Glossary

A   |  B   |  C   |  D   |  E   |  F   |  G   |  H   |  I   |  K   |  L
M   |  N   |  O   |  P   |  Q   |  R   |  S   |  T   |  V   |  W   |  Z

Glossary terms for 'R'

Random errorA departure of a measurement or estimate from the true value due to chance variation. Random error can be reduced by repeating measurements and by increasing the sample size. For example, if the true prevalence of the use of fish oil by persons with coronary disease in the population is 20%, a study that enrolls 100 participants might find that exactly 20% use fish oil, but just by random error, the proportion is likely to be a bit higher or lower than that.
Random sampleA sample drawn by enumerating the units of the population and selecting a subset at random. For example, a random sample of persons with cataracts at an investigator?s clinic would require that the investigator list all of the patients with cataracts and use a table of random numbers or computer-generated random numbers to select the sample. See also probability sample.
Random-effects modelA general term used in multi-level statistical analysis, discussed in this book only with respect to meta-analysis, where it describes a statistical model in which the study weights and variance of the summary effect estimate incorporate a term for the variability between the results of the individual included studies. For example, in a meta-analysis of clinical trials of the effect of practicing yoga on depression, the results of the trials were variable; thus smaller studies contributed more to the summary effect based on the random-effect model, and the confidence interval was wider than with the fixed-effect model. See also fixed-effects model.
RandomizationThe process of randomly assigning eligible participants to one of the study groups in a randomized trial. The number of treatment groups and the probability of being assigned to any group are determined before randomization begins. Although eligible participants are usually assigned to 2 study groups with equal (50%) probability, random assignment can be made to any number of study groups with any predetermined probability. For example, in a study comparing two treatments with a placebo control, randomization could be to 3 groups, with 30% assigned to each of the 2 active treatment groups and 40% assigned to placebo.
Randomized blinded trialA study design in which eligible participants are randomly assigned to the study groups with predetermined probability and study group assignment is not known to investigators, participants or other staff involved in the study. For example, a randomized blinded trial of a new pill for treatment of diarrhea would require that eligible participants be randomly assigned to the new pill or an identical placebo pill (usually with 50% chance of being assigned to each group) and that the investigators, participants and study staff not know if a participant is taking the active medication or placebo.
RateA measure of risk, defined as the number of subjects who develop an outcome divided by the person-time at risk. For example, the rate of stroke in the study was 23 per 1000 person-years. See also hazard .
Recall biasA specific type of bias in which whether a subject remembers his exposure to a risk factor is influenced by another factor, especially by whether the subject is a case or control. For example, recall bias was thought to be the reason why cases of amyotrophic lateral sclerosis were more likely to recall exposure to insecticides than controls.
Receiver operating characteristic (ROC) curveA graphical technique to quantify the accuracy of a diagnostic test and illustrate the trade-off between sensitivity and specificity at different thresholds for considering the test positive. The curve displays the rates of true positives (sensitivity) on the Y-axis and the corresponding rates of false positives (1-specificity) on the X-axis at several cutpoints for considering the test positive. The area under the ROC curve, which ranges from 0.5 for a useless test to 1.0 for a perfect test, is a useful summary of the overall accuracy of the test. For example, the area under the ROC curve for the use of CT scans (which could be interpreted as Clearly positive, Likely positive, Not helpful, Likely normal, or Clearly normal) to diagnose appendicitis was 0.95, substantially better than the value of 0.77 for ultrasound (which had similar categories of interpretation).
RecruitmentThe process of identifying and enrolling eligible participants in a study. Recruitment methods vary depending on the nature of the study. For example, recruitment for the study included identifying eligible patients in specialty clinics, advertising in fliers and newspapers, and using the Internet and social media sites.
RegistryA database of persons with a certain disease or who underwent a particular procedure. Studies can be conducted using registries by collecting outcome data as part of the registry, or by linking registry data to other sources, such as cancer registries or the National Death Index. For example, the San Francisco Mammography Registry obtains data on all women who undergo mammography at the 3 largest mammography centers in San Francisco; investigators have linked it with local cancer registries to estimate mammography accuracy.
Regression to the mean The tendency for outlying (very high or very low) values to be closer to the population mean when repeated. For example, in a group of children selected for a study based on having systolic blood pressures above the 95th percentile, the majority of children were observed to have lower blood pressures at the first follow-up visit, even though they had not yet received any treatment.
