Sir,

I read with interest the article by Jain et al1 titled ‘Screening for age-related macular degeneration using nonstereo digital photographs’. The authors found reasonably high sensitivities and specificities for detection of ARM and age-related macular degeneration (AMD) by graders viewing digital photos. While their results are not in question, I do raise objection to their discussion in which they state that the findings of the study might usefully be extended into a primary care setting. This assumption ignores a pivotal statistical fact and highlights why sensitivities and specificities alone do not tell the whole story when assessing how useful a screening test is.2

The setting of the study involved a contrived selection of cases from a retinal unit database. In this ‘population’, the prevalence of neovascular AMD was 31%. In a primary care setting, of course, the real prevalence will be much lower, say 2% in patients over 65 years. While this difference does not affect the sensitivity or specificity of the screening tool, it does impact very significantly on the positive predictive value.

In the study, for example, for grader 1, the sensitivity was 82.1% and the specificity was 79.7%. The positive predictive value can be calculated as 64.8% in the study ‘population’. If the same sensitivity and specificity are applied to a primary care population, with an AMD prevalence of say 2%, the positive predictive value drops to 7.8%. This means that over 92% of positive results will actually be false positives.

This demonstrates that the utility of a screening tool cannot be evaluated without reference to the prevalence of the disease in the population in which it is to be used.