Dataset Information

Prediction of Post-Treatment Visual Acuity in Age-Related Macular Degeneration Patients With an Interpretable Machine Learning Method.

ABSTRACT:

Purpose

We evaluated the features predicting visual acuity (VA) after one year in neovascular age-related macular degeneration (nAMD) patients.

Methods

A total of 527 eyes of 506 patients were included. Machine learning (ML) models were trained to predict VA deterioration beyond a logarithm of the minimum angle of resolution of 1.0 after 1 year based on the sequential addition of multimodal data. BaseM models used clinical data (age, sex, treatment regimen, and VA), SegM models included fluid volumes from optical coherence tomography (OCT) images, and RawM models used probabilities of visual deterioration (hereafter probability) from deep learning classifiers trained on baseline OCT (OCT0) and OCT after three loading doses (OCT3), fluorescein angiography, and indocyanine green angiography. We applied SHapley Additive exPlanations (SHAP) for machine learning model interpretation.

Results

The RawM model based on the probability of OCT0 outperformed the SegM model (area under the receiver operating characteristic curve of 0.95 vs. 0.91). Adding probabilities from OCT3, fluorescein angiography, and indocyanine green angiography to RawM showed minimal performance improvement, highlighting the practicality of using raw OCT0 data for predicting visual outcomes. Applied SHapley Additive exPlanations analysis identified VA after 3 months and OCT3 probability values as the most influential features over quantified fluid segments.

Conclusions

Integrating multimodal data to create a visual predictive model yielded accurate, interpretable predictions. This approach allowed the identification of crucial factors for predicting VA in patients with nAMD.

Translational relevance

Interpreting a predictive model for 1-year VA in patients with nAMD from multimodal data allowed us to identify crucial factors for predicting VA.

SUBMITTER: Kim N

PROVIDER: S-EPMC11373725 | biostudies-literature | 2024 Sep

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Prediction of Post-Treatment Visual Acuity in Age-Related Macular Degeneration Patients With an Interpretable Machine Learning Method.

Kim Najung N Lee Minsub M Chung Hyewon H Kim Hyung Chan HC Lee Hyungwoo H

Translational vision science & technology 20240901 9

<h4>Purpose</h4>We evaluated the features predicting visual acuity (VA) after one year in neovascular age-related macular degeneration (nAMD) patients.<h4>Methods</h4>A total of 527 eyes of 506 patients were included. Machine learning (ML) models were trained to predict VA deterioration beyond a logarithm of the minimum angle of resolution of 1.0 after 1 year based on the sequential addition of multimodal data. BaseM models used clinical data (age, sex, treatment regimen, and VA), SegM models in ...[more]

PMID: 39226064

Similar Datasets

Project description:ObjectiveTo describe the effects of treatment for 1 year with ranibizumab or bevacizumab on macular morphology and the association of macular morphology with visual acuity (VA) in eyes with neovascular age-related macular degeneration (AMD).DesignProspective cohort study within a randomized clinical trial.ParticipantsParticipants in the Comparison of Age-related Macular Degeneration Treatments Trials.MethodsParticipants were assigned randomly to treatment with ranibizumab or bevacizumab on a monthly or as-needed schedule. Optical coherence tomography (OCT), fluorescein angiography (FA), color fundus photography (FP), and VA testing were performed periodically throughout 52 weeks. Masked readers graded images. General linear models were applied to evaluate effects of time and treatment on outcomes.Main outcome measuresFluid type and location and thickness by OCT, size, and lesion composition on FP, FA, and VA.ResultsIntraretinal fluid (IRF), subretinal fluid (SRF), subretinal pigment epithelium fluid, and retinal, subretinal, and subretinal tissue complex thickness decreased in all treatment groups. A higher proportion of eyes treated monthly with ranibizumab had fluid resolution at 4 weeks, and the difference persisted through 52 weeks. At 52 weeks, there was little association between the presence of fluid of any type (without regard to fluid location) and the mean VA. However, at all time points, eyes with residual IRF, especially foveal IRF, had worse mean VA (9 letters) than those without IRF. Eyes with abnormally thin (<120 μm) or thick (>212 μm) retinas had worse VA than those with normal thickness (120-212 μm). At week 52, eyes with larger neovascular lesions or with foveal scar had worse VA than eyes without these features.ConclusionsAnti-vascular endothelial growth factor (VEGF) therapy reduced lesion activity and improved VA in all treatment groups. At all time points, eyes with residual IRF had worse VA than those without. Eyes with abnormally thin or thick retinas, residual large lesions, and scar also had worse VA. Monthly ranibizumab dosing yielded more eyes with no fluid and an abnormally thin retina, although the long-term significance is unknown. These results have important treatment implications in eyes undergoing anti-VEGF therapy for neovascular AMD.Financial disclosure(s)Proprietary or commercial disclosure may be found after the references.

