Author: Dimitrios Damopoulos (Switzerland)
Co-authors: Luis Mendes, Torcato Santos, Ales Neubert, Beatriz Garcia Armendariz, Robert Weikert, Daniela Ferrara
Purpose
Use of artificial intelligence–based models to predict the risk of patients with non-proliferative diabetic retinopathy (NPDR) to develop clinically significant macular edema (CSME) could inform personalised monitoring and patient follow-up. Our objective was to develop and evaluate machine learning models employing systemic and/or retinal imaging features for predicting whether patients with mild NPDR will develop CSMEwithin 2 years from baseline.
Setting/Venue
Analysis of data from a prospective observational cohort study (NCT00763802) in patients with type 2 diabetes and mild NPDR (Diabetic Retinopathy Severity Scale [DRSS] 20–35).
Methods
Systemic and retinal imaging features from 348 patients (129 female; mean age, 60.9 years), without CSME at baseline, enrolled in the study were pooled. Systemic measurements included glycated haemoglobin; systolic and diastolic blood pressure; total, high-density lipoprotein and low-density lipoprotein cholesterol; and triglycerides. Retinal imaging features, acquired using optical coherence tomography (Stratus) and stereoscopic 7-field colour fundus photographs, included central subfield thickness; inner ring temporal, superior, inferior and nasal thickness; number of microaneurysms; and 6-month microaneurysm turnover. Data were collected at 6 and 24 months from baseline. Linear regression models were evaluated for predicting these events using systemic data only, imaging features only and both combined. Area under the receiver operating characteristic (AUROC) curve was employed as a performance metric. Mean values and 95% CIs were computed over all the iterations of 10 repeats of a 5-fold cross-validation setup.
Results
Of 348 patients without CSME at baseline, 12 developed CSME in the study eye by month 6, and 34 patients by month 24. When using data obtained at baseline, CSME at month 24 was predicted with an AUROC = 0.550 (95% CI, 0.521, 0.580) with systemic data, 0.727 (95% CI, 0.700, 0.753) with imaging features and 0.734 (95% CI, 0.708, 0.760) with both combined. When using data from month 6, the 6 patients who developed CSME by month 6 were excluded. CSME at month 24 for the remaining patients was predicted with an AUROC = 0.536 (95% CI, 0.500, 0.573) with systemic data, 0.734 (95% CI, 0.702, 0.766) with imaging data and 0.713 (95% CI, 0.677, 0.748) with both combined. When using data combined from both month 6 and baseline, CSME at month 24 was predicted with an AUROC = 0.508 (95% CI, 0.471, 0.545) with systemic data, 0.749 (95% CI, 0.719, 0.778) with imaging features and 0.738 (95% CI, 0.707, 0.770) with both combined.
Conlusions
Our results indicate that development of CSME in patients with mild NPDR can be more accurately predicted with retinal imaging features than with systemic data alone. When predicting the development of CSME by month 24 using data obtained at month 6 only, the performance was not significantly different compared with using both baseline and month 6. Extending the evaluation to larger datasets is planned to enable more confident conclusions regarding the performance and predictive potential of the different types of data. Such predictive models of CSME in patients with NPDR could help inform personalised monitoring and follow-up, in both clinical development and clinical practice settings.
Financial Disclosure
Dimitrios Damopoulos: Employee of Hays (Schweiz) AG, Basel, Switzerland, hired by F. Hoffmann-La Roche Ltd., Basel, Switzerland as a contractor. Prof. José Cunha-Vaz: Consultant for Carl Zeiss Meditec, Alimera Sciences, Allergan, Bayer, Gene Signal, Novartis, Pfizer, Oxular, Roche and Sanofi Ales Neubert, Beatriz Garcia Armendariz, Robert Weikert and Fethallah Benmansour: Employees of F. Hoffmann-La Roche Ltd. Daniela Ferrara: Employee of Genentech, Inc. Luis Mendes and Torcato Santos have no relevant financial disclosures to report.
Comments
Please add below extra two co-authors: Co-author 7 first name: José Co-author 7 last name: Cunha-Vaz Co-author 7 affiliation: AIBILI- Association for Innovation and Biomedical Research on Light and Image, Coimbra, Portugal Co-author 8 first name: Fethallah Co-author 8 last name: Benmansour Co-author 8 affiliation: F. Hoffmann-La Roche Ltd, Basel, Switzerland
