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 'C'

CalibrationThe process of ensuring that an instrument gives a consistent reading; usually done by measuring a known standard and then adjusting (calibrating) the instrument accordingly. For example, the scale was calibrated monthly by weighing a 50-kg block of steel.
CaseA subject who has, or who develops, the outcome of interest. For example, cases were defined as those who had unstable angina, myocardial infarction, or sudden death during follow-up. See also control.
Case-cohort studyA research design in which subjects who develop a disease (or other outcome) are selected as cases during follow-up of a larger cohort, and then compared with a random sample of the overall cohort. For example, a case-cohort study enrolled a cohort of 2000 men with early prostate cancer, and compared levels of androgens and Vitamin D from samples obtained at baseline among those who died during follow-up with levels in a random sample of the entire cohort.
Case-control studyA research design in which cases who have a disease (or other outcome) are compared with controls who do not. For example, a case-control study compared average weekly consumption of nuts and seeds among cases of diverticulitis seen in an emergency room with nut and seed consumption of controls who had other gastrointestinal diagnoses.
Case-crossover studyA variant of the case-control design, in which each case serves as his own control, and the value of a specific time-dependent exposure in the period before the outcome occurred is compared with its value during one or more control periods of time. For example, a case-crossover design was used to determine whether patients who presented to an emergency room with a migraine headache were more likely to have eaten chocolate within the previous 2 hours than during a similar time of day one week previously. This design is susceptible to recall bias and is therefore most useful when an exposure can be ascertained objectively.
Categorical variableA variable that can have only several possible values. For example, the investigator transformed her measurements of reported educational level into a categorical variable with four values: less than high school, high school or some college, college degree, or post-college degree. See also continuous variable.
Cause-effectThe concept that a predictor is responsible for producing an outcome?or increasing the likelihood of an outcome?s occurrence. The purpose of most observational studies is to demonstrate cause-effect, though this is difficult to do unless the cause (e.g., a treatment) is assigned randomly. For example, the investigator performed a case-control study to determine whether there was a cause-effect relation between drinking alcohol (the cause) and pancreas cancer (the effect). See also confounding and effect-cause.
Chi-squared testA statistical technique that compares two (or more) proportions to determine if they are significantly different from one another. For example, a study determined whether the risk of dementia was similar among people who exercised at least twice a week as compared with those who exercised less frequently by comparing those risks statistically with a chi squared test.
Clinic-based controlIn the context of a case-control study, the selection of control patients from the same clinics (or practices) from which the cases were chosen. For example, the investigator used clinic-based controls in her study of whether running on pavement for at least two miles per week was associated with radiographic osteoarthritis of the knee.
Clinical prediction ruleAn algorithm that combines several predictors, including the presence or absence of various signs and symptoms and the results of medical tests, to estimate the probability of a particular disease or outcome. For example, the investigators developed a clinical prediction rule for the diagnosis of wrist fracture among post-menopausal women based on information about prior fractures, the characteristics of the fall (if any), physical examination of the forearm, and current medications.
Clinical trialA research design in which subjects receive one of (at least) two different treatments. Usually, the treatment is assigned randomly; thus the term randomized clinical trial. Also called an experiment. For example, the investigator performed a clinical trial to determine whether prophylactic treatment with penicillin reduced the risk of bacterial endocarditis among patients with abnormal heart valves who were undergoing dental procedures.
Cluster randomizationA technique in which groups of participants, known as clusters, are randomly assigned to different treatments, rather than having each participant assigned randomly as an individual. For example, in a study of the effects of noise reduction on recovery from cardiac surgery, the investigator used cluster randomization to assign intensive care units in 40 different hospitals to either a “post-operative quiet” intervention or a “usual care” control.
Cluster sampleA sampling technique in which subjects are selected in groups (clusters) rather than as individuals. Most often used for convenience when sampling large populations. For example, an investigator interested in determining the prevalence of drug use used cluster sampling to enroll 300 patients. First, she identified potential subjects by choosing 10 three-digit prefixes (e.g., 285-, 336-, etc.) within an area code; then she used random digit dialing to find 30 willing subjects within each three-digit cluster.