Project description:ImportanceAlthough anti-vascular endothelial growth factor treatment of neovascular age-related macular degeneration (AMD) results in improved vision overall, loss of substantial vision can occur. Understanding the processes that lead to loss of vision may lead to preventive strategies.ObjectiveTo determine the incidence, characteristics, causes, and baseline predictors of sustained visual acuity loss after 2 years of treatment with ranibizumab or bevacizumab for neovascular AMD.Design, setting, and participantsA cohort study within a randomized clinical trial of participants in the Comparison of Age-Related Macular Degeneration Treatments Trials (CATT).InterventionsParticipants were randomly assigned to treatment with ranibizumab or bevacizumab and to 2 years of monthly or as needed injections or monthly injections for 1 year and as needed injections the following year.Main outcomes and measuresSustained visual acuity loss, defined as loss of 15 or more letters from baseline at weeks 88 and 104.ResultsAmong 1030 participants, 61 eyes (5.9%) developed sustained visual acuity loss in 2 years. Within this group, visual acuity decreased gradually over time, with a mean decrease of 2, 19, and 33 letters from baseline at 4 weeks, 1 year, and 2 years, respectively. At 2 years, eyes with sustained visual acuity loss had more scarring (60.0% vs 41.4%, P = .007), more geographic atrophy (GA) (31.6% vs 20.7%, P = .004), larger lesions (16 vs 8 mm2, P < .001), and higher proportions of intraretinal fluid (82.5% vs 51.0%, P < .001), subretinal hyperreflective material (84.5% vs 44.2%, P < .001), retinal thinning (43.3% vs 23.0%, P < .001), and thickening (20.0% vs 12.1%, P < .001). Likely causes of sustained visual acuity loss included foveal scarring (44.3%), pigmentary abnormalities (27.9%), and foveal GA (11.5%). Baseline factors independently associated with a higher incidence of sustained visual acuity loss were the presence of nonfoveal GA (odds ratio [OR], 2.86; 95% CI, 1.35-6.08; P = .006), larger area of choroidal neovascularization (OR for a >4-disc area vs ≤1-disc area, 3.91; 95% CI, 1.70-9.03; P = .007), and bevacizumab treatment (OR, 1.83; 95% CI, 1.07-3.14; P = .03).Conclusions and relevanceSustained visual acuity loss was relatively rare in CATT. The development of foveal scar, pigmentary abnormalities, or GA contributed to most of the sustained visual acuity loss. Risk was 3% higher among eyes treated with bevacizumab. Treatment that targeted the prevention of scarring or GA may improve vision outcomes.Trial registrationclinicaltrials.gov Identifier: NCT00593450.

