Record Linkage Using The Northern Ireland Longitudinal Study GSS Seminar on Data Matching Mon 29 November, London Fiona Johnston NILS Research Support Unit

  Introduction to the Northern Ireland Longitudinal Study (NILS) incl. the Northern Ireland Mortality Study (NIMS)   Record Linkage Methodology using the NIMS: Issues and Biases   Research Based on the NIMS: Exemplar Projects & Findings   Research Support & Future Plans Presentation Outline

1. 1.Research-Driven   Cross-sectional studies: no information on change over time   Other UK LS   Other international mortality-based LS   Health and socio-demographic profile of NI 2. 2.Legislation   Confidentiality protected, and managed by NISRA, under census legislation   NISRA have consulted the following:   Information Commissioner for Northern Ireland   Office of Research Ethics   Health and Social Care Privacy Advisory Committee 3. 3.Funding   Infrastructure funded by the Health and Social Care R&D Division and NISRA   Research support function funded by ESRC and NI Government (OFMDFM) Background to the NILS and NIMS

1. 1.Northern Ireland Longitudinal Study (NILS) – 28% representative sample of NI population (c. 500,000), based on health card registrations, linked to:   2001 Census returns   vital events (births, deaths and marriages)   demographic & migration events AND   distinct Health & Care datasets 2. 2.Northern Ireland Mortality Study (NIMS) - enumerated population at Census Day (c.1.6 million), linked to:   2001 Census returns   subsequently registered mortality data Both NILS and NIMS linked to contextual and area-based data:   capital value of houses and property attributes   geographical indicators   settlement classifications   deprivation measures Overview of the NILS and NIMS

Contextual dataNIMS Core DataEvents 2001 Census 1.6m enumerated Deaths NIMS Database Individual Project Datasets Geographic indicators Property characteristics Area characteristics Structure of the NIMS

Datasets Routinely Linked to the NIMS Census Datasets 2001Contextual Datasets  Age, sex and marital status  Religion and community background  Family, household or communal type  Housing, including tenure, rooms and amenities  Country of birth, ethnicity  Educational qualifications  Economic activity, occupation and social class  Migration (between 2000 and 2001)  Limiting, long-term illness, self-reported general  health, care-giving  Travel to work LPS Property Data 2010 Capital and rating value (based on 2005 valuation exercise) - Household characteristics (no. of rooms, property type, floor space, central heating) and valuation - Estimated capital value Geographical Indicators Super Output Area Ward Local Government District Settlement Classifications Urban/Rural/Mixed Deprivation Measures 2005 & 2010 Multiple Deprivation Measure Individual Deprivation Domains GRO Death Events Datasets Deaths of sample members

NIMS Record Linkage Methodology   NIMS database based on 1.6m pop. at Census 2001   GRONI deaths data added to NIMS database on a six monthly basis   3 stage matching process:   exact computer matching   fuzzy computer matching   detailed manual searching   Create and run matching queries:   accept exact matches   manually confirm/refute fuzzy matches   clerical searching for unmatched records   check for duplicates and resolve

Linkage rates close to 100% not possible for NIMS – why? 1. 1.Non-enumeration at Census:   One Number Census methodology: imputation for adjusted est. total   Imputation varies by age, gender and geographical area   In NI enumerated 2001 Census total was 1,603,641 - an additional 81,626 people were imputed = overall imputation rate of 4.6% People who came to NI after 2001 and subsequently died: selective unrecorded migration 3. 3.Differences between the info collected on census form and death certificate Record Linkage: Issues and Biases

Study on potential biases: O’ Reilly, D., Rosato, M. & Connolly, S. (2008) Unlinked vital events in census-based longitudinal studies can bias subsequent analysis. Journal of Clin. Epid. 61: What are the characteristics of people whose events are not linked into the LS datasets? What does this mean for analyses using the LS? Record Linkage: Issues and Biases

Record Linkage Rates  59,396 deaths available from to be linked from  6% deaths (3,392) could not be matched ProcessNumber (%) All death records NI59,396 Exact matches45,496 (80.6) Fuzzy matches4,491 (8.0) Manual matches2,093 (3.7) Linkage through HCR951 (1.7) Unlinked3,395 (6.0)

Characteristics of matched & non-matched deaths Based on data from death records and compared by Multivariate Logistic Regression:   Year of registration   Socio-demographic details   age, sex, marital status, social class (NS-SEC)   Place of death   home, hospital, nursing/residential home   Area in which they lived (SOA)   Deprivation (Income domain),   Urban/rural   Population density   Imputation   Cause of death

Age and sex distribution of unlinked death records Number of Deaths Proportion of Deaths

Variation according to demographic characteristics (deaths and results of log. regression) Aged less than 65Aged more than 65 SexDeathsORDeathsOR Male8, , Female4, ***31, * Marital status Married7, , Single3, ***8, *** Widowed ***27, *** Sep/Divorced1, ***1, *** Place of death Home6, , N/R home1, , *** Hospital5, ***31, *** *** P<0.001; ** P< 0.01; * P<0.05

Variation according to relative deprivation (deaths and results of log. regression) Aged less than 65Aged more than 65 DeathsOdds ratiosDeathsOdds ratios Least Deprived 1,831 (6.8%) ,543 (5.7%) nd 2,137 (8.8%)1.1911,103 (5.4%) rd 2,554 (9.5%)1.2011,933 (6.0%) th 2,901 (10.4%) ,534 (5.2%) 0.84 * Most Deprived 3,530 (16.0%) 1.78 ***11,374 (7.2%) 1.23 ** *** P<0.001; ** P< 0.01; * P<0.05

