Exploring factors associated with pressure ulcers: a data mining approach.

Area of Science:

• Healthcare analytics
• Predictive modeling in medicine
• Nursing research methodologies

Background:

• Pressure ulcers significantly increase in-hospital mortality by nearly threefold.
• Existing prediction methods like the Braden scale may not capture all risk factors.
• Data mining techniques offer potential for enhanced pressure ulcer risk assessment.

Purpose of the Study:

• To address the gap in nursing research regarding data mining applications.
• To review, explain, and compare data mining models for pressure ulcer prediction.
• To identify patient-level factors associated with pressure ulcers using a US military hospital dataset.

Main Methods:

• Comparison of logistic regression, decision trees, random forests, and multivariate adaptive regression splines.
• Utilized easily accessible demographic and medical measurement data.
• Evaluated models based on performance and interpretability.

Main Results:

• The random forests model demonstrated the highest predictive accuracy (C-statistic).
• Key predictors identified were: days in hospital, serum albumin, age, blood urea nitrogen, and total Braden score.
• Ranked importance of variables for pressure ulcer prediction was established.

Conclusions:

• Random forests, a data mining technique, are effective for predictive modeling of pressure ulcers.
• Hospitals should leverage their own historical data for tailored risk prediction.
• Risk models should incorporate factors beyond the Braden scale and be specific to patient populations.