Andersons pediatric cardiology 635

then all of the measurements are made at once in the present (see Table 24.2) They are useful for determining the prevalence of risk factors at a chosen point, and the outcomes in a population Associations between risk factors and outcomes are sought, but it is difficult to establish any causal relationships Surveys are the most common methodology used for cross-sectional studies An example of a cross-sectional study would be surveying all adult patients with tetralogy of Fallot and asking about their surgical history, current medications, educational status, and current employment status Case-Control Studies Case-control studies are efficient designs for studying rare risk factors or rare outcomes for which study of a cohort would require an excessively large number of subjects or an excessively long duration of follow-up (see Table 24.2) A population of subjects with a given condition or outcome (the cases) is assembled and a comparison group of subjects is assembled from the same population but known to be free of the given condition or outcomes (the controls) The two groups are then assessed for baseline characteristics and risk factors, usually from nonconcurrent sources of data including chart review or recall from the subjects The relative frequency of risk factors is then determined in the cases compared with the controls Since they are not population-based, these studies do not give estimates of prevalence, incidence, or relative risk An example would be comparing a group of individuals who develop doublechamber right ventricle years after a the closure of a ventricular septal defect (VSD) and comparing them with a similar group of individuals after VSD closure that do not develop double-chamber right ventricle One would then determine differences between the two groups that might be associated with developing double-chamber right ventricle Case-control studies can be very efficient and cost-effective but have several limitations and threats to their validity First, since the groups are differentiated on the basis of a single condition or outcome, this is the only outcome that can be assessed Second, the selection of both cases and controls can be challenging, and sampling bias and nonrepresentativeness are important threats Often only currently known cases are represented, excluding subjects who died or those whose condition resolved without detection Further, cases usually derive from a setting of medical care, this time excluding subjects with missed or incorrect diagnoses or those who do not present for medical care or present elsewhere They may overrepresent subjects with more symptomatic, severe, complex, or prolonged disease or those that had a known risk factor that predisposed them to increased surveillance and diagnosis Likewise, the selection of controls can be challenging The goal ideally is to select the controls randomly from the larger population at risk from which the cases were derived In our example, the larger population is all people who had a VSD closure For convenience, controls can be recruited from the same setting in medical care The controls would then be representative of the cases derived from the same catchment area For instance, one could recruit all the patients from the same outpatient clinic Additionally, controls can be matched regarding important confounding characteristics, commonly age and gender, with the individual cases This is a useful strategy when a known or highly suspect confounder has the potential for being differentially represented in the cases as opposed to the controls Matching based on characteristics decreases the risk of that characteristic causing confounding, but it also eliminates one's ability to assess that characteristic as a risk factor, since every case will be the same as the control for that factor Matching more than one control to every case is common practice and allows increased power, although there is limited gain beyond matching three controls to every case With regards to potential biases, the nature of the assessment of risk factors is particularly at risk There may be a potential for differential measurement in the cases versus the controls This becomes increasingly important with greater subjectivity of the risk factors being assessed Cases may be more likely to have certain measurements, the measurements may be made and recorded by more qualified observers, or the cases may be more motivated to recall certain factors The investigators may be more aggressive in the identification of risk factors if they know that the subject is a case Blinding of assessments wherever possible becomes important, where the assessment occurs without knowing if the subject is a case or a control Nested Case-Control Studies A case-control study can be effectively nested within a cohort study (see Table 24.2) This increases efficiency and cost-effectiveness when assessment of a risk factor might be expensive or undesirable When sufficient subjects have developed an outcome, which is usually rare, a sample of controls is selected from the subjects free of the given outcome within the cohort The subjects chosen as controls may be matched to those representing cases according to key potential confounding characteristics, particularly time or period of observation This allows the risk factor to be assessed in a smaller proportion of the cohort For instance, in studying the incidence of surgical site infection in a cohort of infants undergoing the Norwood procedure, a researcher could identify all patients with surgical site infections and match them to a noninfected Norwood patient based on year of surgery This would allow some control of confounding from changes in the intensive care unit's practice pattern over time and allow examination of patient-specific risk factors Causality From Observational Studies An association is an identifiable relationship between a risk factor and an outcome or disease, but that risk factor may not cause the disease to occur A causal factor is responsible for the occurrence of the outcome A causal factor does not have to lead directly to the outcome or be the primary driver of it, but it must be a part of the true development of the outcome Although associations between risk factors and outcomes can readily be determined from observational studies, often researchers and clinicians are interested in identifying those associations that are causal in nature and free from bias, error, and confounding There are several criteria that must be satisfied before concluding that a relationship is causal in nature, as outlined in Box 24.1 Box 24.1 Criteria for Defining Causality in Relationships Among Independent Variables, Including Management, and Dependent Variables, Including Outcomes or Conditions ■ Is the relationship biologically plausible in terms of the current state of knowledge regarding pathophysiology? ■ Is the relationship strong? ■ Is the temporal relationship correct, in that the risk factor precedes the development of the outcome or condition?