{1}
##LOC[OK]##
{1}
##LOC[OK]##
##LOC[Cancel]##
{1}
##LOC[OK]##
##LOC[Cancel]##
Office of Behavioral & Social Sciences Research
National Institutes of Health
eLearning for Behavioral & Social Sciences Research
Home
About
Contact Us
Setting the Scene
Appropriate Research Methods
'Science' in the Social Sciences
Design Decisions in Research
Theory Development
Social and Behavioral Theories
Describing How
Sample Surveys
Social Survey Data Collection
Administrative Data Systems
Observational Studies
Explaining Why
Qualitative Methods
Conversation Analysis
Software and Qualitative Analysis
What Works
Clinical Trials
Cluster Unit Randomized Trials
Emerging Issues
Ethical Challenges
Multilevel Modeling
Objective Measurement of Subjective Phenomena
Measuring Socioeconomic Status
Evaluating the Quality of Health Care
PatientReported Outcomes
All Chapters
Setting the Scene
Appropriate Research Methods
'Science' in the Social Sciences
Design Decisions in Research
Theory Development
Social and Behavioral Theories
Describing How
Sample Surveys
Social Survey Data Collection
Administrative Data Systems
Observational Studies
Explaining Why
Qualitative Methods
Conversation Analysis
Software and Qualitative Analysis
What Works
Clinical Trials
Cluster Unit Randomized Trials
Emerging Issues
Ethical Challenges
Multilevel Modeling
Objective Measurement of Subjective Phenomena
Measuring Socioeconomic Status
Evaluating the Quality of Health Care
PatientReported Outcomes
Skip to content
This Chapter
Contents
Clinical Trials
1. Learning Objectives
2. Introduction
3. Classification
4. Endpoints
5. Design Issues
6. Erroneous Trial Results
7. Statistics
8. Summary
9. Glossary
10. References
11. Author Biographies
This Chapter
Contents
Appropriate Research Methods
'Science' in the Social Sciences
Design Decisions in Research
Theory Development
Social and Behavioral Theories
Sample Surveys
Social Survey Data Collection
Administrative Data Systems
Observational Studies
Qualitative Methods
Conversation Analysis
Software and Qualitative Analysis
Clinical Trials
Cluster Unit Randomized Trials
Ethical Challenges
Multilevel Modeling
Objective Measurement of Subjective Phenomena
Measuring Socioeconomic Status
Evaluating the Quality of Health Care
PatientReported Outcomes
Tables
Figures
Exercises
Examples

Login
print chapter
print page
text size:
Previous

Next
Page 24 of 26
Clinical Trials
9. Glossary of Terms
Show All
Hide All
GLOSSARY
Bias
Systematic errors associated with inadequacies in the design, conduct, or analysis of a trial on the part of any of the participants of that trial (patients, medical personnel, trial coordinators, or researchers), or in publication of its results, make the estimate of a treatment effect deviate from its true value. Systematic errors are difficult to detect and cannot be analyzed statistically but can be reduced by using randomization, treatment concealment, blinding, and standardized study procedures.
Confidence Intervals
A range of values within which the "true" population parameter (e.g. mean, proportion, treatment effect) is likely to lie. Usually, 95% confidence limits are quoted, implying that there is 95% confidence in the statement that the "true" population parameter will lie somewhere between the lower and upper limits.
Confounding
A situation in which a variable (or factor) is related to both the study variable and the outcome so that the effect of the study variable on the outcome is distorted. For example, if a study found that coffee consumption (study variable) is associated with the risk of lung cancer (outcome), the confounding factor here would be cigarette smoking, since coffee is often drunk while smoking a cigarette which is the true risk factor for lung cancer. Thus we can say that the apparent association of coffee drinking with lung cancer is due to confounding by cigarette smoking (confounding factor). In clinical trials, confounding occurs when a baseline characteristic (or variable) of patients is associated with the outcome, but unevenly distributed between treatment groups. As a result, the observed treatment difference from the unadjusted (univariate) analysis can be explained by the imbalanced distribution of this variable.
