ISD5 Research Design Lecture

Hierarchy of Evidence

Meta-analysis: →Statistical merging of individual studies to estimate pooled effect

Systematic Review: →Review multiple studies to assimilate findings of primary research

Randomised Controlled Trial: →Participants randomised into intervention or controls

Cohort: →Identify participants and assess their outcome of interest

Case-Control: →Identify Cases with disease and match to controls

Case Report: →individual case findings and management

Expert Opinion: →People agree on what is considered best practice

Secondary Studies (Data)

• Meta-analysis

• Systematic Review

Primary Studies (Data)

• Randomised Controlled Trial

• Cohort

• Case Control

Descriptive Studies

• Case Reports

• Cross-Sectional Studies

• Case Series

• Ecological Studies

Analytical/Observational Studies

• Case-Control

• Cohort

• Ecological Studies

Analytical/Experimental Studies

• Randomised CT

• Clinical Trials

• Quasi-experimental

True Experimental Design

1. Introduce an intervention or manipulate a variable 

2. Perform randomisation of participants 

3. Include an experimental control group

Randomised Controlled Trial

• Not feasible to conduct research with everyone 

• The sample should represent the entire population 

• Randomisation ensures confounding factors are evenly distributed 

• Statistical "power" refers to the probability of actually detecting an effect if there is a true effect to detect.

Randomised Control Trial Phases

Phase 0 – Earliest of drug trials – small number with low dose of drug. 

Phase 1 – How much rug is safe to give, what are the side effects, does it help treat the disease 

Phase 2 – Does the drug work well enough for larger phase 3 trial, which disease the drugs work best for, more information on side effects, better evidence for dose 

Phase 3 – Large trials, comparing new drug to current standard of care. 

Phase 4 – After drug has been licensed, investigates longer term effects and more wider use

Primary Hypothesis

• The hypothesis must be tested during the research.

Null HYpothesis

• No association between variable and disease

Independent Variable

• the altered/changed variable

CAUSE

Dependent Variable

• the variable being tested and measured

EFFECT

Primary Outcome Measure

• The outcome that is being investigated 

• Defined before start 

• Contributes to sample size calculation 

• Avoid collecting data and then performing statistical analysis 

DATA DREDGING

Interventions

• Study group which receives the intervention or manipulation of the variable

Controls

• Study group which does not receive an intervention or where the intervention cannot influence the dependent variable (Placebo).

Correlative

 Identifies if there are correlations or associations between different aspects of the study population.

Manipulative

Experimenter changes a variable to assess its impact on an outcome variable

Confounding Factors

• Variables that may falsely create an apparent association which doesn’t exist or which hide actual associations

Strategies for avoiding confounding factors:

1. Exclude: Selection criteria 

2. Randomise: so confounders will be evenly shared between groups 

3. Statistics: Identify the confounders and statistically account for them

Randomisation

• Participants can be assigned to all groups 

• If groups aren’t balanced, then results are biased and not representative 

• Minimises bias 

• Simple 

• Block 

• Stratified

Blinding/Masking

• Hiding the grouping/treatment from participant, investigator, statistician 

Single: Investigator or Participant

 Double: Investigator and participant