Mednotes

L2 EBP Overview of study design

Properties2
tagsebp, lecture
aliasesL2 EBP Overview of study design

10 min read

Overview

This lecture gives a brief overview of epidemiology: the measures used to describe occurrence and association, two organising frameworks (the Public Health Model and the GATE frame/PECOT), and the structure, strengths and challenges of the four main clinical study designs (cross-sectional, cohort, case-control, randomised controlled trial). It then covers the components of internal validity (chance, bias, confounding), causation (Bradford Hill’s Guidelines), external validity, and research ethics (NEAC Guidelines).

What is epidemiology

  • Epidemiology is the study of the distribution and determinants of health-related states or events (including disease), and the application of this study to the control of diseases and other health problems (WHO, 2018).
  • Epidemiological research underpins all clinical disciplines and public health.

The Public Health Model

A four-step flow from problem to response:

  1. Defining & measuring the problem — informed by Who? Where? When?
  2. Describing causes & consequences — informed by What? and Why?
  3. Developing & evaluating interventions — addressing “What can we do about it?”
  4. Disseminating effective policy & practice.

Epidemiological measures

  • Occurrence/frequency measures: incidence proportion, incidence rate, prevalence.
  • Association/effect measures: relative risk, risk difference, odds ratio.
  • [slide does not elaborate calculation or interpretation here; flagged as covered in Tutorial 1]

Frameworks: GATE frame / PECOT

  • PECOT: Population, Exposure/Intervention, Comparison/Control, Outcome, Time.
  • Diagram structure: a source population narrows to a sample population, which splits into an exposed group and a comparison group, with a time arrow showing the direction of follow-up.
  • The standard exposure-by-outcome contingency table has cells a, b, c, d: rows are Outcome Yes (+) and Outcome No (-), columns are exposed vs comparison.

Study designs overview

  • Observational designs: cross-sectional, cohort, case-control.
  • Experimental design: randomised controlled trial.
  • Systematic reviews / meta-analyses are also part of the study design hierarchy.
  • Common problems associated with a study design do not necessarily mean they are present whenever that design is used.

Cross-sectional studies

  • Classification: observational; descriptive or analytical.
  • Measure calculated: prevalence.
  • Steps:
    • Define the source population.
    • Randomly select a sample of the source population.
    • At the same point in time, measure the number of participants with the specified exposures of interest (current or historical) and the number with the outcomes of interest.
    • Calculate the prevalence of exposures and outcomes at that single point in time for the sample population.
  • Usually measured using surveys or routinely collected data.
  • Cannot demonstrate evidence of causation, because exposure and outcome are measured simultaneously rather than exposure preceding outcome.

Cohort studies

  • Classification: observational; analytical.
  • Measures calculated: incidence rate, incidence proportion, relative risk, risk difference.
  • Steps:
    • Define the source population.
    • Randomly select a sample of the source population (participants must be outcome-free at the start).
    • Measure the exposure state of sample participants at the beginning of the study.
    • Group participants as exposed or comparison.
    • Follow up for a set period of time.
    • Count who develops the outcome during follow-up.
    • Calculate incidence measures, then relative risk and risk difference.
  • Strengths: can investigate multiple outcomes; can calculate incidence measures; can calculate measures of association (RR and RD); can provide evidence for temporal sequence; gives observational evidence of causation.
  • Challenges: loss to follow-up; long periods between exposure occurring and outcome developing; difficulty studying rare outcomes.

Case-control studies

  • Classification: observational; analytical.
  • Measure calculated: odds ratio (no measure of occurrence).
  • Participants are selected by outcome status (cases vs controls), not by exposure status; the outcome table is oriented with rows Cases/Controls and columns exposed/not exposed (cells a, c for cases; b, d for controls).
  • Steps:
    • Define the source population.
    • Select cases with the outcome of interest.
    • Select controls (participants without the outcome of interest) from the same source population that the cases belong to.
    • Measure the exposure status of both cases and controls.
    • Calculate the odds ratio.
  • Strengths: can investigate multiple exposures; good for investigating rare outcomes; good for outcomes with long periods between exposure and outcome; if the outcome is rare the OR approximates the RR (so the OR can be interpreted like an RR); gives observational evidence of causation.
  • Challenges: selection bias; information bias (recall bias); rare exposures are hard to study; only one outcome can be investigated per study.

