Conducting and Reporting Randomized Clinical Trials: Important Recent Updates
Arthur M. Nezu, Drexel University
Randomized clinical trials (RCTs) are considered the gold-standard method for assessing the relative efficacy of a given medical or psychosocial intervention. However, recent changes have occurred in the manner in which researchers are now required to report the findings of RCTs in both medical and psychological journals. More importantly, these changes have a significant impact on the manner in which we need to design and conduct RCTs, as well as how to analyze and interpret the data emanating from these investigations. Much of these changes emerge from the recently adopted Consolidated Standards of Reporting Trials (CONSORT) guidelines, as well resulting professional task force reports (e.g., Society for Behavioral Medicine). For example, do you know that you need to register a clinical trial prior to its implementation? The first half of this session will present such guidelines and discuss their implications for CBT researchers. The second half will be very hands-on-therefore, attendees are encouraged to bring in their own studies that can serve as examples to discuss and demonstrate how to apply the "nitty-gritty" implied by such guidelines. In addition, a special emphasis will be made with regard to designing and conducting treatment integrity or fidelity protocols (i.e., therapist adherence and competence evaluations).
You will learn:
How to adopt the CONSORT guidelines to psychosocial interventions;
How to develop a basic treatment integrity protocol for your own research study;
How to register their own clinical trials as required by the government.
Recommended Readings: Boutron, I. et al. (2008). Extending the CONSORT statement to randomized trials of nonpharmacologic treatment: Explanation and elaboration. Annals of Internal Medicine, 148, 295-309. o Nezu, A. M. & Nezu, C. M. (2008). Ensuring treatment integrity. In A. M. Nezu & C. M. Nezu (Eds.), Evidence-based outcome research: A practical guide to conducting randomized clinical trials for psychosocial interventions (pp. 263-281). New York: Oxford University Press. o Perepletchikova, F. et al. (2009). Barriers to implementing treatment integrity procedures: Survey of treatment outcome researchers. Journal of Consulting and Clinical Psychology, 77, 212-218. o Trudeau, K. J. et al. (2008). Explanation of the CONSORT statement with application to psychosocial interventions. In A. M. Nezu & C. M. Nezu (Eds.), Evidence-based outcome research: A practical guide to conducting randomized clinical trials for psychosocial interventions (pp. 25-44). New York: Oxford University Press.
Models for Infrequent Outcomes: A Tutorial on Count Regression and Zero-
David C. Atkins, University of Washington
Robert J. Gallop, West Chester University
Some of the behaviors that we care most about as clinical researchers are also infrequent. Suicide attempts, self-harm, intimate partner violence, criticism, and drug use are all critically important mental health behaviors but will also tend to have strongly skewed distributions, except in specially chosen samples. These data all share something in common: They are typically counts or rates of behaviors (e.g., number of instances of self-harm in the last month, rate of drug use). Count data are often analyzed as if they were continuous data, but they have certain properties that strongly limit the appropriateness of ordinary least-squares (OLS) regression. Importantly, counts or rates are nonnegative integers, and because of this, the distributions of count variables are often highly skewed with a large stack of zeroes (e.g., many individuals report no instances of self-harm). These "zero-inflated" distributions suggest two distinct types of individuals and research questions: (a) those who have never engaged in the behavior (versus those who have), and (b) among individual who have engaged in the behavior, there are those who report higher vs. lower amounts. Oftentimes, we are interested in predictors related to each of these qualities of the data (e.g., examining predictors of those who have ever made a suicide attempt vs. those who have not, but also examining predictors of the number of suicide attempts for those who have made any).
This AMASS will present a tutorial on statistical methods for count data, called Poisson and Negative Binomial regressions. In addition, we will present extensions of these models to distributions with many zeroes, called zero-inflated Poisson (ZIP) and Negative Binomial (ZINB) regression models. These latter models simultaneously model the presence/absence of the dependent variable as well as the amount of the dependent variable when present. The methods will be introduced using several examples from the presenter's own research, and the data and computer code to run the example analyses in R and SAS will be provided (with some discussion of SPSS and Mplus). Finally, extensions to longitudinal and clustered data using generalized estimating equations (GEE) and generalized linear mixed models (GLMM) will be introduced, and practical advice will be given to assist researchers in using these tools with their data.
You will learn:
To define characteristics of count variables and why ordinary least-squares regression is not an appropriate model;
To describe overdispersion-what it is, sources of overdispersion, and why it is an issue that needs to be corrected in count regression;
To correctly interpret the output of count regression models, including Poisson, negative binomial, and zero-inflated models.
Recommended Readings: Atkins, D. C., & Gallop, R. J. (2007). Re-thinking how family researchers model infrequent outcomes: A tutorial on count regression and zero-inflated models. Journal of Family Psychology, 21, 726-735. o Gardner, W., Mulvey, E. P., & Shaw, E. C. (1995). Regression analyses of counts and rates: Poisson, over-dispersed Poisson, and negative binomial models. Psychological Bulletin, 118, 392-404. o Hedeker, D., & Gibbons, R. D. (2006). Mixed-effects regression models for counts (Chapter 12 in Longitudinal Data Analysis). New York: Wiley. o Hilbe, J. M. (2007). Negative binomial regression. New York: Cambridge.