Prognostic prediction models for diabetic retinopathy progression: a systematic review

Area of Science:

• Ophthalmology
• Medical Informatics
• Public Health

Background:

• Diabetic retinopathy incidence and detection are rising, increasing demand for treatment services.
• Prognostic models exist for early detection but are mainly used in screening, not hospital referrals.
• The predictive ability and applicability of these models for high-risk patients need evaluation.

Purpose of the Study:

• To assess the predictive ability and applicability of existing prognostic models for high-risk diabetic retinopathy patients referred to hospitals.
• To identify models suitable for optimizing treatment services for this population.

Main Methods:

• A systematic literature search was conducted across multiple databases (MEDLINE, EMBASE, COCHRANE CENTRAL) and reference lists.
• Studies were screened for relevance, and data on model characteristics, predictive ability, and validation were extracted.
• Quality assessment was performed using PROBAST and CHARMS checklists by two independent reviewers.

Main Results:

• Twenty-two articles on 14 prognostic models were included; 11 had internal validation, 8 external validation.
• Discriminative ability (c-statistics) ranged from 0.57 to 0.91, with studies showing a high risk of bias.
• Most models focused on lower-risk patients, but three showed some applicability for higher-risk groups, though requiring updates and external validation.

Conclusions:

• Existing prognostic models for diabetic retinopathy have limitations in predicting outcomes for higher-risk patients in hospital settings.
• Significant risk of bias and lack of external validation impact the applicability of most models.
• Further updating and multi-center external validation are crucial before implementing these models in clinical practice for high-risk patients.