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 'S'

SampleThe subset of the population that participates in a study. For example, in a study of a new treatment for asthma, where the target population (see below) is all children with asthma and the accessible population (see above) is children with asthma in the investigator?s town this year, the study sample is the children in the investigator?s town this year who actually enroll in the study.
Sample sizeThis term has two meanings. It can either be the number of participants enrolled in a study, or the estimated number of participants needed for a study to be successful. For example, the investigator estimated that she needed to have a sample size of 54 subjects to have 90% power to detect a doubling in the risk of aggressive behavior among third-grade boys exposed to violent video games.
SamplingThe process of selecting participants to invite to enroll in a study when the number of eligible participants is larger than the estimated sample size. For example, the investigator used a “1 in 3” sampling scheme to select, on average, every third eligible subject. See also convenience sampling, consecutive sampling, probability sampling, simple random sampling, systematic sampling, stratified random sampling and cluster sampling.
Sampling biasA systematic error that causes the sample of persons included in a study not to represent the target population. For example, if participants in a study of risk factors for osteoporosis were recruited from among patients hospitalized for hip fracture, falling may falsely appear to be a risk factor for osteoporosis due to sampling bias.
ScaleA common approach to measuring abstract concepts by asking multiple questions that are scored and combined into a scale. For example, the SF36 scale for measuring quality of life asks 36 questions that yield 8 scales related to functional health and well-being. (SF stands for “short form.”) See also Likert scale.
Scientific misconductA general term for intentionally defrauding the scientific community, including research misconduct (fabrication and falsification of data and plagiarism), as well as guest and ghost authorship, and conflict of interest that is not disclosed or managed. For example, the investigator?s institution judged that she was guilty of scientific misconduct because she failed to disclose an equity interest in the company that made the medical device she was studying.
Secondary data analysisUse of existing data to investigate research questions other than the ones for which the data were originally collected. Secondary data sets may include previous research studies, medical records, health care billing data, and death certificates. For example, hospital discharge data and the National Death Index could be used in a secondary data analysis to determine 1-year mortality among patients with a discharge diagnosis of acute pancreatitis.
Secondary research questionQuestions other than the primary research question, often including additional predictors or outcomes. For example, if the primary research question is to determine the association of alcohol consumption among pregnant women and low birth weight infants, a secondary question might be to determine the association of alcohol consumption and anemia during pregnancy.
Selection criteriaRules that define who is eligible to enroll in a study, including the inclusion and exclusion criteria. For example, in a clinical trial of transdermal testosterone to enhance libido in postmenopausal women, the selection criteria might be women aged 45 to 60 years with low libido who are free of coronary disease who have not had more than 3 menstrual periods in the prior year.
SensitivityThe proportion of subjects with disease in whom a test is positive (“positive in disease,” or PID). For example, compared with pathology results on biopsy, the sensitivity of a serum PSA test result > 4.0 ng/mL is about 20% for the detection of prostate cancer; in other words, 20% of men with prostate cancer will have a PSA > 4.0 ng/mL. See also specificity.
Sensitivity analysisUsing different methods (e.g., alternate definitions of predictor or outcome variables, different statistical tests) to determine if the results of the primary analysis are robust. For example, in a meta-analysis of clinical trials of the effect of selective serotonin reuptake inhibitors on depression, in a sensitivity analysis, the investigator might include only the blinded trials to demonstrate that the results are robust when the analysis is restricted to high-quality trials.
Simple hypothesisA hypothesis with only one predictor variable and one outcome variable. For example, the investigator rephrased his complex hypothesis into the simple hypothesis that people who eat fruit at least five times a week are less likely to develop colon cancer. See also complex hypothesis.
Specific aimsIn a research proposal, brief statements of the goals of the research. For example, one specific aim of a randomized trial of the effect of testosterone on bone mineral density in men might be: “To test the hypothesis that compared with men assigned to receive a placebo patch, those assigned to receive the testosterone path will have less bone loss during 3 years of treatment.”
SpecificationA design phase strategy to cope with a confounder by specifying a value of that confounder as an inclusion criterion for the study. For example, in a study of the effect of pacifier use on the risk of sudden infant death syndrome, the investigator might use specification to include only formula-fed infants in the study. If a decreased risk of sudden death was found in pacifier users, it could not be because they were more likely to be breastfed.
SpecificityThe proportion of subjects without the disease being tested for in whom a test is negative (“negative in health,” or NIH). For example, compared with pathology results on biopsy, the specificity of a serum PSA test result of >4.0 ng/mL is about 95% for the detection of prostate cancer; in other words, 95% of men without prostate cancer will have a PSA = 4.0 ng/mL. See also sensitivity.
Spectrum biasThe situation in which the accuracy of a test is different in the sample than it would have been in the population because the spectrum of disease (which affects sensitivity) or non-disease (which affects specificity) in the sample differs from that in the population in which the test will be used. For example, a new serum test designed to diagnose esophageal cancer may be relatively accurate in a study of patients with advanced esophageal cancer compared to healthy medical students, but perform poorly when used in elderly patients with undiagnosed difficulty swallowing .