Covariates
This term is generally used as an alternative to explanatory variables in the regression analysis. However, they more specifically refer to variables that are not of primary interest in an investigation. Covariates are often measured at baseline in clinical trials because it is believed that they are likely to affect the outcome variable, and consequently need to be included to estimate the adjusted treatment effect.
Descriptive/Inferential Statistics
Descriptive statistics are used to summarize and describe data collected in a study. To summarize a quantitative (continuous) variable, measures of central location (i.e. mean, median, and mode) and spread (e.g. range and standard deviation) are often used, whereas frequency distributions and percentages (proportions) are usually used to summarize a qualitative variable. Inferential statistics are used to make inferences or judgments about a larger population based on the data collected from a small sample drawn from the population. A key component of inferential statistics is hypothesis testing. Examples of inferential statistical methods are ttest and regression analysis.
Endpoint
Clearly defined outcome associated with an individual subject in a clinical research. Outcomes may be based on safety, efficacy, or other study objectives (e.g. pharmacokinetic parameters). An endpoint can be quantitative (e.g. systolic blood pressure, cell count), qualitative (e.g. death, severity of disease), or timetoevent (e.g. time to first hospitalization from randomization).
Hazard Ratio
In survival analysis, hazard (rate) represents instantaneous event rate (incidence rate) at certain time for an individual who has not experienced an event at that time. Hazard ratio compares two hazards of having an event between two groups. If the hazard ratio is 2.0, then the hazard of having an event in one group is twice the hazard in the other group. The computation of the hazard ratio assumes that the ratio is consistent over time (proportional hazards assumption).
Hypothesis Testing or Significance Testing
Statistical procedure for assessing whether an observed treatment difference was due to random error (chance) by calculating a Pvalue using the observed sample statistics such as mean, standard deviation, etc. The Pvalue is the probability that the observed data or more extreme data would have occurred if the null hypothesis (i.e. no true difference) were true. If the calculated Pvalue is a small value (like <0.05), the null hypothesis is then rejected, and we state that there is a statistically significant difference.
IntentiontoTreat Analysis
A method of data analysis on the basis of the intention to treat a subject (i.e. the treatment regimen a patient was assigned at randomization) rather than the actual treatment regimen he received. It has the consequence that subjects allocated to a treatment group should be followed up, assessed, and analyzed as members of that group regardless of their compliance to that therapy or the protocol, irrespective of whether they later crossed over to the other treatment group or not or whether they discontinued treatment.
KaplanMeier Estimate and Survival Curve
A survival curve shows an estimate of the fraction of patients who survive over the follow up period of the study without an event of interest (e.g. death). The KaplanMeier estimate is a simple way of computing the survival curve taking into account patients who were lost to follow up or any other reasons for incomplete results (known as censored observations). It usually provides a staircase graph of the fraction of patients remaining free of event over time.
MetaAnalysis
The systematic review and evaluation of the evidence from two or more independent studies asking the same clinical question to yield an overall answer to the question.
Number needed to treat (NNT)
This term is often used to describe how many patients would need to be given a treatment to prevent one event. It is determined from the absolute difference between one treatment and another. In a randomized study the group receiving treatment A had a death rate of 12.5%, and the group receiving treatment B had a death rate of 15.0%. Both groups are matched for size and length of followup. Comparing the two treatments there was an absolute risk reduction of 15%  12.5% = 2.5% for treatment A. From this we can derive that the NNT (= 1/0.025) is 40. This means 40 patients need to be given treatment A rather than B to prevent 1 additional death.
Odds Ratio (OR) and Risk Ratio (RR)
These terms compare the probability of having an event between two groups exposed to a risk factor or treatment. The risk ratio (RR) is the ratio of the probability of occurrence of an event between two groups. The odds ratio (OR) is the ratio of the ratio of patients with and without an event in each group. If the number of deaths in the treatment and control arms (both of sample size 100) of a randomized study are 50 and 25 respectively, the RR = (50/100) / (25/100) = 2. The treatment group has a 2 fold relative risk of dying compared with the control group. The OR = (50/50) / (25/75) = 3 indicates that the odds of death in the treatment arm is 3fold of the control arm.
