About

About the Annual Deprivation Index

The ADI is a multi-domain measure of deprivation in England, published annually for 2014 to 2025 at four geographic levels. It is a product of the Autonomy Data Unit at the Autonomy Institute, with contributions from Rob Calvert Jump.

The three domains

Geography & coverage

Outputs are produced at four levels: LSOA (~33,750 neighbourhoods), local authority district (296), region (9) and England. Domains are processed in 2011 LSOA boundaries and converted to 2021 boundaries with a crosswalk weighted from each publication year's population, then rolled up. The index is not combined into a single composite score; the three domains are kept separate.

pop is the all-age ONS population summed over neighbourhoods included in the release. Each metric also carries the population that actually belongs in its denominator, and its rate is count divided by that metric-specific population. For most metrics that is the covered all-age population; the new QOF-comparable health rates use the covered population in the condition's eligible age range instead.

Estimating health at neighbourhood level

QOF disease registers are published per GP practice, not per neighbourhood. For each disease we use practices with a published register and usable list size, weight them by the number of that neighbourhood's patients they cover, and withhold a source estimate when those practices cover less than 80% of its registrations. registration_coverage separately compares GP registrations with resident population.

For nine conditions, QOF counts only people old enough to be included in that measure: for example, diabetes is measured among people aged 17 and over and osteoporosis among people aged 50 and over. ADI now provides two rates side by side. The original, unchanged rate is the estimated share of all residents. The new “QOF eligible-age” rate is the estimated share of residents in the relevant age range. They are two views of the same condition, not two groups to add together.

The distinction can be large. In 2024-25, England's osteoporosis rate is 0.450% of all residents but 1.201% of residents aged 50 and over; NHS England publishes 1.198% for the eligible population. The measures are not interchangeable. Existing analysis using the original rate remains valid under its existing whole-population definition, while comparison with published QOF prevalence should use the eligible-age rate.

Age-restricted does not mean age-standardised. The new rate leaves in differences in the age profile of people who are eligible; it does not adjust every area to a common age structure. It therefore cannot support a claim that comparisons have been “adjusted for age”. All new eligible-age rates are blank in 2014 because QOF 2013-14 supplied no separate age denominator; seven conditions are available from 2015 and asthma and non-diabetic hyperglycaemia join them from 2021.

The *_afflicted values in both views are modelled resident estimates — rate multiplied by the matching ONS population — not observed patients or QOF register counts. QOF practice lists and resident populations are different administrative measures, so these estimates will not exactly equal national register totals.

Known limitations

• Claimant Count rounding and definition. This is JSA plus the relevant UC component, not the total UC caseload or a count of unique people during the year. Nomis independently rounds each monthly observation to the nearest five. Published zeroes are rounded values, not suppressed blanks, and rounding can move low-count annual values either upward or downward.
• Forest of Dean claimant anomaly. Forest of Dean 010C rises from 0.7363% in 2023 to 9.9438% in 2024 and falls to 1.9172% in 2025. The rise is present in current Nomis monthly, sex, age and parent-MSOA data and not in neighbouring LSOAs, so it is not created by ADI averaging or crosswalking; no reliable local explanation was found, and the retained source value remains locally unverified.
• Crime definition and coverage. Anti-social behaviour is governed separately and is excluded from the 13-category headline recorded-crime total. Incidents without an LSOA code are dropped, exact duplicate identified incidents are removed, and British Transport Police is excluded because national rail-passenger exposure has no comparable resident denominator. A territorial force-year is blank if it lacks 12 non-empty monthly files or falls below 90% LSOA geocoding. Current unavailable footprints are Avon and Somerset (2016–2019, 2025), Staffordshire (2018), Lancashire, Thames Valley and Suffolk (2019), Greater Manchester (2019–2025), and Gloucestershire (2020–2022). Higher-level rates use only metric-specific coverage populations.
• QOF coverage and missing values. Practices without a usable register are not treated as zero. Covered-practice weights are renormalised, and estimates below 80% disease-specific registration coverage are withheld. Interior gaps of at most two years may be linearly interpolated between observations; leading, trailing and longer gaps remain blank. At the series endpoints, 64 LSOAs have no health values in 2014, one has none in 2024 and two have none in 2025.
• QOF interpretation. QOF reflects GP-diagnosed and recorded conditions and can understate underdiagnosed disease or vary with recording practice. The nine eligible-age groups are asthma (6+), rheumatoid arthritis (16+), diabetes (17+), CKD, depression, epilepsy, NDH and obesity (18+), and osteoporosis (50+); obesity used 16+ in output year 2015. The all-resident and eligible-age rates answer different questions and must not be combined. The eligible-age rate is age-restricted, not age-standardised, so area comparisons can still reflect age structure within the eligible population.
• Population vintage. For 2021+ the 2011-vintage population is unavailable, so 2020 is used inside domain processing. Published outputs are reset to each year's LSOA 2021 estimate through 2024. The LSOA-level series has not published mid-2025, so ADI carries mid-2024 forward for one year at every level to preserve aggregation consistency, although ONS mid-2025 figures exist for LADs, regions and England. ADI's included-LSOA England value is 58,611,150 versus ONS mid-2025's 58,834,812 (0.38% lower); the pipeline refuses a second stale year.
• Complex boundary changes. Six target LSOAs are excluded from the 2011→2021 conversion: St Albans 021C, Stevenage 013A, Welwyn Hatfield 017C, East Hertfordshire 019C, Gateshead 029D and Northumberland 043F. The release therefore contains 33,749 of 33,755 English LSOA 2021 areas and its England population is about 9,000 lower than the complete ONS LSOA total. The local effect reaches 1.90% of Stevenage's 2021 population.
• Health years are offset by one. QOF runs April to March, and the ADI labels a QOF year by the year it ends — so health “2021” is QOF 2020-21. Employment and crime years are calendar years.
• QOF 2020-21 comparability. NHS Digital warns that implementation changes may make indicator values inaccurate and comparisons with earlier years unreliable. Obesity was particularly affected, while asthma and COPD register definitions changed; this is not evidence that every condition moved downward.
• Health exclusions and corrections. Impossible rates and implausible one-year spikes are rejected as missing, not capped. Eight epilepsy LSOAs in 2016 and seven heart-failure LSOAs in 2021 are excluded from aggregate coverage. Depression 2024 and osteoporosis 2015 are interpolated only where both adjacent-year anchors exist; unanchored LSOAs remain blank. CVD primary prevention is available for 2014–2020 and blank after withdrawal; one-year smoking and hypothyroidism groups are not published.

An automated validator checks the pipeline outputs at LSOA, LAD, Region and England level; see the pipeline repository.

The paper

The methodology behind the ADI is set out in full in the paper “An annual deprivation index for neighbourhoods in England”, by Lukas Kikuchi, Robert Calvert Jump, Jo Michell and Will Stronge (2024).

Download the paper (PDF) ↓

Data & reproducibility

The complete dataset is available as CSV, at every geographic level. The full pipeline (fetch, process and aggregate) is open source and reproducible end to end. All sources are public; no API keys are required.

Download the data → See what changed → Pipeline on GitHub ↗ How it complements the IMD →

Data is published under the Open Government Licence. Boundaries © ONS / Crown copyright.