Real Data Applications of Learning Curves In Cardiac Devices and Procedures.

Area of Science:

• Medical device safety
• Health services research
• Biostatistics

Background:

• Learning effects significantly influence medical device procedure outcomes and safety.
• Accurate estimation of learning curves is critical for medical device safety surveillance.
• Physician learning rates vary within institutions, necessitating hierarchical modeling.

Purpose of the Study:

• To evaluate a flexible method for modeling learning effects in medical device procedures.
• To demonstrate the method's applicability across diverse datasets and procedures.
• To improve the accuracy of safety surveillance by capturing physician learning hierarchies.

Main Methods:

• Employed a unique learning curve modeling approach incorporating institutional and physician hierarchies.
• Utilized generalized estimating equations (GEE) for hierarchical data analysis.
• Applied the model to new datasets including off-pump coronary artery bypass (CABG) and radial access procedures, with mediation analyses.

Main Results:

• The choice of learning curve shape (power series, logarithmic, exponential) is dataset-specific, influencing the modeling of fast or slow learning.
• Power series and logarithmic shapes are generally better for slower learning, while exponential shapes suit faster learning.
• Mediation analysis shows promise for adapting learning curve modeling.

Conclusions:

• The developed method demonstrates flexibility in various applications, including those with multiple procedures per patient.
• Accurate capture of learning curve rates for physicians nested within institutions is achievable.
• This approach can enhance medical device and procedure safety surveillance across the board.