Project description:PurposeTo evaluate associations of morphologic features with 5-year visual acuity (VA) in the Comparison of Age-related Macular Degeneration Treatments Trials (CATT).DesignCohort study within a randomized clinical trial.ParticipantsParticipants in CATT.MethodsEyes with age-related macular degeneration-associated choroidal neovascularization (CNV) and VA between 20/25 and 20/320 were eligible. Treatment was assigned randomly to ranibizumab or bevacizumab and to 3 dosing regimens for 2 years and was at the ophthalmologists' discretion thereafter.Main outcome measuresVisual acuity, thickness and morphologic features on OCT, and lesion size and foveal composition on fundus photography (FP) and fluorescein angiography (FA).ResultsVisual acuity and image gradings were available for 523 of 914 participants (57%) alive at 5 years. At 5 years, 60% of eyes had intraretinal fluid (IRF), 38% had subretinal fluid (SRF), 36% had subretinal pigment epithelium (RPE) fluid, and 66% had subretinal hyper-reflective material (SHRM). Mean (standard deviation) foveal center thickness was 148 μm (99) for retina, 5 μm (21) for SRF, 125 μm (107) for subretinal tissue complex, 11 μm (33) for SHRM, and 103 μm (95) for RPE + RPE elevation. The SHRM, thinner retina, greater CNV lesion area, and foveal center pathology (all P < 0.001) and IRF (P < 0.05) were independently associated with worse VA. Adjusted mean VA letters were 62 for no pathology in the foveal center; 61 for CNV, fluid, or hemorrhage; 65 for non-geographic atrophy (GA); 64 for nonfibrotic scar; 53 for GA; and 56 for fibrotic scar. Incidence or worsening of 8 pathologic features (foveal GA, foveal scar, foveal CNV, SHRM, foveal IRF, retinal thinning, CNV lesion area, and GA area) between years 2 and 5 was independently associated with greater loss of VA from years 2 to 5 and VA loss from baseline to year 5.ConclusionsAssociations between VA and morphologic features previously identified through year 1 were maintained or strengthened at year 5. New foveal scar, CNV, intraretinal fluid, SHRM and retinal thinning, development or worsening of foveal GA, and increased lesion size are important contributors to the VA decline from years 2 to 5. A significant need to develop therapies to address these adverse pathologic features remains.

Project description:PurposeTo describe the association between morphologic features on fundus photography (FP), fluorescein angiography (FA), and optical coherence tomography (OCT) and visual acuity (VA) in the second year of the Comparison of Age-related Macular Degeneration Treatments Trials (CATT).DesignProspective cohort study within a randomized clinical trial.ParticipantsParticipants in the CATT.MethodsStudy eye eligibility required angiographic and OCT evidence of choroidal neovascularization (CNV) secondary to age-related macular degeneration (AMD) and VA between 20/25 and 20/320. Treatment was assigned randomly to ranibizumab or bevacizumab with 3 different dosing regimens over a 2-year period.Main outcome measuresFluid type, location, and thickness; retina and subretinal tissue complex thickness on OCT; size and lesion composition on FP and FA; and VA.ResultsAmong 1185 CATT participants, 993 (84%) had fluid on OCT at baseline and completed 2 years of follow-up. At 2 years, intraretinal fluid (IRF), subretinal fluid (SRF), sub-retinal pigment epithelium (RPE) fluid, and subretinal tissue complex thickness decreased in all treatment groups. Ranibizumab monthly was best able to resolve each type of fluid. Eyes with SRF in the foveal center on OCT had better mean VA than eyes with no SRF (72.8 vs. 66.6 letters; P = 0.006). Eyes with IRF in the foveal center had worse mean VA than eyes without IRF (59.9 vs. 70.9 letters; P < 0.0001). Eyes with retinal thickness <120 μm had worse VA compared with eyes with retinal thickness 120 to 212 and >212 μm (59.4 vs. 71.3 vs. 70.3 letters; P < 0.0001). At 2 years, the mean VA (letters) of eyes varied substantially by the type of subfoveal pathology on FP and FA: 70.6 for no pathology; 74.1 for fluid only; 73.3 for CNV or pigment epithelial (RPE) detachment; 68.4 for nongeographic atrophy; and 62.9 for geographic atrophy, hemorrhage, RPE tear, or scar (P < 0.0001).ConclusionsThe associations between VA and morphologic features identified through year 1 were maintained or strengthened during year 2. Eyes with foveal IRF, abnormally thin retina, greater thickness of the subretinal tissue complex on OCT, and subfoveal geographic atrophy or scar on FP/FA had the worst VA. Subretinal fluid was associated with better VA.

Dataset Information

Prediction of Post-Treatment Visual Acuity in Age-Related Macular Degeneration Patients With an Interpretable Machine Learning Method.

Purpose

Methods

Results

Conclusions

Translational relevance

Publications

Prediction of Post-Treatment Visual Acuity in Age-Related Macular Degeneration Patients With an Interpretable Machine Learning Method.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets