Presentation is loading. Please wait.

Presentation is loading. Please wait.

Instructor Resource Chapter 16 Copyright © Scott B. Patten, 2015.

Similar presentations


Presentation on theme: "Instructor Resource Chapter 16 Copyright © Scott B. Patten, 2015."— Presentation transcript:

1 Instructor Resource Chapter 16 Copyright © Scott B. Patten, 2015.
Permission granted for classroom use with Epidemiology for Canadian Students: Principles, Methods & Critical Appraisal (Edmonton: Brush Education Inc.

2 Chapter 16. Other study designs

3 Objectives Describe important features of the following study designs: nested case-control studies, case- crossover studies, retrospective cohort studies, randomized controlled trials, case-cohort studies, and ecological studies. Discuss advantages and disadvantages of these designs over other designs.

4 Study designs covered so far
So far, we have covered the “classic” epidemiologic study designs: cross-sectional studies case-control studies prospective cohort studies.

5 Why is study-design classification important?
Identifying a study’s design is an early step in critical appraisal. It begins the process of thinking through a study’s vulnerability to error. It helps organize information about the study and focus critical appraisal on likely vulnerabilities.

6 Nested case-control studies
Nested case-control studies are not very different from classic case-control studies. They have a special name because of the way they are conducted. A nested case-control study is a case-control study situated within a prospective cohort study.

7 Nested case-control studies (continued)
The prospective cohort study generates cases and potential controls for the nested case-control study. This provides a well-defined source population from which cases and controls both arise. Whenever a case occurs, a control can be selected from among the cohort members who do not have the disease at that point in time (the risk set), providing a firm procedure for control selection.

8 Nested case-control studies (continued)
By assessing exposure in the case group and in a sample of the noncase group (controls), the nested case-control approach applies the familiar efficiency of the case-control design. It is often more efficient to properly select a subset of eligible cases rather than the whole group of all eligible cases.

9 Nested case-control studies (continued)
Nested case-control studies can make selection of cases and controls easier, because the cohort is being actively followed and is presumably strongly engaged in research. For the same reason, they can often achieve high response rates. They can benefit from work already done by the host study, which may have already measured many relevant confounding or effect-modifying variables. They may it easier to ensure that cases are incident cases, because close follow-up in the host study can lead to greater clarity of the temporal relationship between exposure and disease.

10 Case-crossover studies
Case-crossover studies are another variant of case- control studies. Like case-control studies, they have a backward logical and temporal direction: The investigation begins with the identification of a series of cases. Exposure is retrospectively assessed. Unlike case-control studies, case-crossover studies use the same people for cases and controls

11 Case-crossover studies (continued)
A case-crossover study assesses the frequency of exposure in the cases immediately before the onset of disease, and compares this to another time when the same people did not develop the disease. In this sense, these studies also have a retrospective temporal direction. If an exposure precipitates disease, that exposure should occur more frequently during the interval before disease onset than during some other interval when disease did not occur. The case-crossover design is used to study outcomes that rapidly, and typically temporarily, follow exposure.

12 Case-crossover studies (continued)
Because the comparisons occur within individuals, individual characteristics that don’t change over time and might contribute to the outcome are “matched” and cannot therefore act as confounders of the exposure-disease relationship. Matching controls confounding due to fixed characteristics such as genetic factors, stable psychological characteristics (personality), education, and stable aspects of health status. Much like pair-matched data in case-control studies, the nonindependence of these observations requires an approach that accounts for nonindependence, so case-crossover studies use analysis statistics that are designed for matched data.  

13 Retrospective cohort studies
Retrospective cohort studies have a forward logical direction: they start with exposure and ask whether an increased risk of disease follows exposure. They have a retrospective (backward) temporal direction: they look back in time

14 Retrospective cohort studies (continued)
Retrospective cohort studies require exposure data from the past. Examples include the use data from: occupational settings—some occupational exposures can be defined by participation in a particular industry (e.g., uranium mining) data collected routinely for safety purposes in occupational settings (e.g., radiation exposure badges) a group that was exposed to a disaster or accident (e.g., a nuclear meltdown), which qualifies the group as an exposed cohort

15 Retrospective cohort studies (continued)
Many retrospective cohort studies compare occupational (or other) cohorts to the general population. However, some occupational groups tend to be dominated by young people. Some may have an imbalanced sex ratio. This produces an obvious vulnerability to confounding by the nonmodifiable variables of age and sex. Retrospective cohort studies often use indirect standardization to address this vulnerability: reporting standardized mortality, standardized morbidity ratios (SMRs), and standardized incidence ratios (SIRs).

16 Retrospective cohort studies (continued)
These studies do not always based their comparisons on a general population referent. Some use referent cohorts consisting of other workers—for example, people who have worked in a similar setting or who have been employed by the same company. These studies can employ the same measures of association typical of prospective cohort studies: incidence proportion ratios (risk ratios), incidence rate ratios, and hazard ratios.  

17 Randomized controlled trials (continued)
Randomized controlled trials (RCTs) have a forward logical direction: they start with an exposure and then determine outcome. They also have a forward temporal direction: they follow their participants from the present into the future. However, randomized controlled trials, unlike prospective cohort studies, are not observational studies: they are interventional studies. This means that the investigators assign subjects to the exposure groups, and in an RCT they do this by a random process.