Coefficient of variation (CV)A measure of the precision of a measurement, obtained by dividing the standard deviation of a series of measurements performed on single sample by the mean of those measurements. Sometimes, the CV is obtained for values at the middle and the extremes of the measurement. For example, the lab determined that its coefficient of variation for serum estradiol levels was 10% in a sample from a peri-menopausal woman (in whom the estradiol level was very low), but only 2% in a younger woman.
Cohort studyA prospective cohort study involves enrolling a group of subjects (the cohort), performing some baseline measurements, and then following them forward in time to observe outcomes; a retrospective cohort study involves identifying a group of subjects (the cohort) in whom the measurements have already been made, and in whom some or all of the follow-up has already occurred. For example, an investigator did a retrospective cohort study of whether the results of an emotional intelligence test administered when soldiers enlisted in the Army was associated with the subsequent likelihood of developing post-traumatic stress disorder (PTSD).
CointerventionAn intervention that occurs after randomization, other than the intervention being studied, that affects the likelihood of an outcome. Co-interventions that occur at different rates in the study groups can bias the outcome and make it difficult to ascribe causality to the original intervention. For example, a study of the effect of a breast-feeding promotion intervention on subsequent allergic disease in infants was hard to interpret because the women in the intervention group not only breastfed longer, but were also more likely than the control group to delay the introduction of solid foods and to purchase hypo-allergenic formula. Blinding is the main approach to controlling co-intervention.
Complex hypothesisA research hypothesis that has more than one predictor or outcome variables. Complex hypotheses should be avoided, because they are difficult to test statistically. For example, the investigators reformulated their complex hypothesis (“That a new program in case management would affect both length of stay and the likelihood of readmission”) into two simple hypotheses (“That a new program in case management would affect length of stay” and also “That a new program in case management would affect the likelihood of readmission.”) See also simple hypothesis.
ConcordanceA measure of agreement between two (or more) observers about the occurrence of a phenomenon. For example, the concordance between radiologists A and B was 96% for the presence of a lobar pulmonary infiltrate, but only 76% for cardiomegaly. See also kappa.
Confidence intervalOften misunderstood, a confidence interval is best thought of as a measure of precision: the narrower a confidence interval, the more precise the estimate. Confidence intervals are closely related to statistical significance: A (1-α)% confidence interval (approximately) includes the range of values that were not statistically significantly different (at significance level α) from what was observed. Confidence intervals are often erroneously interpreted as direct statements about posterior probability (e.g., that there is a 95% probability that the true value is contained within the 95% confidence interval). This is incorrect because posterior probability depends on other information besides what was found in the study. For example, a relative risk of 1.6, with a 95% confidence interval from 0.9 to 2.8, would not be statistically significant at an alpha of 0.05, because the interval includes “no effect” (a relative risk of 1.0). See also alpha and P value.
ConfounderSee confounding.
ConfoundingAn epidemiologic phenomenon in which an association between a predictor and an outcome is due to a third variable (called the confounder or the confounding variable), rather than being a cause-effect relation between the predictor and the outcome. For example, the apparent association between cigarette consumption and cervical cancer was confounded by human papilloma virus (HPV) infection, because women who smoked were also more likely to have (multiple sexual partners and) HPV infection. See also effect-modification.
Confounding by indicationA specific form of confounding in which one of the indications for a treatment is the confounder; usually occurs in observational studies of the association between a treatment and an outcome. For example, the reviewers of an observational study were concerned that the reported association between a new treatment for bipolar disorder and increased suicide risk might have occurred because patients with more severe underlying disease had been selectively treated with the new medication.
Confounding variableSee confounding.
Consecutive samplingThe process by which subjects are chosen for a study one after another, until the sample size is achieved. Usually used to refer to the intended sample (see below); may also refer to the actual sample when performing medical records reviews, since informed consent may not be required. For example, the investigators performed consecutive sampling to review the charts of the first 100 patients with rheumatoid arthritis seen in the rheumatology clinic, beginning January 15, 2013.
