Risk Assessment Models for Late AMD When patients are told by their ophthalmologist that they have early signs of age-related macular degeneration (AMD), they often ask about the risk of their AMD progressing. Long-term epidemiological studies that have collected information on the prevalence, incidence, and natural history of AMD provide data to answer this question. These studies have shown that risk is influenced by many things, including the severity of the signs of AMD that are already present (eg, drusen size, area of involvement, and type and presence of pigmentary abnormalities), demographic factors (eg, age, sex, and race/ethnicity), family history of the disease, lifestyle factors (eg, smoking, heavy drinking, and physical activity), medical conditions (eg, inflammatory and pulmonary disease), environmental exposures (eg, light), and ocular factors (eg, history of cataract surgery). Risk assessment models account for multiple exposures that may influence risk. These have been useful for heart disease and have recently been developed for AMD. These models may also help to decide on the interval of follow-up and whether to suggest options for preventive care (eg, dietary supplements and specific foods).

The new risk assessment model proposed by Klein et al is based on findings from the Age-Related Eye Disease Study (AREDS) for the development of advanced AMD. It was designed for “the identification of those individuals with early AMD who are at greatest risk to progress to advanced, vision-threatening AMD (geographic atrophy or neovascular AMD) ... and to predict when progression to advanced AMD might occur.” The model uses the modified simple scale score of AMD severity from the AREDS including indicator variables for the presence of very large drusen and the presence of late AMD in 1 eye, age, smoking history, history of AMD in a first-degree relative, and the presence or absence of 2 genetic variants (CFH Y402H and ARMS2 A69S) for the individual patient. It was restricted to white patients. The area under the receiver operating characteristic curve showed good discrimination between those who would progress to late AMD and those who would not. Two previous models also showed similar overall performance, although the end points differed between studies.

Of particular interest in the article by Klein and colleagues is that the best model was the one that included 3 groups of risk factors: demographic/environmental (age, smoking), phenotypic (severity of early AMD based on the simple AREDS severity scale), and genetic (ARMS2 and CFH Y402H). This model was just slightly better than the model including only the first 2 of these groups; genetic factors added only 0.007 to the C statistic, the estimate of the improvement of the model. Figure 1 in the article by Klein and colleagues graphically depicts the very small contribution of the presence of “risk” alleles to decreasing prediction error in the model. The difference between models including phenotypic and demographic characteristics and including only demographic and genetic factors may reflect the fact that phenotype includes effects of the several other genes influencing lesions of AMD, gene X environment interactions, and other possible environmental effects such as diet and supplement deficiencies that may have influenced the phenotype before study entry. Knowing the severity of the lesions of AMD that are already present coupled with knowledge of the important lifestyle factors (eg, smoking history) gives most of the important information about risk of progression. It is possible that the model would perform differently in other populations.

A question arising from the fact that the risk assessment model was derived from data from participants in a trial is whether the model is applicable to persons in a general population who may be more like an average patient in a clinic setting. Having replicated the validity of the model in participants from another trial does not assure the usefulness of this model in clinical practice. For this reason, we examined the performance of the proposed risk formula (kindly provided by Klein and colleagues) in a group of 1575 persons aged 55 to 80 years at baseline in the population-based Beaver Dam Eye Study (BDES). These persons were white, participated in a 5-year follow-up examination, and were at risk for developing late AMD in at least 1 eye. In all categories, the model predicted greater risk of progression to late AMD than actually occurred in the BDES sample. For example, in the BDES subgroup predicted to have a 20% to 30% 5-year risk of progression to late AMD, the actual 5-year incidence of late AMD was 4%. The model should be tested in other population-based cohort studies.

Klein and colleagues did not include information on supplement intake, a risk factor that the AREDS investigators are uniquely able to examine. They inform the reader that one could expect a 10% to 15% reduction or increase in calculated risk depending on the treatment assignment. Why not consider this a person-specific risk factor in the risk assessment analysis?

The importance of the genetic factors as a component of the model in assessment tools in which the phenotype has been characterized invites further consideration as to the usefulness of such factors for predicting risk in clinical practice.

Read more...
Ophthalmology. 2011 Feb;118(2):332-8