Design Issues in Case-Control Studies
Ørnulf Borgan, Norman E. Breslow, Nilanjan Chatterjee, Mitchell H. Gail, Alastair Scott, Christopher J. Wild in Handbook of Statistical Methods for Case-Control Studies, 2018
Selection bias refers to any aspect of the study design that may tend to make the sample of study subjects unrepresentative of their source population. For example, in a case-control study, it may not be possible to include all cases within the defined geographic area or time period because some of them may have died, moved away, could not be located, or refused to participate; if exposure is related to case-fatality, for example, this will distort the true relationship with disease incidence. Likewise, controls should represent the source population that gave rise to the cases, but this is often difficult to accomplish if there is no population registry. Selection bias is difficult to assess quantitatively without data on the subjects who did not participate, hence it can be helpful to try to obtain at least limited demographic information on as many of them as possible. The potential for selection bias needs to be considered in hospital-based case-control studies. For example, cases with severe disease may be over-represented in a teaching hospital, and the patients with control diseases may not be representative of the source population with respect to the risk factor under study.
Infectious Disease Data from Surveillance, Outbreak Investigation, and Epidemiological Studies
Leonhard Held, Niel Hens, Philip O’Neill, Jacco Wallinga in Handbook of Infectious Disease Data Analysis, 2019
Systematic data errors only lead to bias when they are differential, i.e., when the extent of it depends on study participants’ characteristics and their outcomes. Bias is defined as a systematic deviation of results or inferences from truth [3]. In general, we can try to avoid bias by having an optimal design, collection, analysis, interpretation, and reporting of a study. Bias in epidemiological studies can arise in many ways, and dozens of types of biases have been described. The two major types of bias we will describe here are selection bias and information bias. We will not discuss confounding bias, since it is mainly a problem with epidemiological interpretation of results of studies into effects of determinants, while selection bias and information bias are also applicable to broader use of data.
Sampling Theory
Marcello Pagano, Kimberlee Gauvreau in Principles of Biostatistics, 2018
The ideal population we would like to describe is known as the target population. In the preceding example, the target population consists of all 15to 17‐year‐olds living in Massachusetts. In many situations, the entire target population is not accessible. If we are using school records to select our sample of teenagers, for instance, individuals who do not attend high school would have no chance of being included. After we account for practical constraints, the group from which we can actually sample is known as the study population. A list of the elements in the study population is called a sampling frame. Note that a random sample, although representative of the study population from which it is selected, may not be representative of the target population. If the two groups differ in some important way—perhaps the study population is younger on average than the target population—the selected sample is said to be biased. Selection bias is a systematic tendency to exclude certain members of the target population.
Challenges in estimating influenza vaccine effectiveness
Published in Expert Review of Vaccines, 2019
Kylie E. C. Ainslie, Michael Haber, Walt A. Orenstein
Bias may be present in observational studies for the estimation of influenza VE because differences in the risk of outcomes (e.g., infection) may exist that are caused by factors other than vaccination status. Depending on the study design, this bias may be a result of differences in the selection of cases and controls, such as a) differing health-care seeking behavior between vaccinees and non-vaccinees or the effect of vaccination on the probability of developing non-influenza ARI; b) confounding, such as the effect of health status or frailty on both the probability of vaccination and the probability of influenza ARI; or c) misclassification of influenza virus infection and/or vaccination status. In the following section we discuss three types of bias: selection bias, confounding, and information bias. However, the sources of bias discussed below are not exhaustive and additional bias in influenza VE studies may result from sparse data, missing data, and volunteer bias, among others.
Prevalence of extended high-frequency hearing loss among adolescents from two rural areas in Colombia
Published in International Journal of Audiology, 2021
Daniel Peñaranda, Lucía C. Pérez-Herrera, Diana Hernández, Sergio Moreno-López, Ilene Perea, Mario Jacome, Nancy Suetta-Lugo, Juan Manuel García, Augusto Peñaranda
Within the strengths of this study, selection bias control (which was conducted through a simple random sampling) stands out. As a result, no significant differences were expected between the population attending the educational institutions included in the study and the remainder of the teenage population in the rural environment. Additionally, data collection was performed by trained medical practitioners, and the completion time of the questionnaires was controlled. The questionnaire was designed by an expert in the audiological field, and the application of these instruments was standardised. Concerning the sample size of the study, we highlight that the actual size included for the robust logistic regression analysis was 694 ears. Therefore, our sample size meets the statistical criteria required to perform these analyses. Besides, we included the entire adolescent population who meet our inclusion criteria in both municipalities.
Recovery Dharma: Exploration of a Buddhist-based mutual help organization for the treatment of addiction
Published in Journal of Social Work Practice in the Addictions, 2022
Onawa LaBelle, Matthew Meeks, Noel Vest, Maurissa Hastings, Tayler Harding
The current study is not without limitations. First, we acknowledge that this sample does not characterize all individuals who engage in the RD program. We cast a wide net in our recruitment efforts to include the RD newsletter, RD website, and RD social media groups and pages; however, all of these locations are online; individuals without a computer, mobile device, or internet access did not participate. Thus, our sample represents the RD members who have access to the internet and may not represent the RD membership as a whole. Second, much like other MHOs, there is no membership database that could be used for recruitment purposes, which prevents any comparisons between people who chose to complete the survey and those who did not complete it. Self-selection bias may have been a factor in our sample. Third, although the measure we used for perceived support has been previously used in studies to characterize recovery samples (Curtis et al., 2019), it has not yet been validated for this purpose. The validation of this instrument or a similar measurement tool that assesses the perceived support from a specific recovery program is an additional important next step for future work.
Related Knowledge Centers
- Endometrial Cancer
- External Validity
- Internal Validity
- Lead Time Bias
- Sampling Bias
- Statistical Inference
- Menopause
- Sampling
- Self-Selection Bias
- Length Time Bias