Construct validityA term that describes how well a measurement corresponds to the theoretical definitions of the trait (the “construct ”) that is being measured. For example, a measurement of social anxiety was thought to have construct validity, because there were substantial differences in its values among people whose friends described them as “fun-loving” and “extroverted” as compared with those who were described as “shy” and “unlikely to go to parties.” See also content validity and criterion-related validity.
ContaminationThe undesirable process by which some or most of the effects of an intervention also affect subjects in the control group. For example, a study of the effects of whether teaching children to count backwards improved their overall arithmetic skills was plagued by contamination, because the children in the intervention group couldn?t resist teaching that skill to their friends in the control group.
Content validityA term that describes how well a measurement represents several aspects of the phenomenon being studied. For example, a measurement of insomnia was thought to have content validity, because it measured total amount of sleep, episodes of nighttime awakening, early morning awakening, energy on arising for the day, and daytime sleepiness. See also construct validity and criterion-related validity.
Continuous variableA measurement that, in theory, can have an infinite number of possible values. In practice, the term is often used for measurements that have “many” (some say 10 or more, others say 20 or more) possible values. For example, systolic blood pressure was measured as a continuous variable in mm Hg using a mercury sphygmomanometer. See also categorical variable.
ControlA term that has two distinct meanings. First, control refers to a subject who does not have the outcome of interest, and is therefore a member of a comparison group to which those with the outcome (the cases, see above) are compared. For example, for a study of risk factors for peptic ulcer disease, controls were selected from patients hospitalized during the study period with a non-gastrointestinal diagnosis. Second, control refers to the inactive “treatment” (i.e., a placebo or “usual care”) received by participants in a clinical trial who did not receive the study intervention; in that context, control is also used to refer to a participant who received that treatment. For example, the controls were given placebo tablets that looked identical to the active drug. See also case and intervention.
Convenience sampleA group of subjects who were selected for a study simply because they were relatively easy to access. For example, the investigator used a convenience sample of patients from her clinic to serve as controls for her case-control study of risk factors for meningioma.
Correlation coefficientA statistical term that indicates the degree to which two continuous measurements are related linearly, such that a change in one measure is associated with a proportional change in the other. Often abbreviated as “r.” For example, height and weight were correlated in a sample of middle-aged women with r = 0.7.
Cox modelAlso called Cox proportional hazards model. A multivariable statistical technique that measures the individual effects of one or more predictor variables on the rate (hazard) at which an outcome occurs in a sample, accounting for differing lengths of follow-up among subjects. See also logistic regression model.
Criterion-related validityA term that describes how well a measurement correlated with other ways of measuring the same phenomenon. For example, a measurement of depression in adolescents was thought to have criteria-related validity, because it had a high correlation with scores from the Beck depression inventory. See also construct validity and content validity.
Cross-overA term used to describe a subject, usually in a clinical trial, who starts out in one group (say, the active treatment) and switches to the other group (say, usual care) during the study. Most commonly occurs when the active treatment involves a procedure. For example, 15 subjects with prostate cancer who were initially assigned to watchful waiting crossed-over to receive radiation therapy or surgery during the trial.
Cross-sectional studyA research design in which subjects are selected and measurements made within a limited period of time, usually to estimate the prevalence (see below) of an exposure or a disease. For example, the prevalence of myopia was estimated in a cross-sectional study of 1200 college students in Berkeley, CA.
Crossover studyA research design in which all the subjects from one treatment (or control) group are switched to the other group, usually at the midway point of the study. Sometimes, there is a washout period (see below) between the two phases. This design, which enables all subjects to receive the active treatment, is only useful for conditions that wax and wane, not for those that are permanent. For example, patients with migraine were involved in a crossover study comparing a new drug with a placebo for the prevention of migraine.
Cumulative incidence See incidence proportion.

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