The data was arranged following an orderly review of published population-focused studies on vision impairment and blindness by the VLEG. This review included studies published between Jan 1, 1980, and Oct 1, 2018, incorporating grey literature as well. Relevant studies from this review were combined with data from Rapid Assessment of Avoidable Blindness (RAAB) studies by VLEG. Data was also sourced from the US National Health and Nutrition Examination survey and the WHO Study on Global Ageing and Adult Health, contributed by the GBD team. More detailed methods are published elsewhere10,3 and discussed in brief as follows.
In total, VLEG pinpointed 137 studies and pulled data from 70 studies in their 2010 review, and 67 additional studies in their 2014–18 review. Most of these studies were national or subnational cross-sectional surveys. VLEG also arranged for the production of 5-year age-segregated RAAB data from the RAAB repository. To qualify, studies had to meet specific criteria: vision acuity data must be gathered through a test chart compatible with the Snellen scale, and the sample had to be representative of the population. Subjective reports of vision loss were not included. The criteria for vision loss was defined by the International Classification of Diseases 11th edition as employed by WHO. It was based on the vision in the better eye upon presentation. Moderate vision loss was defined as a visual acuity of 6/60 or better but less than 6/18, severe vision loss as a visual acuity of 3/60 or better but less than 6/60, and blindness as a visual acuity of less than 3/60 or less than 10° visual field around central fixation (although the visual field definition was rarely used in population-based eye surveys).
We split the original data into several datasets, creating separate envelopes for each degree of vision loss (mild, moderate, and severe) and blindness. This data was then fed into a meta-regression tool designed by the Institute for Health Metrics and Evaluation (IHME) known as MR-BRT (meta regression; Bayesian; regularised; trimmed). The benchmark for each severity level was presenting vision impairment.
Data about uncorrected refractive errors were pulled straight from the data sources when possible, and if not, they were calculated by subtracting the best-corrected vision impairment from presenting vision impairment prevalence at each level of severity. Other causes were factored into the best-corrected estimates for each level of vision impairment.
Our models for distance vision impairment and blindness were based on the most commonly reported causes found in the literature, and the minimum age for inclusion of data on AMD was 45 years. We created estimates of MSVI and blindness that were specific to location, year, age, and sex using Disease Modelling Meta-Regression (Dismod-MR) 2.112. Its data processing steps have been outlined elsewhere3. Briefly, Dismod-MR 2.1 models were run for all vision impairment by severity (moderate, severe, blindness) regardless of cause and, separately, for MSVI and blindness due to each modelled cause of vision impairment. Then, models of MSVI due to specific causes were split into moderate and severe estimates using the ratio of overall prevalence in the all-cause moderate presenting vision impairment and severe presenting vision impairment models. Next, prevalence estimates for all causes by severity were scaled to the models of all-cause prevalence by severity. This produced final estimates by age, sex, year, and location for each individual cause of vision impairment by severity. We age-standardised our estimates using the GBD standard population13. Data on blindness and MSVI due to AMD were presented by seven super-regions (Southeast Asia/East Asia/Oceania, Central Europe/Eastern Europe/Central Asia, High-income, Latin America and Caribbean, North Africa and Middle East, South Asia, and Sub-Saharan Africa and globally. Data on other causes of vision impairment and blindness will be presented in separate publications.