To develop comprehensive predictive models for choroidal neovascularization (CNV) and geographic atrophy (GA) incidence within 3 years that can be applied realistically to clinical practice.
Retrospective evaluation of data from a longitudinal study to develop and validate predictive models of CNV and GA.
The predictive performance of clinical, environmental, demographic, and genetic risk factors was explored in regression models, using data from both eyes of 2011 subjects from the Age-Related Eye Disease Study (AREDS). The performance of predictive models was compared using 10-fold cross-validated receiver operating characteristic curves in the training data, followed by comparisons in an independent validation dataset (1410 AREDS subjects). Bayesian trial simulations were used to compare the usefulness of predictive models to screen patients for inclusion in prevention clinical trials.
Logistic regression models that included clinical, demographic, and environmental factors had better predictive performance for 3-year CNV and GA incidence (area under the receiver operating characteristic curve of 0.87 and 0.89, respectively), compared with simple clinical criteria (AREDS simplified severity scale). Although genetic markers were associated significantly with 3-year CNV ( CFH : Y402H; ARMS2 : A69S) and GA incidence ( CFH : Y402H), the inclusion of genetic factors in the models provided only marginal improvements in predictive performance.
The logistic regression models combine good predictive performance with greater flexibility to optimize clinical trial design compared with simple clinical models (AREDS simplified severity scale). The benefit of including genetic factors to screen patients for recruitment to CNV prevention studies is marginal and is dependent on individual clinical trial economics.
Age-related macular degeneration (AMD) is the leading cause of visual impairment in the Western world. The development of anti–vascular endothelial growth factor therapies has led to a significant protection from further vision loss for many choroidal neovascularization (CNV) patients. Although there are potential geographic atrophy (GA) therapies in clinical trials, there is currently no effective treatment for GA, which represents approximately 38% of advanced AMD cases in white populations and 5% to 30% of advanced AMD cases in East Asian populations. Apart from the use of nutritional supplements (antioxidants plus zinc), there are currently no prophylactic treatments recommended for preventing the advance to CNV or GA. Because the onset of CNV and GA is accompanied by progressive vision loss, major unmet needs are the ability to predict which patients are at greatest risk of developing CNV or GA and the availability of therapeutics to prevent progression to CNV and GA in those patients.
Risk factors for AMD include genetic factors, advanced age, smoking, and body mass index. Many of these risk factors have been assessed for their contribution to predicting progression to any advanced AMD (i.e., CNV and GA). The Age-Related Eye Disease Study (AREDS) simplified severity scale currently is the most practical model of advanced AMD incidence to apply in the clinic, requiring the assessment of clinical features only, for each eye. Other models use a combination of clinical, demographic, environmental, and genetic factors to predict progression to any advanced AMD. However, we are unaware of previous reports of predictive models that both offer prediction of the patients at highest risk of conversion within the time frames most relevant to clinical practice (≤ 3 years) and that are designed to differentiate progression to CNV or GA, as opposed to any advanced AMD. Because therapeutic interventions for CNV and GA may differ, development of predictive models specific for CNV or GA thus is warranted. To inform design of clinical trials for therapies that may reduce progression to advanced AMD, this study developed and compared models for predicting conversion to CNV or GA within 3 years of evaluation. Our models of CNV and GA incidence included clinical, environmental, demographic, and genetic risk factors. We addressed whether there is additional predictive value in the inclusion of genetic markers in the models, and if so, whether there is sufficient additional predictive value to justify genetic assessment for patient enrollment in clinical trials.
Study Cohort and Dataset
This report uses data from the AREDS, a multicenter, randomized clinical trial that evaluated the effect of antioxidants and zinc on the progression of AMD and cataract. The AREDS included baseline and annual visit assessments for 4757 subjects over a median follow-up period of 10.6 years. We used data from 2011 subjects for predictive model development and data from 1404 subjects for validation. The clinical dataset used for the analyses described in this article were obtained from the AREDS database at the database of Genotypes and Phenotypes (dbGaP). Each eye was categorized based on longitudinal records of ocular examination and fundus photography conducted during annual visits. The AMD grading schemes used for our analysis were the AREDS severity scale and the AREDS simplified severity scale. Subjects in the model development set (n = 2011) include all white persons who provided general research use consent, had a quality assessed DNA sample available for genetic evaluation, and graded fundus photograph data available for the 2- and 3-year visits. Clinical, demographic, environmental, and genetic data were evaluated for the study population, which included at baseline, subjects with no or small drusen, as well as subjects with early AMD and unilateral advanced AMD ( Table 1 ). The validation dataset (n = 1404) comprised the white subjects who were not used for model generation who provided general research use consent and had graded fundus photograph data available for the 2- and 3-year visits, but did not have a genetic sample. Baseline variables for the validation population are provided in the Supplemental Materials ( Supplemental Table 5 , available at AJO.com ).