Spurious associationAn association between a predictor and an outcome that is seen in a study but that is not true in the population, either due to chance or bias. For example, observational studies of found decreased risk of cardiovascular disease among persons who took beta carotene supplements. However, a randomized trial of beta carotene supplements found no effect on risk of cardiovascular disease, suggesting that the association observed in the observational studies was spurious.
StandardizationSpecific, detailed instructions for how to perform a measurement designed to maximize reproducibility and precision of the measurement. For example, in a study that measures blood pressure, standardization of the measurement could include instructions on preparing the participant, what size cuff to use, where to place the cuff, how high to inflate and deflate the cuff, and which heart sounds indicate systolic and diastolic blood pressure.
Steering committeeIn a multi-center study, a committee that provides overall governance for the study. It is generally composed of the principal investigators of each study site, the coordinating center, and representatives of the sponsor. For example, the study?s steering committee decided whether proposed ancillary studies should be conducted.
StratificationAn analysis phase strategy for controlling confounding by segregating the study participants into strata according to the levels of a potential confounder and analyzing the association between the predictor and outcome separately in each stratum. For example, in a study of the association between exercise and the risk of stroke, not exercising regularly might be associated with increased risk of stroke because many people who don?t exercise are obese, and obesity increases stroke risk. To minimize the potential confounding effect of obesity, participants were stratified by their body mass index, and the analyses were carried out separately in those who were normal weight, overweight, or obese at baseline.
Stratified blocked randomizationA randomization procedure designed to ensure that equal numbers of participants with a certain characteristic (usually a confounder) are randomly assigned to each of the study groups. Randomization is stratified by the characteristic of interest; within each stratum, participants are randomly assigned in blocks of predetermined size (see blocked randomization). For example, in a trial of a drug to prevent fractures, a history of vertebral fracture is such a strong predictor of the outcome and of response to many treatments that it would be best to ensure an equal number of participants with and without vertebral fracture in each of the study groups. Therefore, participants were divided into those with vertebral fractures and those without such fractures; within each group, randomization was carried out in blocks of six to ten subjects.
Stratified random sampleA sampling technique in which potential participants are stratified into groups based on characteristics, such as age, race, or sex, and a random sample is taken from each stratum. The strata can be weighted in various ways. For example, in a study of the prevalence of pancreatic cancer in California, strata of race/ethnicity could be weighted to reflect the proportion of each race/ethnic group in the state or to oversample minorities.
Subgroup analysisComparisons between randomized groups in a subset of the trial participants. For example, in a randomized trial of the effect of a selective estrogen receptor modulator (SERM) on recurrence of breast cancer, the investigators performed a subgroup analysis of the effect of treatment by stage of cancer, comparing the effect of the SERM to placebo among women with stage I, stage II, and stages III and IV disease.
SubjectSee participant.
Subject biasSee recall bias.
Summary effectIn a meta-analysis, the weighted average effect seen in the included studies; the formula for the weights depends on the model. For example, in a meta-analysis of randomized trials of the effect of an angiotensin-converting enzyme (ACE) inhibitor on mortality in patients with coronary disease,, the summary effect with the fixed effect model was the weighted mean relative risk, weighted by the inverse of the variance of the relative risk in each included study. See also fixed-effects models and random-effects models.
Surrogate markerA measurement thought to be associated with meaningful clinical outcomes. A good surrogate marker usually measures changes in an intermediate factor in the main pathway that determines the clinical outcome. For example, an increased CD4 lymphocyte count in patients with human immunodeficiency virus (HIV) infection is a good surrogate marker for the effectiveness of antiretroviral drugs because it predicts lower risk of opportunistic infections.
SurveyA cross-sectional study in a specific population, usually involving a questionnaire. For example, the National Epidemiologic Survey on Alcohol and Related Conditions enrolled a representative sample of adults in the US and asked questions about present and past alcohol consumption, alcohol use disorders, and utilization of alcohol treatment services.
Survival analysisA statistical technique used to compare times to an outcome (not necessarily survival) among groups in a study. For example, in a randomized trial of the effect of coronary artery bypass surgery compared with percutaneous coronary angioplasty for the prevention of myocardial infarction and death, survival analysis could be used to compare time from starting treatment to either of those outcomes in the two groups.
Systematic errorAnother term for bias that distorts a study?s findings or measurements. See bias.
Systematic reviewA review of the medical literature that uses a systematic approach to identify all studies of a given research question, clear criteria to include a study in the review, and standardized methods to extract data from the included studies. A systematic review may also include a meta-analysis (see above) of the study results. For example, the investigator did a systematic review of all studies that tested whether zinc supplements reduced the risk of developing colds.
Systematic sampleA sample that is drawn by enumerating the units of the eligible population and selecting a subset of the population using a pre-specified process. For example, in the Framingham Heart Study, investigators constructed a list of all adult residents of the town of Framingham, Massachusetts, and then selected every other resident to be included in the study.

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