People in the 53 practices allocated to the "universal" arm of the trial were given a visual acuity test as part of a detailed health assessment by the practice nurse. Visual acuity was measured at 3 metres with a Glasgow Acuity Chart which measures the minimal angle of resolution on a logarithmic scale[6]. Vision was measured both as presenting vision (with spectacle or contact lenses) and also in each eye. People with binocular presenting vision better than 6/9 were defined as "good vision", those with presenting vision less than 6/9 to 6/18 as reduced vision. Visual impairment was defined as presenting binocular acuity of less than 6/18 (logMAR score 0.5 or more). In 49 practices, the cause of visual impairment was assessed by medical record review[7]. Information on hospital admission was collected by trial nurses on an ongoing basis for each patient from the date of the baseline assessment. The baseline assessments were conducted during four years 1995 to 1998; hospital admissions were collected by study nurses for each patient for a 2-year period from baseline (invitation to assessment) from hospital discharge letters in the patients' GP records. The definition of a hospital admission included an overnight stay. The trial and additional data collection on the cause of visual loss was approved by the relevant local research ethics committees.

All analyses were done using Stata version 10.0 and took into account the cluster design of the study using "svy" commands (Stata Corporation, College Station, Texas 77845, USA). People with an MMSE score of less than 12 were excluded from these analyses because measurement of visual acuity in this group is likely to be inaccurate. The rate ratio of hospital admission associated with visual impairment was estimated using poisson regression. In all analyses the reference group was those with "good vision". The following potential confounding factors and effect modifiers were considered: age, sex, marital status (single/married/widowed), living alone, housing tenure (home owner/not home owner/sheltered accommodation), financial difficulties (difficulties making ends meet and/or managing finances), looked after someone with a serious illness in the last year, death of loved one in last year, social support (no relative/friend to call on and/or no help at night), alcohol consumption (never/ex/current below median/current above median), smoking (never/ex/current), body mass index (BMI) (quintiles), depression (6 or more on Geriatric Depression Scale), diabetes, hearing impairment (failed whispered voice test), reported major illness (heart attack, stroke, Parkinsons disease, cancer), self-reported health (excellent or very good/good/fair or poor), activities of daily living (ADL) score (unable to do 0–1 ADLs, unable to do 2–4 ADLs, unable to do 5–8 ADLs), falls in last six months (none/one/two or more), self-reported activity (very or fairly/not very or not at all), cognitive impairment (Mini Mental State Examination (MMSE) score 12-17/18-23/24-30).

Table ​Table22 shows the rate of hospital admission in visually impaired people. The rate of hospital admission in visually impaired people was 238.7/1000 person-years compared to 169.7/1000 person-years in people with good vision. The rate ratio adjusted for age and sex was 1.32 (95% CI 1.19 to 1.47). Adjusting for a wide range of potential explanatory social and co-morbidity factors eliminated this increased risk of hospital admission in visually impaired people. The rate ratio adjusting for all factors was 1.06 (95% CI 0.94 to 1.20). We repeated the analyses excluding variables potentially on the causal pathway. The rate ratio for hospital admission was 1.19 (95% CI 1.06 to 1.34).

Table ​Table33 shows the rate of hospital admission by cause of visual impairment. After adjustment for confounders there were no significant associations between cause of vision impairment and risk of admission.

The reason for hospital admission was specified for 87% of admissions, i.e. the speciality of the admission was recorded. 1.5% of admissions were classified as "accident and emergency". However, this figure probably represents admissions to small short stay wards found in some (but not all) casualty departments and is unlikely to reflect true visits to emergency departments. There was little association between visual impairment and risk of an "accident and emergency" admission (risk ratio controlling for all explanatory factors 0.73, 95% CI 0.29 to 1.84).

There were 303 admissions recorded as "ophthalmology". Unsurprisingly people with visual impairment were more likely to be admitted for one or more ophthalmology admissions (risk ratio controlling for all explanatory factors 3.10, 95% CI 2.24 to 4.29). Excluding these ophthalmology admissions did not change the results: the rate ratio adjusted for age and sex was 1.31 (95% CI 1.17 to 1.46); the rate ratio adjusting for all factors was 1.04 (95% CI 0.92 to 1.18); and the rate ratio adjusted for variables not potentially on the causal pathway was 1.17 (95% CI 1.04 to 1.32).

There are not many other studies on this topic. Jack et al measured the prevalence of visual impairment in 200 people aged 65 years and above admitted to one UK hospital with an acute medical illness[1]. They found that half of these people were visually impaired. Although this is a high prevalence of visual impairment it is difficult to interpret as the age structure of the hospital admissions is unlikely to reflect population age structure and the prevalence of visual impairment increases rapidly over the ages of 65 years and above[8,9]. Jacobs et al studied 839 people aged 77 in Israel and found that visual impairment was associated with an increased likelihood of attending for emergency care in men but not in women[2]. However, these analyses were not controlled for age or any other potentially confounding factors. Morse et al in New York State, USA, analysed routinely collected data on length of hospital stay[3]. Visual impairment was associated with an extra two days length of stay in hospital, after controlling for age, sex, payer source and other co-morbid conditions. We similarly found a non-significant increased length of stay in visually impaired older people of two days, after adjusting for explanatory factors not on the causal pathway.