Business Context
A public health institute is analyzing a dataset of 1 million patients to estimate the effect of a specific lifestyle change ('Active') vs 'Sedentary' on recovery days. Due to the data volume, they opt for the Inverse Probability Weighting (IPW) method for efficiency.
About the Set : causalanalysis
Causal inference analysis and effect estimation.
Discover all actions of causalanalysis Data Preparation
Simulation of a large dataset (1 Million rows) to test performance and stability.
Copied!
| 1 | |
| 2 | DATA mycas.health_study; |
| 3 | call streaminit(999); |
| 4 | DO i = 1 to 1000000; |
| 5 | bmi = rand('NORMAL', 25, 4); |
| 6 | age = rand('UNIFORM', 18, 80); |
| 7 | IF rand('UNIFORM') < (1 / (1 + exp(-( -1 + 0.02*age)))) THEN lifestyle = 'Active'; |
| 8 | ELSE lifestyle = 'Sedentary'; |
| 9 | recovery_days = 14 - 2*(lifestyle='Active') + 0.1*bmi + rand('POISSON', 2); |
| 10 | OUTPUT; |
| 11 | END; |
| 12 | |
| 13 | RUN; |
| 14 |
Étapes de réalisation
1
Pre-calculation of Propensity Scores (Treatment Probabilities) using Logistic Regression.
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | logistic TABLE={name='health_study'}, class={'lifestyle'}, model={depvar='lifestyle', effects={'age', 'bmi'}}, OUTPUT={casout={name='health_scored', replace=true}, copyVars='ALL', predProbs=true}; |
| 4 | |
| 5 | RUN; |
| 6 |
2
Execution of caEffect using IPW method on the large dataset.
Copied!
| 1 | |
| 2 | PROC CAS; |
| 3 | causalanalysis.caEffect TABLE={name='health_scored'}, method='IPW', treatVar={name='lifestyle'}, outcomeVar={name='recovery_days', type='CONTINUOUS'}, pom={{trtLev='Active', trtProb='_PredProbs_'}, {trtLev='Sedentary', trtProb='_PredProbs_'}}, difference={{evtLev='Active', refLev='Sedentary'}}; |
| 4 | |
| 5 | RUN; |
| 6 |
Expected Result
The action processes the 1 million records efficiently without memory errors. It returns the 'POMs' table utilizing the pre-calculated probability weights (_PredProbs_) to adjust for the confounding variables (Age, BMI) via IPW.