Epidemiology Week 3: Confounders and how to reduce

the four flaws of research studies

• confounding, selection bias, experimenter bias, social desirability bias

• can lead to incorrect conclusions

correlation does not (BLANK) causation

imply nor prove

spurious relationship

correlation w/o causation

confounder variable

an external, uncontrolled factor in a study that influences both the independent and dependent variables, creating a misleading or inaccurate association between them

a variable that can lead to a spurious relationship

what term is known as the cause?

exposure

what term is known as the effect?

outcome

common confounders in dentistry

• age

• Gender

• Economic status

• Smoking status

• Number of times brushing per day

• Number of times visiting the dentist per year

• Further medical/dental characteristics (e.g. periodontitis, amount of plaque, BMI, diabetes)

five ways to reduce/avoid confounding

• restriction

• matching

• randomization

• stratification

• multivariate analysis

which strategy to reduce confounding?

• limit the study to indivs who fall w/in a specified category of each confounder

• ex: all same age and non smokers

restriction

which strategy disadvantage?

• reduce generalizability (finding may only apply to chosen group)

• might not find enough participants to fit in said category

restriction

which strategy to reduce confounding?

• each person has a counterpart in the other group w same age and status

• ex: group 1: taking regimen A group 2: taking regimen B but both 45 yo non smokers

matching

which strategy disadvantage?

• hard to find people who align perfectly

matching

which strategy to reduce confounding?

• randomly assign subjects to regimen A and B

• groups are likely approximately balanced in terms of potential confounders not previously considered

randomization

which strategy disadvantage?

• no guarantee groups are adequately balanced; unlucky

  • problem of small sample sizes

• not always ethical/feasible

  • can’t make people smoke

  • can’t assign people a certain age

randomization

which strategy to reduce confounding?

• separate analysis for each combination of confounders

• ex: allow people of diff ages and smoking statuses in the study then break into subdivisions when analyzing and provide separate results; regimen A and B and then each combo of confounders like age and smoking

stratification

which strategy disadvantage?

• deciding subdivisions/groups: what is young? why these categories

• different results with different subdivision cutoffs

• is a cut off point arbitrary

• possible small sample size in each subdivisions; main divisions like age maybe be equal but smokers vs non in subdivisions may be unequal

stratification

which strategy to reduce confounding compares “like vs like” making it fair?

stratification

which strategy to reduce confounding can you see which regimen does better for subdivisions?

stratification

which strategy to reduce confounding?

• get data on all variables including (multiple) confounders and plug the data into a formula

  • multiple linear regression

  • multiple logistic regression

• ex: allow people of all ages and smoking status, then when analyzing use advanced statistical methods to solve the confounding problem

multivariate analysis

which strategy disadvantage?

• if study design is bad, statistical formula cannot save the study

multivariate analysis

limit study to indiv who fall w/in specified category of each confounder; all 30 yo non smokers

restriction

select subjects such that each person in one group has a counterpart in the other group; same age and smoking status

matching

base group assignments on chance

randomization

separate analysis for each combo of confounders; subdvidisons of confounders

stratification

adjust for confounders using a single formula

multivariate analysis