Constructing Causal Networks Through Regressions: A Tutorial.

Area of Science:

• Causal inference and network analysis in healthcare settings.
• Application of statistical modeling for adverse event analysis.

Background:

• Network models offer advanced causal analysis, yet their adoption by hospital risk managers is limited.
• Understanding root causes of adverse events is crucial for improving patient safety and hospital operations.

Purpose of the Study:

• To introduce causal networks and their analysis methods for hospital risk managers.
• To demonstrate how to construct causal networks using readily available regression software.
• To identify root causes of adverse events, using excessive emergency department boarding as a case study.

Main Methods:

• Causal networks represent cause-and-effect relationships, defined by temporal precedence, mechanistic pathways, and non-spurious associations.
• Construction involves iterative least absolute shrinkage and selection operator (LASSO) regression, identifying direct causes.
• Network parameters are estimated by fitting the structure to data, with a demonstration using simulated emergency department boarding data.

Main Results:

• A causal network accurately modeled simulated adverse event data, with 156 possible links precisely recovered.
• The analysis successfully identified direct and root causes of excessive emergency department boarding.
• Hospital occupancy rate was identified as the primary root cause, surpassing emergency department efficiency.

Conclusions:

• Causal networks provide valuable insights into the root and direct causes of adverse events.
• These models offer a framework for empirically testing the causes of adverse events in healthcare.
• Hospital risk managers are encouraged to adopt these network analysis methods for improved risk management.