Optimization of anesthetic decision-making in ERAS using Bayesian network

Yu‐Wen Chen, Yiziting Zhu, Kunhua Zhong, Zhiyong Yang, Yujie Li, Xin Shu, Dandan Wang, Peng Deng, Xuehong Bai, Jianteng Gu, Kaizhi Lu, Zhang Ju, Lei Zhao, Tao Zhu, Ke Wei, Yi BinFrontiers in Medicine 2022 · DOI

TL;DR

Researchers have developed a novel Bayesian Network (BN) approach to optimize anesthetic decision-making within Enhanced Recovery After Surgery (ERAS) protocols. By introducing a new structure-learning algorithm called LSHCT, the study successfully mapped the complex, causal dependencies between specific anesthesia choices and patient outcomes, such as hospital stay duration and total medical costs.

What Problem Does This Paper Address?

Enhanced Recovery After Surgery (ERAS) is a multidisciplinary approach designed to minimize the body's stress response to surgery, leading to faster recovery. While ERAS is widely used, the "anesthesia" component remains a bit of a black box. Clinical decisions—like which painkillers to use or how to manage fluids—are often made in isolation, even though they likely interact in complex ways.

Traditional statistical methods, like logistic regression, often fail to capture these intricate, non-linear dependencies and uncertainties. The authors address this gap by using Bayesian Networks (BNs) to provide a "white-box" graphical model that anesthesiologists can use to understand why certain combinations of decisions lead to better or worse outcomes.

Methods

The authors proposed a multi-stage pipeline to move from raw clinical data to a functional decision-support model:

  1. Feature Selection: They first used statistical tests (T-tests, Chi-square) and a simplified BN to identify 18 key indicators (out of an initial 50+) that significantly impacted outcomes.
  2. LSHCT Algorithm: The core innovation is a new structure-learning method: Strongly Connected Components Local Structure determination by Hill Climbing Twice (LSHCT).
  3. It uses a "divide-and-conquer" strategy: calculating mutual information between nodes, building small "sub-BNs," and then fusing them.
  4. It identifies Strongly Connected Components (SCCs) and applies the Hill Climbing algorithm twice to determine the direction of causal edges (e.g., does Decision A cause Outcome B, or vice versa?).
  5. Expert Refinement: Recognizing that pure data-driven models can sometimes suggest clinically nonsensical links, the resulting Directed Acyclic Graph (DAG) was manually adjusted by experts to ensure clinical validity.

Datasets

The study utilized a real-world clinical dataset from three major hospitals in China (Southwest Hospital, Xuan Wu Hospital, and West China Hospital).

  • Initial Pool: 49,768 surgical patients.
  • Final Cohort: 1,827 patients who underwent laparoscopic surgery for benign gynecological tumors.
  • Outcomes: The model focused on two primary targets: Postoperative Length of Stay (LOS) and Total Cost (TC).

Key Insights and Ideas

  • The "Positive Balance" Hub: One of the most significant findings was that "positive balance" (fluid management) acted as a "parent node" in the network. It directly influenced fluid ratios and urine volume, which in turn affected the length of stay.
  • Interconnected Decisions: The model revealed that anesthesia decisions are not independent. For example, the choice to use opioids for postoperative analgesia was linked to ETCO2 levels (end-tidal carbon dioxide), which together influenced recovery speed.
  • Cost vs. Recovery: The factors influencing cost were largely related to drug choices (like the use of dexmedetomidine), whereas recovery speed was more closely tied to physiological management (blood pressure and fluid balance).
  • Interpretability Matters: Unlike "black-box" neural networks, the BN provides a DAG that allows clinicians to trace the path from a decision to an outcome, making it a more trustworthy tool for medical professionals.

Key Results

The study identified specific "optimal states" for anesthesia indicators:

  • For Length of Stay (LOS): High urine volume/length of operation ratio, normal mean blood pressure, and avoiding postoperative compound nerve blocks were associated with a shorter stay (< 4.5 days).
  • For Total Cost (TC): The use of dexmedetomidine and non-steroidal anti-inflammatory drugs (NSAIDs) were pivotal indicators. Interestingly, while some drugs are expensive, they can reduce total costs by preventing complications or shortening recovery.
  • Probabilistic Success: The model identified combinations of indicators that resulted in a >90% probability of achieving the "optimal state" (lower cost and shorter stay), providing a roadmap for individualized anesthetic protocols.

Limitations and Future Work

The authors acknowledge several limitations:

  • Narrow Scope: The data is limited to benign gynecological tumors; the model's performance on malignant cases or other surgical specialties is unknown.
  • Missing Preoperative Data: The study focused heavily on intraoperative and postoperative factors, leaving out preoperative patient optimization which is a core pillar of ERAS.
  • Outcome Isolation: LOS and TC were analyzed as separate networks. Future work aims to integrate multiple outcomes into a single, unified model to better understand the trade-offs between cost and recovery speed.

Summary updated

Abstract

Enhanced recovery after surgery (ERAS) can accelerate patient recovery. However, little research has been done on optimizing the ERAS-related measures and how the measures interact with each other. The Bayesian network (BN) is a graphical model that describes the dependencies between variables and is also a model for uncertainty reasoning. In this study, we aimed to develop a method for optimizing anesthetic decisions in ERAS and then investigate the relationship between anesthetic decisions and outcomes. First, assuming that the indicators used were independent, the effects of combinations of single indicators were analyzed based on BN. Additionally, the impact indicators for outcomes were selected with statistical tests. Then, based on the previously selected indicators, the Bayesian network was constructed using the proposed structure learning method based on Strongly Connected Components (SCC) Local Structure determination by Hill Climbing Twice (LSHCT) and adjusted according to the expert's knowledge. Finally, the relationship is analyzed. The proposed method is validated by the real clinical data of patients with benign gynecological tumors from 3 hospitals in China. Postoperative length of stay (LOS) and total cost (TC) were chosen as the outcomes. Experimental results show that the ERAS protocol has some pivotal indicators influencing LOS and TC. Identifying the relationship between these indicators can help anesthesiologists optimize the ERAS protocol and make individualized decisions.
BibTeX
@inproceedings{chen2022optimization,
  author = {Yu‐Wen Chen and Yiziting Zhu and Kunhua Zhong and Zhiyong Yang and Yujie Li and Xin Shu and Dandan Wang and Peng Deng and Xuehong Bai and Jianteng Gu and Kaizhi Lu and Zhang Ju and Lei Zhao and Tao Zhu and Ke Wei and Yi Bin},
  title = {{Optimization of anesthetic decision-making in ERAS using Bayesian network}},
  booktitle = {Frontiers in Medicine},
  year = {2022},
  doi = {10.3389/fmed.2022.1005901},
}
Figure