The effect of partial noncompliance on the power of a clinical trial

doi:10.1016/0197-2456(90)90010-Y

Controlled Clinical Trials

Volume 11, Issue 3, June 1990, Pages 157-168

https://doi.org/10.1016/0197-2456(90)90010-Y Get rights and content

Abstract

Noncompliance is an important concern in randomized trials and must be taken into account when calculating sample sizes. The effect of noncompliance is usually assessed through a simple binary model that assumes that the patient either does or does not comply with the allocated treatment. In this article I present an example from cancer prevention, and show that different time courses of noncompliance can have different effects on the power, with widely varying estimates of the sample size required. The time course as well as the level of noncompliance should be considered when planning trials with long-term treatments.

References (23)

M Halperin et al.
Sample size for medical trials with special reference to long-term therapy
J Chron Dis
(1968)
E Lakatos
Sample sizes for clinical trials with time-dependent rates of losses and noncompliance
Controlled Clin Trials
(1986)
DP Byar et al.
Clinical trials in diet and cancer
Prev Med
(1989)
S Self et al.
Statistical design of the Women's Health Trial
Controlled Clin Trials
(1988)
MA Schork et al.
The determination of sample size in treatment-control comparisons for chronic disease studies in which dropout or nonadherence is a problem
J Chron Dis
(1967)
M Wu et al.
Sample sizes for long-term medical trials with time-dependent dropout and event rates
Controlled Clin Trials
(1980)
JM Lachin et al.
Evaluation of sample size and power for analyses of survival with allowance for nonuniform patient entry, losses to follow-up, non-compliance, and stratification
Biometrics
(1986)
E Lakatos
Sample sizes based on the log-rank statistic in complex clinical trials
Biometrics
(1988)
D Anbar
The relative efficiency of Zelen's prerandomization design for clinical trials
Biometrics
(1983)
RG Newcombe
Explanatory and pragmatic estimates of the treatment effect when deviations from allocated treatment occur
Stat Med
(1988)

U Carlsson et al.

Evaluation of possibilities for mass screening for colorectal cancer with Hemoccult fecal blood test

Dis Colon Rectum

(1986)

Cited by (35)