Relative riskSee risk ratio.
Representative sampleA sample of people enrolled in a study that represents the target population. For example, in the Framingham Heart Study, the target population was all adults. The accessible population (to investigators located in Boston) was the adult population of the town of Framingham, MA. Investigators enumerated adults in Framingham and asked every second resident to enroll in the study. This approach could result in a representative sample, but some people refused to enroll and were replaced by volunteers. Since volunteers often have more healthy habits than non-volunteers, the sample may have over-represented healthy persons. In addition, the population of Framingham (which was mostly white) does not represent all US adults, and certainly does not represent adults on other countries.
Reproducibility studyA study in which the reproducibility of a measurement is the main research question, typically performed by comparing the results of a measurement done multiple times by the same person or machine (intraobserver reproducibility) or the results of the same measurement done by different persons or machines (interobserver reproducibility). For example, the investigators performed a reproducibility study to determine whether a new electronic stethoscope could improve the ability to detect diastolic heart murmurs.
Research hypothesisA statement by the investigator that summarizes the main elements of the study, including the population of interest, the predictor and outcome variables, and an anticipated result. For statistical purposes, the research hypothesis is stated in a form that establishes the basis for tests of statistical significance, generally including a null and alternative hypothesis. For example, the research hypothesis was that migraine headaches would be associated with at least a 20% increase in risk of stroke.
Research misconductIllegal or unethical conduct of research, including plagiarism and fabrication or falsification of research data. For example, a research coordinator at the VA Medical Center in Albany, NY was found to have repeatedly submitted false documentation to allow persons who did not qualify for a study to be enrolled. All data from the Albany site were subsequently excluded, such that the participants? time and effort were wasted. See also scientific misconduct.
Research proposalA document written for the purpose of obtaining research funding that describes the proposed study design, participants, measurements, statistical analyses, and ethical issues. For example, the National Institutes of Health receives thousands of research proposals annually from investigators who seek funding for their studies.
Research questionThe question a research project is intended to answer. A good research question should include the predictor and outcome of interest and the population that will be studied. Research questions generally take the form of “Is A associated with B in population C?”, or (for a clinical trial) “Does A cause B in population C?”. For example, “Does regular use of dental floss reduce the risk of coronary events in persons with diabetes?”
Response rateThe proportion of eligible participants who respond to a questionnaire or to a particular item on it. A low response rate can decrease the internal validity of the study and bias the outcome. For example, in a survey of high school students, a response rate of 20% to a question on marijuana use would suggest the result is not likely to be a valid estimate of the real rate of marijuana use among students. See also missing data.
Retrospective cohort studyA cohort study in which assembly of the cohort, baseline measurements, and follow-up happened in the past. For example, to describe the natural history of thoracic aortic aneurysms, an investigator conducting a retrospective cohort study in 2012 could obtain data from hospital discharge records of patients who had a diagnosis of aortic aneurysm in 2007 and use hospital discharge records and the National Death Index to determine which patients subsequently had ruptured aortic aneurysm or died before 2012.
Risk differenceThe risk for an outcome in one group minus the risk in a comparison group. For example, if the risk for venous thromboembolic events among women who are current users of estrogen is 5/1000 (0.5%) and the risk among never users of estrogen is 2/1000 (0.2%), the risk difference among women using estrogen compared to nonusers is 3/1000 (0.3%). See also number needed to treat.
Risk ratio (relative risk)The risk for an outcome in one group divided by the risk in a comparison group. For example, if the risk for venous thromboembolic events among women who are current users of estrogen is 5/1000 (0.5%) and the risk among never users of estrogen is 2/1000 (0.2%), the relative risk among women using estrogen compared with nonusers is 2.5. See also odds ratio and hazard ratio.
Run-in periodIn a clinical trial, a brief period during which eligible participants take either the placebo or the active intervention; only those who achieve a certain level of adherence, tolerate the intervention, or have a beneficial effect on an intermediate outcome, are eligible for the main trial. For example, in the Cardiac Arrhythmia Suppression Trial, only those who had a satisfactory reduction in premature ventricular contractions on active medication during the run-in period were randomized to continue medication or switch to placebo.

Glossary material from Hulley SB et al. Designing Clinical Research, 4th ed. Philadelphia, Lippincott Williams & Wilkins, 2013.