Variation by cause of death (deaths and results of log. regression) All agesUnder 65 years old Deaths (%unmatched) All causes70,289 (6.9%)13,071 (11.1%) I.H.D13,970 (5.6%)2,064 (9.4%) Stroke7,211 (6.8%)542 (8.9%) Respiratory Disease9,722 (7.0%)802 (9.9%) Cancer18,572 (5.6%)4,846 (8.1%) All External causes2,634 (15.2%)1,648 (20.3%) Accidents1,719 (12.3%)830 (18.2%) Suicides702 (19.9%)649 (21.4%) Other Causes12,840 (8.9%)2,579 (13.6%)

Research conclusions: small proportion of events are not linked – biases:   increase in months immediately after Census Day 2001   increase with ‘distance’ from the census   are non-random and more frequent in … younger males, older females people who are perhaps more socially isolated amongst residents of nursing/residential homes deprived areas where enumeration is low Non-linkage may limit the ability to study some causes of death and potentially lead to an underestimation of social gradients Record Linkage: Issues and Biases

  potential biases yet: statutory obligation to record death events and is therefore complete & good quality data – long experience of use for mortality analyses AND always be biases in every linkage study ≠100% - this research shows that biases can be quantified   small number problems i.e. falling death rates, population sub-groups (minority ethnics), cause-specific mortality (suicides, trauma & specific cancers) yet: can increase length of follow-up study, aggregate sub- populations & increase cohort size However ….

Health & Mortality:   Temperature-related mortality and housing (DSD)   Socio-demographic and area correlates of suicides   Distribution of cancer deaths in Northern Ireland by population and household type (NI Cancer Registry)   Variations in alcohol related deaths in Northern Ireland Demography:   Vital events: Standard Table Outputs (DMB) Section 75 (Equality Analyses)   Equality assessment of health outcomes: cause-specific mortality for Section 75 groups (DHSSPS)   Mortality rates and life-expectancy: Section 75 groups and social disadvantage (OFMDFM)   Religious affiliation and self-reported health   Denominational differences in short-term mortality   Mortality risk for carers Research Based on the NIMS

Exemplar Project & Research Findings   A study of the socio-demographic and area correlates of suicides in NI (Project 005) O’ Reilly, D., Rosato, M., Connolly, S. and Cardwell, C. (2008) Area factors and suicide: 5-year follow-up of the Northern Ireland population. Br J Psychiatry 192(2):

Background: Suicide rates vary between areas: is this due to individual characteristics (composition) or area characteristics (context)? Aim: To determine if area factors are independently related to suicide risk after adjustment for individual and family characteristics. Method: A 5-year record linkage study, based on the NIMS database, was conducted of c. 1.1 million individuals (not in communal establishments) aged 16–74 years, enumerated at the 2001 Northern Ireland Census. - data anonymised and held in a safe setting Area Factors & Suicide (i)

Results: 1. 1.The cohort experienced 566 suicides during follow-up Suicide risks: i. i.lowest for women and for those who were married or cohabiting; ii. ii.strongly related to individual and household disadvantage and economic and health status The higher rates of suicide in the more deprived and socially fragmented areas disappeared after adjustment for individual and household factors There was no significant relationship between population density and risk of suicide. Area Factors & Suicide (ii)

Conclusions: Differences in rates of suicide between areas are predominantly due to population characteristics rather than to area-level factors. Policy implication? Policies targeted at area-level factors are unlikely to significantly influence suicides rates. Area Factors & Suicide (iii)

Suicide (Daily Mirror)

NILS Research Support Unit   Based: Centre for Public Health (QUB) and NISRA HQ (McAuley House)   Support: 2 full-time and 1 half-time Research Support Officers   Set-up: April 2009 Remit:   raise awareness of the NILS research potential;   assist with development of research ideas and projects;   facilitate access to NILS data;   training & advice in use and analysis of NILS datasets;   promote policy relevance; and   enhance NILS research capacity incl: specific duty to assist government researchers and to undertake exemplar public policy research. Research Support

NILS data are sensitive and access is highly controlled:   researchers can access data only within a ‘secure setting’ (NILS- RSU office at McAuley House); arrangements can be made to run analyses remotely;   researchers must sign and abide by user licenses & security policies;   disclosure control thresholds in place to protect confidentiality of the data: no tabulated cell counts less than 10; and   all outputs must be cleared by NISRA staff. Access

Ongoing/Pending:   Inter-Censal Migration Flows   Mortality after death of a spouse: Is risk the same for all groups?   Religion, fertility and space: impacts on the future school population of Northern Ireland.   Exploratory analysis of the use of antibiotics by demographic and area characteristics Potential:   Pharmaco-epidemiological studies using Prescribing data   Cancer research Northern Ireland Cancer Registry data   Hospital admissions using Hospital Inpatient System data   Healthcare Associated Infections using laboratory testing data Current Project Activity

The help provided by the staff of the Northern Ireland Longitudinal Study and the Northern Ireland Mortality Study (NILS and NIMS) and the NILS Research Support Unit is acknowledged. The NILS and NIMS are funded by the Health and Social Care Research and Development Division of the Public Health Agency (HSC R&D Division) and NISRA. The NILS-RSU is funded by the ESRC and Northern Ireland Government. The authors alone are responsible for the interpretation of the data. Acknowledgements

NILS Research Support Unit Northern Ireland Statistics and Research Agency McAuley House 2-14 Castle Street Belfast BT1 1SA Tel: Website: nils-rsu.census.ac.uk