Sponsoring Section/Society: ASA Biometrics Section
Session Slot: 10:30-12:20 Thursday
Estimated Audience Size: 300-350
AudioVisual Request: Overhead
Session Title: The Impact of Patient Compliance
The need to assess patient compliance in clinical trials involving medication is widely recognized. The failure of patients to adhere adequately to prescribed medication is an important medical problem. Studies have demonstrated the prevalence of poor adherence across all types of regimens and diseases. Treatment efficacy cannot adequately be evaluated without valid measures of patient adherence. The use of careful and sensitive instruments to measure adherence behavior in a clinical trial is important in developing new therapies.
Unless appropriate account is taken of compliance problems in
clinical trials, policies based on these trial findings may be
misguided. Moreover, overlooking compliance problems in the general
population may yield health care policy recommendations that are
impractical or ineffective in actual application.
Theme Session: No
Applied Session: Yes
Session Organizer: Ting, Lee Mei-Ling Harvard Medcal School
Address: Channing Laboratory/BWH Harvard Medical School 181 Longwood Avenue Boston, MA 02115
Session Timing: 110 minutes total (Sorry about format):
Opening Remarks by Chair - 5 or 0 minutes First Speaker - 30 minutes (or 25) Second Speaker - 30 minutes Third Speaker - 30 minutes Discussant - 10 minutes (or none) Floor Discusion - 10 minutes (or 5 or 15)
Session Chair: Sun, Jianguo University of Missouri
Address: Dept. of Statistics Univ. of Missouri 321 Math Sci Building Columbia, MO 65211
1. Sensitivity Analysis and Non Complance in Randomized Trials
Robins, James M., Harvard School of Public Health
Address: James M. Robins Room 3-823 Harvard School of Public Health 677 Huntington Avenue Boston, MA 02115
Abstract: In randomized trials with noncompliance, it is often of interest to try to estimate the effect of treatment had there been complete compliance. However the effect under full compliance is not identified without strong untestable aasumptions. As a consequence, three approaches are available. First, compute bounds for the effect under full compliance. Second quantify uncertainty in a prior distribution and do a Bayesian analysis. Third conduct a sensitivity analysis. This talk offers a new approach to sensitvity analysis and shows how it can recover both results obtained under the bound approach and the Bayesian approach, while offering greater flexibility and insight. Our appoach derives from a theorem stating that although the odds of becoming noncompliant at each time t conditional on the possibly unobserved ``outcome under complete compliance'' is not idenified, nonetheless given this ``selection bias'' odds as a function of t, the distribution of the ``outcome under complete compliance'' is nonparametrically just identified.Hence in a sensitivity analysis, we vary these ``selection bias'' odds.
2. Estimating the Effect of Observed Exposure in Experimental Settings
Goethebeur, Els, University of Ghent, Belgium
Address: Els Goetghebeur University of Ghent Department of Applied Mathematics and Informatics Krijgslaan 281-S9 9000 Ghent BELGIUM
Phone: 0032-(0)9- 264 4811
Fax: 0032-(0)9- 264 4995
Abstract: In their 1991 paper, Efron and Feldman model the relationship between potential placebo response, and the effect of a possible drug exposure. They allow explicitly for an interaction between placebo response and drug dose effect. In doing so they provide measures of effectiveness in complier subgroups (like the effect of full compliance for the subgroup of patients who were full compliers with their assigned treatment), and efficacy at different potential exposures (like the effect of full compliance exposure over the entire study population). These parameters are estimated under the assumption of some form of exchangeability between compliance with placebo and compliance with treatment. In a simulation, Albert and Demets (1994) demonstrate how strong the biases can be in such model, when the assumption of comparable patients in similar compliance subgroups on treatment and placebo are not fulfilled.
In this talk we show that the structural mean model (Robins, 1994, Goetghebeur and Lapp, 1997) can be used to estimate all the parameters of interest in the model of Albert and Demets, provided a baseline covariate can be identified which is associated with placebo response or treatment compliance, but which does not interact with the added treatment effect. These estimators are consistent even for selective noncompliance and noncomparable compliance groups between treatment and placebo. In return, they rely on the parametric description of the `causal' or structural relationships. Ocne this is correctly specified, the SMM model also allows to find the average relationship between treatment compliance and placebo response. We will move on to discuss sensitivity to this new level of structural assumptions.
3. The Impact of Patient Compliance on Steady State Pharmacokinetics
Wang, Wenping, Janssen Research Foundation
Hsuan, Francis, Temple University
Chow, Shein-Chung, Biostatistics and Data Management, Covance
Abstract: Physicians commonly prescribe drug products in a multiple dosage regimen for prolonged therapeutic activity. To study the effect of multiple dosing on drug concentration in blood, researchers often employ a deterministic model with the assumptions that drugs are administered at a fixed dosage, with fixed (usually constant) dosing intervals. In practice, as it is well known in the medical community, patients may not follow such a rigid schedule. Hence, two possible scenarios might occur: patients might not take the prescribed dosage, resulting in variable dosing size; or they might not adhere to the dosing schedule, resulting in irregular dosing times. This paper intends to lay out a probability framework to model these two types of non-compliance and consequently study their impact on the steady state pharmacokinetics.
When studying multiple dose pharmacokinetics, the principle of superposition is the key tool. In this paper, the principle of superposition in the presence of non-compliance is formulated generally as a recursive formula. With this formula, we are able to generalize the notion of steady state in multiple dose pharmacokinetics given non-compliance. Using the compliance models and the principle of superposition, important pharmacokinetic parameters are rigorously studied. Factors affecting the steady state trough concentration are characterized through a simulation study. The relationship between the compliance index and the average concentration at steady state is established. This result generalize the classic result about the equality between the single dose area under the curve (AUC) and the multiple dose AUC. Using theophylline (an antiasthma agent) as an example, we demonstrate that non-compliance causes the drug concentration time curve to exhibit an increased fluctuation. The increase in fluctuation due to non-compliance cannot be explained with the use of the classical deterministic multiple dose model. A practical question - the number of doses needed to reach steady state, is answered both asymptotically and in a small sample study.
Discussant: Zelen, Marvin Dana Farber Cancer Institute and Harvard School of Public Health
Address: Dana Farber Cancer Institute Department of Biostatistical Sciences Mayer 427 44 Binney Street, Boston, MA 02115-6084
List of speakers who are nonmembers: None