How to minimize the dropout and crossover in an infertility trial?
2024, Fertility and Sterility
Randomized controlled trials and intent-to-treat analyses are important for infertility clinical studies. Dropouts or crossovers during the study process will disrupt the randomization design and affect the intent-to-treat analysis. In this review, we have briefly introduced the occurrence of dropout and crossover from our previous Reproductive Medicine Network and other related studies and provided some experience obtained from these studies on how to minimize and reduce the occurrence of dropout and crossover for infertility randomized clinical studies.
Predictors of participant retention in infertility treatment trials
2015, Fertility and Sterility
To identify variables associated with retention (or dropout) in infertility clinical trials. Retention of subjects in randomized controlled clinical trials (RCTs) has received considerable attention, but there have been few consistent findings.
Secondary analysis of data from RCTs.
Academic medical centers.
Women with polycystic ovary syndrome (PCOS) or couples with unexplained infertility, aged 18–40 years.
This study is not an intervention study, but the patients in the original RCTs were treated with any or combination of metformin, clomiphene citrate (CC), letrozole, and gonadotropins.
Successful retention versus dropout during the RCTs.
Race, ethnicity, body mass index (BMI), insurance coverage, history of smoking, and history of alcohol use were significantly associated with retention whether they were considered in bivariate analyses or a multivariable logistic model. Specifically, white race, higher income, having graduate degrees, normal weight, better insurance coverage, nonsmokers, and those who reported current use of alcohol at the start of the trial, had higher retention rates.
We identified several additive and persistent predictors of retention that can be used to guide the conduct of RCTs and improve the retention rate. Given the limitation of our association analysis, methodologically sound and theoretically grounded research are warranted so that high quality data can be collected to improve our understanding on the causes of dropout.
NCT00068861 (PPCOS-I), NCT00719186 (PPCOS-II), and NCT01044862 (AMIGOS).
Usefulness of a run-in period to reduce drop-outs in a randomized controlled trial of a behavioral intervention
2008, Contemporary Clinical Trials
Citation Excerpt :
When evaluating the outcomes of a RCT, it is of great importance to obtain complete information on randomized participants, and this can be achieved by retaining as many participants as possible. For studies that require multiple visits or interventions, a run-in period can increase the likelihood of follow-up and decrease the number of drop-outs or incomplete data after randomization, thus allowing for more precise statistical analyses, greater internal validity and stronger statistical power [4,5]. Though randomization is critical to a RCT, it is equally important that the subjects included in the analysis yield an unbiased assessment of treatment effects, i.e. that missing data is random [6,7].
We evaluated the usefulness of a simple run-in period to reduce drop-outs in a behavioral intervention to improve blood pressure (BP). In a pilot study where a run-in period was not used, we had a 25% drop-out rate.
A prospective evaluation was performed in the context of a blinded 3-arm randomized trial. Participants are eligible if they have uncontrolled BP on 2 consecutive visits. Potential participants are approached during a routine visit, informed, consented and enrolled. After a 1-month run-in period during which all participants receive a phone call to: i) verify phone availability, ii) get basic information on treatment, and iii) confirm the baseline visit, participants return for a baseline visit. They are then randomized to one of the three treatment arms: usual care, non-tailored counseling, or tailored counseling. Participants make return visits at 3, 6 and 12 months.
Of the 1275 potential participants who received detailed study information, 301 consented to participate, of whom 226 were enrolled. During the run-in period, 73 withdrew consent and 153 participants were randomized; 7 subsequently dropped out. There were no differences (p > .1) between the 73 cancelled and the 153 randomized patients. There were fewer drop-outs than in the pilot study (5% vs. 25%, p < .0001).
The run-in period reduces the number of drop-outs after randomization and improves statistical power. In order to retain external validity, it is important to compare participants who remain in the study and those that cancel, and incorporate that in generalizing from the study.
Sample size correction for treatment crossovers in randomized clinical trials with a survival endpoint
2002, Controlled Clinical Trials
Sample size determination in randomized clinical trials usually relies on the determination of survival rates at the time of analysis in both groups, under the null and the alternative hypotheses, the type I and II error rates and on other assumptions, such as proportional hazards in most cases. However, in numerous clinical trials for malignant chronic diseases, it is currently common that a patient allocated to the conventional treatment group would receive the experimental treatment in case of disease progression or relapse. Such crossovers are usually not taken into account when computing the sample size of the trial, but generally result in a decreased power of the trial. In this work, we aimed to correct the sample size of such trials to control the power, under an exponential survival assumption.
Comparison of three composite compliance indices in a trial of self-administered preventive therapy for tuberculosis in HIV-infected Ugandan adults
1998, Journal of Clinical Epidemiology
Compliance with tuberculosis preventive therapy in a randomized placebo-controlled trial in 2736 HIV-infected Ugandans was measured using urinary isoniazid metabolite testing, clinic attendance, and self-report. Overall, 77% of urine tests were positive, subjects kept 85% of their scheduled visits while on therapy, and 69% reportedly never forgot to take their medication. Different strategies were used for constructing three composite compliance indices in active arms: (1) an unweighted index of the summed scores on scaled compliance measures; (2) a weighted index using weights obtained from a survey of experts on tuberculosis; and (3) a statistically weighted index using principal components analysis. Composite indices were evaluated for reliability, validity, and practical utility. Understanding of the regimen, study arm, subsequent follow-up, tuberculosis status, and urine spot-check result were associated with composite compliance scores. The unweighted index in this study performed as well as the weighted indices.
Outcomes of a placebo run-in period in a head and neck cancer chemoprevention trial
1997, Controlled Clinical Trials
This study describes the outcomes of an eight-week placebo run-in period in a head and neck cancer chemoprevention trial. Of 391 former cancer patients who entered the run-in over the first two years of the trial, 91% were randomized. Pill counts showed that adherence rates ranged from 0% to 120% (mean 96%, SD = 15%). The trial did not randomize subjects who were no longer interested in trial participation (n = 20), who did not return within 10 weeks of enrollment date (n = 3), or who did not achieve a drug adherence level of at least 75% (n = 9). Three subjects were not randomized for other reasons.
Univariate predictors of run-in outcome (randomized or not randomized) included ethnicity, education level, cancer site, cancer stage, and Karnofsky performance score. Multivariate analyses resulted in a logistic model with Karnofsky performance and education level as significant predictors of randomization. Persons with a Karnofsky score of 100 had 2.3 higher odds of randomization (95% CI = 1.1, 4.9) than persons with compromised Karnofsky scores, and persons with more than a high school education had 2.1 higher odds of randomization (95% CI = 1.0, 4.9) than persons with less education. These results suggest that the use of a run-in period may compromise the external validity of randomized prevention trials. More research is needed to understand further the behavioral factors underlying the observed differences so that prevention researchers can develop effective interventions for facilitating trial participation, especially in under represented, trial-eligible groups. Investigators should expand the objectives of a runin period to (1) evaluate why eligible persons refuse trial enrollment or fail to be randomized at the end of the run-in and (2) use the run-in period for a systematic evaluation of levels and costs of intervention strategies designed to promote trial enrollment and adherence.

View all citing articles on Scopus

View full text

The effect of partial noncompliance on the power of a clinical trial

Abstract

J Chron Dis

Controlled Clin Trials

Prev Med

Controlled Clin Trials

The determination of sample size in treatment-control comparisons for chronic disease studies in which dropout or nonadherence is a problem

J Chron Dis

Sample sizes for long-term medical trials with time-dependent dropout and event rates

Controlled Clin Trials

Evaluation of sample size and power for analyses of survival with allowance for nonuniform patient entry, losses to follow-up, non-compliance, and stratification

Biometrics

Sample sizes based on the log-rank statistic in complex clinical trials

Biometrics

The relative efficiency of Zelen's prerandomization design for clinical trials

Biometrics

Explanatory and pragmatic estimates of the treatment effect when deviations from allocated treatment occur

Stat Med

Evaluation of possibilities for mass screening for colorectal cancer with Hemoccult fecal blood test

Dis Colon Rectum