Confounders in Research & Epidemiology

Definition

"A confounder is a factor that is associated with the outcome (e.g., disease) and the exposure. The confounder may account for the apparent effect of the exposure on the disease or mask a true association."

• Berek & Novak's Gynecology

The 3 Criteria for a Confounder

Classic Examples

Example 1: Obesity and Heart Disease

• Exposure: Obesity
• Outcome: Heart disease
• Confounder: Age
• Age is related to both obesity (older people tend to be more obese) and heart disease (older people have more heart disease). If the obese group is older, the observed association between obesity and heart disease may partly reflect the effect of age, not obesity alone.
• Goldman-Cecil Medicine

Example 2: Ice Cream Sales and Violence (famous teaching example)

• Exposure: Ice cream sales
• Outcome: Violence/crime rates
• Confounder: Hot weather
• Hot weather causes both increased ice cream sales AND increased outdoor activity/irritability leading to more violent incidents. Ice cream does not cause violence - hot weather is the confounder.
• The Harriet Lane Handbook

Example 3: Foot Size and Reading Ability (in children)

• Exposure: Large foot size
• Outcome: Better reading ability
• Confounder: Age/grade level
• Older children have bigger feet AND better reading skills. Grade level confounds the apparent relationship.

Types of Confounding

Confounders vs. Effect Modifiers

• The Harriet Lane Handbook: "An effect modifier is a variable that modifies the observed effect of an exposure on disease. Can be controlled for by stratification."

How to Control for Confounding

At the Design Stage

1. Randomization - The best method. Random assignment of subjects to exposure groups ensures known AND unknown confounders are equally distributed between groups. This is why RCTs are the gold standard.
2. Restriction - Limit the study to a narrow range of the confounder (e.g., only study men aged 40-50). Eliminates confounding but limits generalizability.
3. Matching - Match each case to a control with the same value of the potential confounder (e.g., match by age and sex). Common in case-control studies.

At the Analysis Stage

1. Stratification - Analyze the exposure-outcome relationship separately within each stratum (subgroup) of the confounder. Produces stratum-specific estimates.
2. Multivariable analysis (regression) - Simultaneously adjust for multiple confounders using logistic regression, Cox regression, etc. Produces an adjusted odds ratio (OR) or relative risk (RR) that reflects the exposure-outcome association after accounting for the confounders included in the model.
3. Propensity score analysis - The probability of receiving the exposure is calculated based on subject characteristics. Used to balance groups in observational studies.

"A propensity score is the probability that a study subject would receive the exposure or treatment of interest, based on the subject's characteristics and the clinical environment. Using the propensity score, it is possible to adjust for the effect of known confounding variables."

• Goldman-Cecil Medicine

Important limitation: "All techniques to reduce confounding depend entirely on data about relevant covariates. If such data are missing, statistical adjustment cannot produce an accurate result." Statistical methods can only control for known and measured confounders. Unknown or unmeasured confounders remain a persistent threat in observational studies.

Directed Acyclic Graphs (DAGs)

• Identify which variables are true confounders
• Distinguish confounders from mediators and colliders
• Decide which variables to adjust for in analysis
• The Harriet Lane Handbook: "Directed acyclic graphs (DAGs) can be drawn to help assessment for confounders, effect modifiers, colliders, and other biases."

Colliders: A Special Case

Why Confounding Matters

• Confounding is the primary reason why observational studies (cohort, case-control, cross-sectional) can be misleading
• It is why an RCT result is considered stronger evidence than an observational study
• Failure to account for confounders can lead to incorrect clinical conclusions, flawed drug approvals, or misguided public health policies
• In observational studies, it is seldom safe to assume that pre-exposure factors are equally distributed between exposure groups - NCBI Bookshelf on confounding bias

Quick Summary

Confounder = associated with BOTH exposure AND outcome + NOT on causal pathway

Control methods:
  Design stage  → Randomization > Restriction > Matching
  Analysis stage → Stratification > Multivariable regression > Propensity score

Remember: You can only adjust for confounders you KNOW about and MEASURE.