PerProtocol Analysis
A method of analysis in which only the subset of subjects who complied sufficiently with the protocol are included. Protocol compliance includes exposure to treatment, availability of measurements, correct eligibility, and absence of any other major protocol violations. This approach contrasts with the more conservative and widely accepted "intentiontotreat" analysis.
Power
The probability of rejecting the null hypothesis (e.g. no treatment difference) when it is false. It is the basis of procedures for calculating the sample size required to detect an expected treatment effect of a particular magnitude.
Random Error
An unpredictable deviation of an observed value from a true value resulting from sampling variability. It is a reflection of the fact that the sample is smaller than the population; for larger samples, the random error is smaller, as opposed to systematic errors (bias) that keep adding up because they all go in the same direction.
Regression Analyses
Methods of explaining or predicting outcome variables using information from explanatory variables. Regression analyses are often used in clinical trials to estimate the adjusted treatment effect taking into account of differences in baseline characteristics, and in epidemiological studies to identify prognostic factors while controlling for potential confounders. Commonly used regression models include linear, logistic, and Cox regression methods.
Treatment Effect
An effect attributed to a treatment in a clinical trial, often measured as the difference in a summary measure of an outcome variable between treatment groups. Commonly expressed as difference in means for a continuous outcome, a risk difference, risk ratio, or odds ratio for a binary outcome, and hazard ratio for a timetoevent outcome.
Bias
Systematic errors associated with the inadequacies in the design, conduct, or analysis of a trial on the part of any of the participants of that trial (patients, medical personnel, trial coordinators or researchers), or in publication of its the results, that make the estimate of a treatment effect deviate from its true value. Systematic errors are difficult to detect and cannot be analyzed statistically but can be reduced by using randomization, treatment concealment, blinding, and standardized study procedures.
Confidence Intervals
A range of values within which the "true" population parameter (e.g. mean, proportion, treatment effect) is likely to lie. Usually, 95% confidence limits are quoted, implying that there is 95% confidence in the statement that the "true" population parameter will lie somewhere between the lower and upper limits.
Confounding
A situation in which a variable (or factor) is related to both the study variable and the outcome so that the effect of the study variable on the outcome is distorted. For example, if a study found that coffee consumption (study variable) is associated with the risk of lung cancer (outcome), the confounding factor here would be cigarette smoking, since coffee is often drunk while smoking a cigarette which is the true risk factor for lung cancer. Thus we can say that the apparent association of coffee drinking with lung cancer is due to confounding by cigarette smoking (confounding factor). In clinical trials, confounding occurs when a baseline characteristic (or variable) of patients is associated with the outcome, but unevenly distributed between treatment groups. As a result, the observed treatment difference from the unadjusted (univariate) analysis can be explained by the imbalanced distribution of this variable.
Covariates
This term is generally used as an alternative to explanatory variables in the regression analysis. However, more specifically refer to variables that are not of primary interest in an investigation. Covariates are often measured at baseline in clinical trials because it is believed that they are likely to affect the outcome variable, and consequently need to be included to estimate the adjusted treatment effect.
Descriptive/Inferential Statistics
Descriptive statistics are used to summarize and describe data collected in a study. To summarize a quantitative (continuous) variable, measures of central location (i.e. mean, median, and mode) and spread (e.g. range and standard deviation) are often used, whereas frequency distributions and percentages (proportions) are usually used to summarize a qualitative variable. Inferential statistics are used to make inferences or judgments about a larger population based on the data collected from a small sample drawn from the population. A key component of inferential statistics is hypothesis testing. Examples of inferential statistical methods are ttest and regression analysis.
Endpoint
Clearly defined outcome associated with an individual subject in a clinical research. Outcomes may be based on safety, efficacy, or other study objectives (e.g. pharmacokinetic parameters). An endpoint can be quantitative (e.g. systolic blood pressure, cell count), qualitative (e.g. death, severity of disease), or timetoevent (e.g. time to first hospitalization from randomization).
Hazard Ratio
In survival analysis, hazard (rate) represents instantaneous event rate (incidence rate) at certain time for an individual who has not experienced an event at that time. Hazard ratio compares two hazards of having an event between two groups. If the hazard ratio is 2.0, then the hazard of having an event in one group is twice the hazard in the other group. The computation of the hazard ratio assumes that the ratio is consistent over time (proportional hazards assumption).
