrome (AIDS) defining diagnosis, in a multicenter prospective study of the pulmonary complications of HIV infection. Participants were homosexual/bisexual men, injection drug users (IDUs), or female sexual contacts of HIV-infected men. Centers were university-based and geographically dispersed across the United States. Standardized questionnaires were administered on entry and at three-month or six-month intervals; we correlated use of general and HIV-related preventive measures before entry and during the first three years in study with clinical/epidemiologic characteristics. Overall use of preventive interventions was low; only one third of study entrants had used such measures. Use was greatest among those with advanced HIV infection, but only half used preventive measures on entry; IDUs were less likely than homosexuals to use these services. Although use of interventions such as anti-Pneumocystis and antiretroviral agents increased during study participation, general measures such as pneumococcal vaccine and tuberculosis prophylaxis were used by less than 30% of those eligible for use. Among IDUs, cumulative use of these measures remained below 20% during the first three years of this study. We conclude that HIV-infected persons underuse preventive interventions, particularly general measures.(ABSTRACT TRUNCATED AT 250 WORDS)

Monitoring multi-armed trials.

We propose and discuss several methods of monitoring multi-armed trials comparing means or survival. These methods combine multiple comparison procedures such as Fisher's LSD, Newman-Keuls and Tukey's with monitoring boundaries such as those of O'Brien and Fleming and Lan and DeMets. Tables of boundaries are provided for the equal variance or equal censoring distribution case.

Monitoring pairwise comparisons in multi-armed clinical trials.

This paper proposes a method for monitoring multi-armed clinical trials on the basis of pairwise comparisons between arms. The set of pairwise test statistics is examined during the course of the trial in order to make decisions about hypotheses, continuation of treatment arms, and continuation of the trial. Strong control of the Type I error rate is achieved by modifying two-armed group sequential procedures of Pocock (1977, Biometrika 64, 191-199), O'Brien and Fleming (1979, Biometrics 35, 549-556), and Lan and DeMets (1983, Biometrika 70, 659-663) to multi-armed trials. In the fixed-sample situation, these methods reduce to either Dunnett's or Tukey's procedure for multiple comparisons. A simpler, more flexible approximation based on the Bonferroni inequality is suggested, as well as an analogue to a sequentially rejective procedure.

The effect of screening on some pretest-posttest test variances.

The clinical trial design in which the endpoint is measured both at baseline and at the end of the study is used in a variety of situations. For two-group designs, test such as the t test or analysis of covariance are commonly used to evaluate treatment efficacy. Often such pretest-posttest trials restrict participation to subjects with a baseline measurement of the endpoint in a certain range. A range may define a disease, or it may be thought that subjects with extreme measurements are more responsive to treatment. This paper examines the effect of screening on the analysis of covariance and t-test variances relative to the population (i.e., unscreened) variances. Bivariate normal and bivariate gamma distributions are assumed for the (pretest, posttest) measurements. Because the sample size required to detect a specified difference between treatment and control is proportional to the variance, the results have direct application to setting sample size.