Table 1 Open in new tab Recommendations...

Recommendations for observational studies

Be transparant about the variable selection method

Detail automated selection procedure (forward, backward, which in-exclusion criteria)

Select the appropriate covariate adjustment method for your purpose

• Baseline covariates should be included irrespective of univariate significance with outcome or collinearity.

Add sensitivity analyses for robustness of the results

Evaluate the validity of the results

Detail handling of multiplicity

Detail handling of missing data

Table 1

Recommendations for covariate adjustment in observational studies

Recommendations for observational studies
Be transparant about the variable selection method
Motivation of pre-specified variables
Detail automated selection procedure (forward, backward, which in-exclusion criteria)
Detail threshold in a univariate analysis for inclusion
Select the appropriate covariate adjustment method for your purpose
Regression model:
• Allows for identifying prognostic/predictive covariates
• Be mindful of potential over-fitting, which leads to bias
• Be mindful of the correct functional form
• Covariates measured after treatment should not be included
Propensity score:
• Does not allow for identifying prognostic/predictive covariates
• Baseline covariates should be included irrespective of univariate significance with outcome or collinearity.
• Covariates measured after treatment should not be included
Add sensitivity analyses for robustness of the results
Vary method for covariate adjustment
Vary variable selection method
Vary number of covariates
Evaluate the validity of the results
Assess validity of model assumptions (proportional hazards, functional form, …)
Assess influence of extreme observations
Assess multi-collinearity
Detail handling of multiplicity
Detail handling of missing data

Recommendations for observational studies
Be transparant about the variable selection method
Motivation of pre-specified variables
Detail automated selection procedure (forward, backward, which in-exclusion criteria)
Detail threshold in a univariate analysis for inclusion
Select the appropriate covariate adjustment method for your purpose
Regression model:
• Allows for identifying prognostic/predictive covariates
• Be mindful of potential over-fitting, which leads to bias
• Be mindful of the correct functional form
• Covariates measured after treatment should not be included
Propensity score:
• Does not allow for identifying prognostic/predictive covariates
• Baseline covariates should be included irrespective of univariate significance with outcome or collinearity.
• Covariates measured after treatment should not be included
Add sensitivity analyses for robustness of the results
Vary method for covariate adjustment
Vary variable selection method
Vary number of covariates
Evaluate the validity of the results
Assess validity of model assumptions (proportional hazards, functional form, …)
Assess influence of extreme observations
Assess multi-collinearity
Detail handling of multiplicity
Detail handling of missing data