This repository contains scripts and datasets for our paper 'Comparisons of the Treatment and Side Effects of Several Bariatric Surgery Procedures: An Observational Study via Random Forest-based and Neural Network-based Approaches'. In this work, we study the effect and risk of different bariatric surgery procedures from a causal inference view.
This study investigates the causal impact of bariatric surgery on various health outcomes using observational data from the MBQIP (Medicare Beneficiary Quality Improvement Project). Bariatric surgery encompasses several prominent procedures, including Sleeve Gastrectomy, Roux-en-Y Gastric Bypass (RYGB), Adjustable Gastric Band (AGB), Biliopancreatic Diversion with Duodenal Switch (BPD/DS), and Single Anastomosis Duodeno-Ileal Bypass with Sleeve Gastrectomy (SADI-S). In this analysis, we consider Sleeve Gastrectomy as the baseline procedure and evaluate the effects and risks associated with the other procedures from a causal perspective.
We employ state-of-the-art average treatment effect (ATE) estimation methods - causal forest and double machine learning from EconML, and rewrite the method - DragonNet - in pytorch by ourselves. We apply these methods to the MBQIP datasets to estimate the average treatment effect (ATE) for changes in BMI before and after surgery as the variable to study the effects, while we estimate the relative risk (RR) for the probability of death, intervention, readmission, and reoperation to study the risks.
We post our Python scripts to conduct the mentioned methods and the MBQIP datasets used in this work.
- Python 3.8+
- numpy
- pandas
- scikit-learn
- tensorflow
- keras
- econml
- dragonnet
-
Clone this repository:
-
Set up a virtual environment and activate it:
-
Install the required dependencies:
pip install -r requirements.txt
This repository contains code for analyzing causal risks and treatment effects. The repository is organized into several folders for ease of use:
risk/: All causal risk-related calculations can be found here.effect/: All treatment effect-related calculations and models are stored here.utils/: Handy tools and utilities for performing calculations and other common tasks.dataset/: This folder contains the datasets. Please preprocess the data before usage.
-
Preprocess the data before running any analysis.
-
Navigate to the relevant folder (
risk/oreffect/) and run the desired script:cd risk/ # or cd effect/ bash script_name.sh
The results of the causal inference study, including estimated treatment effects and performance metrics, will be saved in the results directory. Additionally, detailed logs and plots will be provided for further evaluation.
This project is licensed under the MIT License.