02.Causation04a.Heterogeneity of effect subgroup analysis - sporedata/researchdesigneR GitHub Wiki

1. Use cases: in which situations should I use this method?

  • Mediation analysis is conducted when there is a "variable in the middle" between a cause and an outcome. For example, low hematocrit after cardia surgery leads to transfusion, which leads to postoperative complications.

  • Moderation (also known as interaction) is used when two simultaneously present risk factors increase the risk of an outcome at a level that is greater than the sum of each individual risk factor. For example, diabetes and smoking might increase the risk of a postoperative complication by 10%, but when they both occur in the same individual the risk is of 30%.

  • When a large number of interactions are of interest, one can make use of logic regresssion models (not to be confused with logistic models) - see The use of Logic regression in epidemiologic studies to investigate multiple binary exposures: an example of occupation history and amyotrophic lateral sclerosis

2. Input: what kind of data does the method require?

  1. Pre-requisits for casual-analysis.
  2. Literature to back up the hypothesis that the effect of two predictors interact, or that a potential confounder is a mediator in the causal relationship between a predictor and an outcome

3. Algorithm: how does the method work?

Model mechanics

Mediation analysis will elucidate the causal role from the risk factor (low hematocrit) and the mediator (transfusion) in causing the outcome (complications).

The most common type of moderation evaluated in the literature is multiplicative, which means that in a regression model the variables will be represented by the multiplication of both risk factors. For example Y = risk1*risk2 + risk1 + risk2. Notice that each risk is represented both in the interaction as well as a separate variable.

Reporting guidelines

A framework for the analysis of heterogeneity of treatment effect in patient-centered outcomes research

  • PCORI has well-established guidelines for the evaluation of the heterogeneity of effect.

Data science packages

Suggested companion methods

Common companion models include models that can perform data-driven discovery of subgroup patterns preceding the modeling of heterogeneity of effects.

Learning materials

  1. Books
    • Introduction to Mediation, Moderation, and Conditional Process Analysis, Second Edition: A Regression-Based Approach [1].
  2. Articles
    • Confounding in Statistical Mediation Analysis: What It Is and How to Address It [2].
    • US Food and Drug Administration Approvals of Drugs and Devices Based on Nonrandomized Clinical Trials] [3].
    • Mediation analysis [4].
    • Common references for causation.
  3. Posts

4. Output: how do I interpret this method's results?

Mock conclusions or most frequent format for conclusions reached at the end of a typical analysis.

Tables, plots, and their interpretation

5. SporeData-specific

Templates

Data science functions

References

[1] Hayes AF. Introduction to Mediation, Moderation, and Conditional Process Analysis, Second Edition: A Regression-Based Approach. Guilford publications; 2017 Dec 13.

[2] Valente MJ, Pelham III WE, Smyth H, MacKinnon DP. Confounding in Statistical Mediation Analysis: What It Is and How to Address It. Journal of counseling psychology. 2017 Nov;64(6):659.

[3] Razavi M, Glasziou P, Klocksieben FA, Ioannidis JP, Chalmers I, Djulbegovic B. US Food and Drug Administration Approvals of drugs and devices based on nonrandomized clinical trials: a systematic review and meta-analysis. JAMA network open. 2019 Sep 4;2(9):e1911111-.

[4] Lee H, Herbert RD, McAuley JH. Mediation analysis. Jama. 2019 Feb 19;321(7):697-8.

⚠️ **GitHub.com Fallback** ⚠️