Hypothesis Testing or Significance Testing
Statistical procedure for assessing whether an observed treatment difference was due to random error (chance) by calculating a Pvalue using the observed sample statistics such as mean, standard deviation, etc. The Pvalue is the probability that the observed data or more extreme data would have occurred if the null hypothesis (i.e. no true difference) were true. If the calculated Pvalue is a small value (like <0.05), the null hypothesis is then rejected, and we state that there is a statistically significant difference.
IntentiontoTreat Analysis
A method of data analysis on the basis of the intention to treat a subject (i.e. the treatment regimen a patient was assigned at randomization) rather than the actual treatment regimen he received. It has the consequence that subjects allocated to a treatment group should be followed up, assessed, and analyzed as members of that group regardless of their compliance to that therapy or the protocol, irrespective of whether they later crossed over to the other treatment group or not or whether they discontinued treatment.
KaplanMeier Estimate and Survival Curve
A survival curve shows an estimate of the fraction of patients who survive over the follow up period of the study without an event of interest (e.g. death). The KaplanMeier estimate is a simple way of computing the survival curve taking into account patients who were lost to follow up or any other reasons for incomplete results (known as censored observations). It usually provides a staircase graph of the fraction of patients remaining free of event over time.
MetaAnalysis
The systematic review and evaluation of the evidence from two or more independent studies asking the same clinical question to yield an overall answer to the question.
Number needed to treat (NNT)
This term is often used to describe how many patients would need to be given a treatment to prevent one event. It is determined from the absolute difference between one treatment and another. In a randomized study the group receiving treatment A had a death rate of 12.5%, and the group receiving treatment B had a death rate of 15.0%. Both groups are matched for size and length of followup. Comparing the two treatments there was an absolute risk reduction of 15%  12.5% = 2.5% for treatment A. From this we can derive that the NNT (= 1/0.025) is 40. This means 40 patients need to be given treatment A rather than B to prevent 1 additional death.
Odds Ratio (OR) and Risk Ratio (RR)
These terms compare the probability of having an event between two groups exposed to a risk factor or treatment. The risk ratio (RR) is the ratio of the probability of occurrence of an event between two groups. The odds ratio (OR) is the ratio of the ratio of patients with and without an event in each group. If the number of deaths in the treatment and control arms (both of sample size 100) of a randomized study are 50 and 25 respectively, the RR = (50/100) / (25/100) = 2. The treatment group has a 2 fold relative risk of dying compared with the control group. The OR = (50/50) / (25/75) = 3 indicates that the odds of death in the treatment arm is 3fold of the control arm.
PerProtocol Analysis
A method of analysis in which only the subset of subjects who complied sufficiently with the protocol are included. Protocol compliance includes exposure to treatment, availability of measurements, correct eligibility, and absence of any other major protocol violations. This approach contrasts with the more conservative and widely accepted "intentiontotreat" analysis.
Power
The probability of rejecting the null hypothesis (e.g. no treatment difference) when it is false. It is the basis of procedures for calculating the sample size required to detect an expected treatment effect of a particular magnitude.
Random Error
An unpredictable deviation of an observed value from a true value resulting from sampling variability. It is a reflection of the fact that the sample is smaller than the population; for larger samples, the random error is smaller, as opposed to systematic errors (bias) that keep adding up because they all go in the same direction.
Regression Analyses
Methods of explaining or predicting outcome variables using information from explanatory variables. Regression analyses are often used in clinical trials to estimate the adjusted treatment effect taking into account of differences in baseline characteristics, and in epidemiological studies to identify prognostic factors while controlling for potential confounders. Commonly used regression models include linear, logistic, and Cox regression methods.
Treatment Effect
An effect attributed to a treatment in a clinical trial, often measured as the difference in a summary measure of an outcome variable between treatment groups. Commonly expressed as difference in means for a continuous outcome, a risk difference, risk ratio, or odds ratio for a binary outcome, and hazard ratio for a timetoevent outcome.
Previous

Next
Page 24 of 26
Office of Behavioral & Social Sciences Research
National Institutes of Health
Department of Health
and Human Services