Estimating misclassification error in a binary performance indicator: case study of low value care in Australian...

Area of Science:

• Health Services Research
• Health Informatics
• Quality Improvement

Background:

• Hospital administrative data indicators are prone to misclassification errors, particularly when relying on detailed clinical information.
• Accurate identification of low-value care is crucial for healthcare quality improvement and resource optimization.

Purpose of the Study:

• To develop and apply a modified logistic regression modeling approach to assess misclassification rates (false-positive and false-negative) of low-value care indicators.
• To estimate the true prevalence of low-value care by adjusting for identified misclassification errors.

Main Methods:

• Utilized a New South Wales administrative dataset (2012-2015) for 19 low-value care indicators.
• Fitted four logistic regression models (no misclassification, false-positive only, false-negative only, both) for each indicator.
• Assessed model fit using posterior probabilities to estimate misclassification rates.

Main Results:

• False-positive rates were generally low, indicating high confidence when an indicator flagged care as low value.
• False-negative rates were higher and less precisely estimated.
• The initial estimated low-value care rate of 12% increased to 35% after adjusting for misclassification errors.

Conclusions:

• Modified logistic regression modeling offers a valuable initial validation step for administrative data indicators, reducing the need for extensive chart reviews.
• This approach helps identify and revise potentially flawed indicators before resource-intensive validation.
• Adjusting for misclassification significantly alters the estimated prevalence of low-value care, highlighting the importance of this modeling technique.