Objective:

To systematically evaluate the methodological quality and clinical applicability of prediction models for postoperative cognitive dysfunction (POCD) in adults, highlighting the uncertainty in their clinical relevance.

Key Findings:

• 13 studies comprising 14 prediction models were included from 2,060 initial records, with sample sizes ranging from 82 to 687.
• Models were developed using logistic regression (8 models) and machine learning (6 models).
• Reported discriminatory performance was high (AUC range: 0.710–0.973), primarily from internally validated models, raising concerns about generalizability.
• Common predictors included age (10 models) and preoperative hemoglobin concentration (4 models).
• High risk of bias was noted due to insufficient sample sizes and lack of external validation, which may limit the applicability of these models.

Interpretation:

Current prediction models for POCD show promising discriminatory performance but are limited by methodological issues such as short follow-up periods and lack of external validation, which may hinder their clinical applicability.

Limitations:

• Limited sample sizes in studies.
• Models primarily validated internally rather than externally.
• Incorporation of non-routine predictors may affect generalizability and complicate clinical implementation.

Conclusion:

Future research should focus on large prospective cohorts, longitudinal predictors, extended follow-up, and rigorous external validation to enhance the clinical applicability of POCD prediction models, addressing the identified methodological issues.

Sources:

• Systematic Review Registration