18 Randomized controlled trials (continued)
All randomized controlled trials are interventional studies, but not all interventional studies are randomized controlled trials. Interventional studies that use nonrandom procedures to assign exposure are called quasi- experiments. Some authors describe randomization as a key feature of experimental studies, but the term experiment typically connotes a study conducted in a highly controlled setting, such as a laboratory.

19 Randomized controlled trials (continued)
What is so special about randomization? Recall that confounding occurs when an independent risk factor is unequally distributed between exposure groups. Randomization helps to ensure that there is no inequality in the distribution of extraneous disease determinants. The law of large numbers helps to ensure that all extraneous variables will be equally distributed between the exposure groups.

20 Randomized controlled trials (continued)
Restriction is another strategy to control confounding. By simply eliminating participants exposed to a potential disease determinant, the possibility of confounding by those determinants is eliminated. Randomized controlled trials tend to liberally employ restriction, before randomization. This affects the generalizability of the results, such that these studies are viewed as assessing efficacy of treatments rather than effectiveness.

21 Randomized controlled trials (continued)
It is sometimes said that randomization also controls for selection bias. There is some truth in this assertion. For selection bias to occur, a process affecting participation in a study must unfold in a way that depends both on exposure and disease. Normally, a prospective cohort study is protected from such bias because the outcome has not occurred at the time of selection. Since randomization ensures that the allocation of exposure does depend on outcome, it prevents selection bias from occurring.

22 Randomized controlled trials (continued)
Even though randomization provides protection against selection bias, attrition can still lead to the occurrence of such bias. An interesting way of dealing with attrition in randomized trials is to conduct an intention-to-treat (ITT) analysis. This means that everyone randomized is included in the analysis, even if they do not actually comply with the treatment or if they leave the study. This requires the use of an imputation procedure to complete the data set, such as “carrying forward” the last available observation. This strategy is believed to minimize selection bias resulting from attrition.

23 Randomized controlled trials (continued)
Measurement bias can compromise the validity of trials. Trials must use blinding to prevent raters, who classify outcome status, from knowing treatment status. Otherwise, differential misclassification bias may occur. Trials should be double blind, which means that both participants and the trial staff are blinded. This will prevent the bias that could otherwise occur if outcome raters tended to rate randomized groups differently.

24 Randomized controlled trials (continued)
Randomized controlled trials are widely viewed as the “gold standard” study design. However, they are only feasible when intervention is feasible and they can suffer from selection bias due to attrition and a lack of generalizability. Specialized procedures such as ITT and blinding are required to safeguard against bias.

25 Case-cohort studies A drawback of the prospective cohort study design is its inefficiency. It needs a large investment of resources to track cohorts over time—and, typically, only a small proportion of a cohort will ever develop the outcomes under investigation. The case-cohort study design seeks to improve efficiency by making a subcohort its focus, rather than the entire cohort.

26 Case-cohort studies (continued)
A case-cohort study compares a series of cases to a subcohort; the subcohort is drawn from the larger cohort that gives rise to the cases. As the name suggests, the study design is a kind of hybrid between case-control and cohort methodologies .

27 Case-cohort studies (continued)
The subcohort represents the entire cohort from which the cases arose. This is done by selecting the subcohort from the population at risk at the start of the cohort’s follow-up interval. Typically, such studies are analytic in their orientation and therefore select incident cases for their case groups. The subcohort does not include prevalent cases. But since the subcohort is selected at the beginning of follow-up, some of its members may develop the disease during follow-up .

28 Ecological studies Ecological studies are a distinct type of study design, differing from all of those discussed so far. Ecological studies use a unit of analysis not based on individual people. The unit of analysis in ecological studies consists of groups of people, such as the population of neighbourhoods, cities, or countries.

29 Ecological studies (continued)
Ecological studies assess correlation between exposure and disease, both measured at an aggregate level. Because correlation is so often the method of analysis in such studies, they are sometimes called correlational studies. Examples of aggregate measures of exposure include: mean sodium consumption per person in different countries, or average number of grams of fish consumed in different provinces. Typical aggregate outcomes might be age-standardized mortality rates in different countries or cancer incidence in different provinces.

30 Ecological studies (continued)
Sometimes, investigators are interested in evaluating etiological hypotheses through ecological studies. In other words, they are not merely concerned with aggregate statistics as an easily accessible proxy for individual exposures, but with characteristics that are best conceptualized at an aggregate level such as income inequality.

31 Ecological studies (continued)
A concern with these studies is the ecological fallacy. This concept emphasizes the danger of making inferences about individual people based on correlations between aggregate units of individuals. For example, the observation that higher rates of admission for psychotic disorders tend to occur in areas with a higher percentage of immigrants does not necessarily mean that immigrants have higher rates of admission. Individual-level data would be required to confirm this.

32 Ecological studies (continued)
Ecological studies also have a very limited ability to address the issue of confounding, owing to their lack of individual-level data. A large drawback to ecological studies is that their target of estimation, usually a correlation, is not easily interpreted in terms of risk and probability. A correlation between aggregate exposures and outcomes doesn’t have the same intuitive meaning as many of the other parameters encountered in the epidemiological literature.

33 End


Download ppt "Instructor Resource Chapter 16 Copyright © Scott B. Patten, 2015."

Similar presentations


Ads by Google