|Total No. of Subjects at Baseline||3-Year Incident Choroidal Neovascularization a||3-Year Incident Geographic Atrophy b|
|% (n) or Mean (SD)||OR (95% CI)||P Value||% (n)||OR (95% CI)||P Value|
|Total||2011||5.7% (114/2011)||2.9% (59/2011)|
|No CNV or GA||1728||3.0% (51/1728)||0.1 (0.1 to 0.2) c||9.4 × 10 −29||2.1% (36/1728)||0.2 (0.1 to 0.4) c||2.2 × 10 −7|
|Unilateral CNV||253||23.3% (59/253)||9.4 (6.3 to 14.0) d||1.4 × 10 −28||7.1% (18/253)||3.2 (1.8 to 5.7) d||6.3 × 10 −5|
|Unilateral GA||30||10.0% (3/30)||1.9 (0.6 to 6.3) e||.3||16.7% (5/30) c||7.1 (2.6 to 19.4) e||.0001|
|AMD grade in worse eye|
|1||503||0% (0/503)||5.5 (4.0 to 7.5) f||1.4 × 10 −26||0% (0/503)||3.5 (2.4 to 5.0) f||9.3 × 10 −12|
|2||479||0.4% (2/479)||0% (0/479)|
|3||750||6.7% (50/750)||4.9% (37/750)|
|4||279||22.2% (62/279)||7.9% (22/279)|
|AREDS simplified severity scale|
|0||788||0.1% (1/788)||2.8 (2.3 to 3.4) f||1.2 × 10 −27||0% (0/788)||4.0 (2.8 to 5.8) f||1.1 × 10 −14|
|1||360||1.7% (6/360)||0% (0/360)|
|2||315||3.5% (11/315)||1.0% (3/315)|
|3||239||13.0% (31/239)||6.3% (15/239)|
|4||304||20.7% (63/304)||13.5% (41/304)|
|Missing Data||5||40.0% (2/5)||0% (0/5)|
|Female||1151||5.9% (68/1151)||Reference||1.9% (22/1151)||Reference|
|Male||860||5.3% (46/860)||0.9 (0.6 to 1.3)||.6||4.3% (37/680)||2.3 (1.3 to 3.9)||.002|
|Never||976||4.2% (41/976)||Reference||2.0% (20/976)||Reference|
|Former||922||6.5% (60/922)||1.6 (1.1 to 2.4)||.03||3.3% (30/922)||1.6 (0.9 to 2.9)||.1|
|Current||113||11.5% (13/113)||3.0 (1.5 to 5.7)||.001||8.0% (9/113)||4.1 (1.8 to 9.3)||.0006|
|≤High school||645||7.8% (50/645)||Reference||4.3% (28/645)||Reference|
|≥College||1366||4.7% (64/1366)||0.6 (0.4 to 0.9)||.006||2.3% (31/1366)||0.5 (0.3 to 0.9)||.01|
|Placebo||596||5.2% (31/596)||Reference||2.7% (16/596)||Reference|
|Antioxidants||624||5.4% (34/624)||1.1 (0.6 to 1.7)||.8||2.6% (16/624)||1.0 (0.5 to 1.9)||.9|
|Zinc||396||5.6% (22/396)||1.1 (0.6 to 1.9)||.8||3.3% (13/396)||1.2 (0.6 to 2.6)||.6|
|Antioxidants + zinc||395||6.8% (27/395)||1.3 (0.8 to 2.3)||.3||3.5% (14/395)||1.3 (0.6 to 2.8)||.4|
|No||1304||5.6% (73/1304)||Reference||2.5% (33/1304)||Reference|
|Yes||707||5.8% (41/707)||1.0 (0.7 to 1.5)||.9||3.7% (26/707)||1.5 (0.9 to 2.5)||.1|
|No||1829||5.5% (101/1829)||Reference||2.7% (49/1829)||Reference|
|Yes||182||7.1% (13/182)||1.3 (0.7 to 2.4)||.4||5.5% (10/182)||2.1 (1.0 to 4.3)||.04|
|Age (years)||68.7 (4.9)||71.3 (5.0)||1.1 (1.1 to 1.2)||1.4 × 10 −8||70.4 (5.5)||1.1 (1.0 to 1.1)||.006|
|BMI (kg/m 2 )||27.5 (4.7)||28.1 (4.0)||1.0 (1.0 to 1.1)||.1||28.3 (4.2)||1.0 (1.0 to 1.1)||.2|