Predicting the Climate Impact of Healthcare Facilities Using Gradient Boosting Machines.

Area of Science:

• Environmental science
• Health informatics
• Machine learning

Background:

• Healthcare contributes significantly to U.S. greenhouse gas (GHG) emissions (9-10%).
• Hospital-level GHG emission monitoring is crucial for sector decarbonization.
• Accurate data collection for emissions estimation is challenging, particularly for smaller hospitals.

Purpose of the Study:

• To explore the efficacy of gradient boosting machines (GBM) in imputing missing resource consumption data for GHG emission calculations in hospitals.
• To assess the feasibility of using readily available administrative data for this imputation process.

Main Methods:

• Utilized a 2020 survey dataset from 283 hospitals in Practice Greenhealth.
• Employed gradient boosting machines (GBM) to impute missing values for electricity, beef, and desflurane consumption.
• Used administrative data accessible to most hospitals for the imputation models.

Main Results:

• GBM successfully predicted electricity and beef consumption (R²=0.82) and desflurane use (R²=0.51).
• Estimated GHG emissions from electricity, beef, and desflurane totaled over 3 million metric tons of CO₂ equivalent (MTCO₂e) across the hospitals.
• Electricity consumption represented the largest carbon footprint (2.4 MTCO₂e), followed by beef (0.6 million MTCO₂e) and desflurane (0.03 million MTCO₂e).

Conclusions:

• Gradient boosting machines offer a viable approach for imputing missing hospital resource consumption data to estimate GHG emissions.
• This method can help individual hospitals estimate their total emissions and refine data collection strategies.
• The approach supports the development of targeted interventions for healthcare sector decarbonization.