Estimating undiagnosed dementia in England using capture recapture techniques.

Background: To our knowledge capture-recapture techniques have not been used to estimate dementia prevalence using routinely collected data in England, nor have they been used to estimate changes in undiagnosed dementia over time. In this study we aimed to use routinely collected electronic health records to estimate the number of undiagnosed dementia cases there are in England and how this has changed over time. We also aimed to assess whether proportion of undiagnosed cases differed by age group, ethnicity, socioeconomic deprivation and sex.

Methods: We used routinely collected primary care data linked to hospital episode statistics from 1997 to 2018. We tabulated capture of dementia in each of the two datasets and used the Lincoln-Petersen estimator to estimate numbers of missing dementia diagnoses per year along with the estimated total number of cases and the proportion of cases identified. We calculated age and sex-adjusted prevalence of dementia for each year and used proportion of cases identified to estimate the underlying population prevalence of dementia per year. We conducted beta regression to estimate how sex, age band, deprivation and ethnic group affects the proportion of dementia cases identified, adjusting for year.

Results: Proportion of cases out of the estimated total that were identified, rose from 42.4% in 1997 to 84.4% in 2018. Estimated population prevalence of dementia rose from 1997 to a high of 4.4% in 2018 in those aged ≥ 65. Proportion of dementia cases identified did not vary by sex but a lower proportion of those from the South Asian ethnic group were diagnosed compared to the White population (coeff -0.115, 95% CI -0.218 to -0.011). Compared to those aged 65-74, those aged 75-84 and 85 + had higher proportions of dementia diagnosed (75-84 Coeff 0.259, 95% CI 0.153-0.366; 85 + Coeff 0.185, 95% CI 0.079-0.291). Those living in the two most deprived areas had a higher proportion of dementia diagnosed compared to the least deprived area (IMD quintile 4 vs 1 coeff 0.093, 95% CI 0.014 to 0.173, IMD quintile 5 vs 1 coeff 0.162, 95% CI 0.083 to 0.242).

Conclusions: Proportion of dementia cases identified has increased over time and results indicate that underlying prevalence of dementia may be lower than previously estimated but this needs replication. Greater focus needs to be given to those with dementia onset at younger ages and those from South Asian backgrounds as dementia is relatively under-diagnosed in these groups.