Randomised controlled trials

  • Classification: experimental; analytical.
  • Measures calculated: incidence rate, incidence proportion, relative risk, risk difference.
  • Steps:
    • Define the source population.
    • Randomly select a sample of the source population.
    • Randomise sample participants into groups (intervention or control).
    • Confirm the randomisation process has been successful.
    • Apply protocols to the intervention and control groups.
    • Follow up participants for a defined period of time.
    • Measure the outcome.
    • Complete an intention-to-treat analysis.
    • Calculate measures of occurrence (incidence proportion and incidence rate) and measures of association (RR and RD).
  • Genuine (clinical) equipoise must exist for it to be ethical to randomise participants.
  • Strengths: successful randomisation controls for both known and unknown confounders; controls for some forms of selection bias; provides experimental evidence; a well-conducted RCT provides strong evidence of causation; uses intention-to-treat analysis.
  • Challenges: some residual confounding will still remain; randomisation does not control for all types of bias.

Internal validity

  • Chance (random error) affects precision. Toolbox: confidence intervals, p-values, sample size.
  • Bias (selection and information) affects accuracy: requires ways to limit bias entering a study, a step-by-step method to assess the impact of bias on a study measure, and consideration of whether a harmful or protective effect is being investigated.
  • Confounding affects accuracy: requires knowing the characteristics of a confounder, the possible effects a confounder may have on the measure of association, reducing confounding at the study design phase, and adjusting for confounding during data analysis.

Causation: Bradford Hill’s Guidelines

Mnemonic BESTCDS:

  • Biological plausibility
  • Experimental evidence
  • Strength of associations
  • Temporal sequence
  • Consistency with other studies
  • Dose response
  • Specificity

External validity

  • Generalisability: who can these results realistically be applied to?
  • What are the key messages of the research?
  • What should be done now, i.e. the implications of the research?

Ethics

NEAC Guidelines — underlying ethical considerations for research:

  1. Respect for persons
  2. Justice
  3. Beneficence and non-maleficence
  4. Integrity
  5. Diversity
  6. Addressing conflict of interest

Self-test

  1. Define epidemiology, including who publishes this definition and what the study is applied to.
  2. Describe the four steps of the Public Health Model, in order, and state the questions that inform the first two steps.
  3. List the two categories of epidemiological measures covered in this lecture, naming the specific measures under each.
  4. What does each letter of PECOT stand for in the GATE frame?
  5. In the standard exposure-by-outcome contingency table used in the GATE frame, what do the cells a, b, c and d represent?
  6. Distinguish observational from experimental study designs, naming which specific designs belong to each category.
  7. Describe the steps of a cross-sectional study.
  8. Which measure is calculated in a cross-sectional study, and why can this design not demonstrate causation?
  9. Describe the steps of a cohort study, from selecting the sample through to calculating measures of association.
  10. List the strengths of a cohort study.
  11. List the challenges of a cohort study.
  12. Describe the steps of a case-control study, including how cases and controls are selected.
  13. Distinguish a case-control study from a cohort study in terms of how participants are selected into the sample and which measure of association each design calculates.
  14. List the strengths of a case-control study.
  15. List the challenges of a case-control study.
  16. Describe the steps of a randomised controlled trial.
  17. What must exist before an RCT can ethically randomise participants?
  18. List the strengths of a randomised controlled trial.
  19. List the challenges of a randomised controlled trial.
  20. Distinguish chance from bias and confounding in terms of what aspect of a study result each affects, and give the toolbox used to address chance.
  21. Describe how confounding can be addressed, both at the study design stage and during data analysis.
  22. List Bradford Hill’s guidelines for assessing causation.
  23. What three questions define a study’s external validity/generalisability?
  24. List the six NEAC underlying ethical considerations for research.
  25. A researcher wants to study a rare occupational exposure and a rare disease with a long latency period between exposure and outcome. Which study design is most appropriate, and why?

Answers

Graph View

Backlinks