Meta-Analysis

Definition

"Meta-analysis is a statistical method for drawing inferences from a sample of studies to derive a summary estimate and confidence interval for a parameter measured by the included studies. It allows for a formal combination of results while estimating and accommodating both the within-study and between-study variations."

• Goldman-Cecil Medicine

Systematic Review vs. Meta-Analysis

"A systematic review is a scientific investigation that focuses on a specific question and uses explicit, planned methods to identify, select, assess, and summarize the findings of similar but separate studies. A meta-analysis is the process of using statistical methods to quantitatively combine the results of similar studies... Thus a meta-analysis includes a systematic review, but a systematic review does not necessarily include a meta-analysis."

• Creasy & Resnik's Maternal-Fetal Medicine

Defining Features of a Systematic Review (the foundation)

1. A clear clinical question to be addressed
2. An extensive, explicit search strategy to find all eligible studies (published and unpublished)
3. Explicit inclusion/exclusion criteria for studies
4. A mechanism to assess risk of bias in each study
5. (Sometimes) statistical synthesis via meta-analysis

Why Meta-Analysis is Valuable

• Increases statistical power - by pooling data from many small studies, it can detect effects that no single trial could find on its own
• Improves precision - the pooled confidence interval is narrower than individual study CIs
• Resolves conflicting results - when individual trials disagree, meta-analysis provides an overall estimate
• Detects subgroup effects - stratified analyses can reveal who benefits most

"Meta-analysis can especially help detect benefits when individual trials are inadequately powered. For example, the benefits of streptokinase thrombolytic therapy in acute MI demonstrated by ISIS-2 in 1988 were evident by the early 1970s through meta-analysis."

• Harrison's Principles of Internal Medicine (22e)

The Forest Plot - Reading It

Figure: Forest plot from Creasy & Resnik's Maternal-Fetal Medicine (Cochrane data)

How to read a forest plot:

• Each of 11 small trials showed a non-significant trend toward benefit for antibiotics
• Pooled RR = 0.67 (95% CI: 0.52 to 0.85) - meaning antibiotics reduced neonatal infection by 33%, and the CI does not cross 1.0, so this IS statistically significant

Statistical Models: Fixed vs. Random Effects

"Fixed effect meta-analysis assumes that all available studies have estimated the same statistic... Random effects meta-analysis assumes that there may be many small differences between the different studies, and that it is not justified to assume that the effect is the same in every study."

• Tietz Textbook of Laboratory Medicine

Heterogeneity - The Key Challenge

Types:

• Clinical heterogeneity: Differences in patient populations, interventions, or outcome definitions
• Methodological heterogeneity: Differences in study design or risk of bias
• Statistical heterogeneity: Variation in results greater than expected by chance alone

Measuring heterogeneity:

"Although there are statistical tools for taking heterogeneity into account (i.e., random effects models), it is incumbent on the reader to make such a determination."

• Creasy & Resnik's Maternal-Fetal Medicine

Publication Bias

• Detection: The funnel plot - a scatter plot of effect size vs. study precision. If asymmetrical (missing studies in one corner), publication bias is likely.
• Prevention: Meta-analyses should actively seek unpublished "negative" trials (grey literature, trial registries)

"Ideally, unpublished trials should be identified and included to avoid publication bias (i.e., missing 'negative' trials that may not be published)."

• Harrison's Principles of Internal Medicine

Summary Measures Reported

• Pooled Relative Risk (RR) or Odds Ratio (OR) with 95% CI
• Absolute Risk Reduction (ARR)
• Number Needed to Treat (NNT) = 1/ARR - how many patients need treatment to prevent one outcome event

Limitations of Meta-Analysis

Quality Reporting Standards

• PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) - the guideline for transparently reporting meta-analyses, subscribed to by NEJM, Lancet, JAMA
• CONSORT - equivalent standard for individual RCTs
• Cochrane Library - produces high-quality systematic reviews and meta-analyses across medical topics

Position in the Evidence Hierarchy

Systematic Reviews & Meta-Analyses (highest level of evidence)
         ↑
Randomized Controlled Trials (RCTs)
         ↑
Cohort Studies
         ↑
Case-Control Studies
         ↑
Case Series / Expert Opinion (lowest)