Presentation on theme: "Nursing, Midwifery and Allied Health Professions Research Unit e-Health data linkage Nadine Dougall & Paul Lambert."— Presentation transcript:
Nursing, Midwifery and Allied Health Professions Research Unit e-Health data linkage Nadine Dougall & Paul Lambert
Scotland a good laboratory for eHealth research –One of a handful of countries with indexed electronic health records spanning decades –Community Health Index (CHI) number a unique patient identifier introduced in 1970s (~10M records), full coverage since 1988, in widespread use secondary and primary care –1989: creation of permanently linked national datasets, Scottish Morbidity Records (SMRs) –Several significant social surveys with relevant health data e.g. Scottish Health Survey (SHeS) and British Household Panel Survey (BHPS)/ Understanding Society –SHeS-SMR linked dataset
eHealth research project DAta Management through E-social Science (DAMES) –Collaboration Universities of Stirling & Glasgow –Inter-disciplinary research spanning sociology, economics, public health and computer science –8 programme themes –deals with data on occupations, education, ethnicity/migration, social care and e-Health Funded by Economic & Social Research Council Chief Investigator Dr Paul Lambert, University of Stirling
DAMES eHealth research DAMES eHealth topic area & study design –Interested in mental health and risk factors for suicide –Data request and permissions sought NHS to do record linkage on all deaths coded as suicide or undetermined deaths since records began (~30 years) –Retrospective suicide cohort study using vital events death data linked to hospital episodes –NHS provided pseudonymised extracted datasets of suicide completers age 15y –Stored on secure server at University of Stirling, linked datasets using Stata v10
DAMES eHealth Summary statistics for the linked suicide cohort –Deaths: individuals –SMR01 (physical health): individuals with 5419 hospital episodes –SMR02 (maternity): 329 individuals with 1082 hospital episodes –SMR04 (mental health): 4207 individuals with hospital episodes
Summary data – ICD-10 main condition codes
Summary data – hospital episodes by gender
Summary data – suicide by occupation
DAMES eHealth Conclusions from SMR e-Health data: -Different socio-economic factors are associated with suicide rate e.g. occupation, education, deprivation -Significant variation in secondary care e-Health records for all prior admissions -75% males have no mental health records compared with 58% females -there is substantial variation in admission patterns males to females; many have no previous admissions, many have 10 or more
DAMES eHealth and population surveys Questions: -What factors are associated with people who have symptoms of possible psychiatric disorder, but do not see their GP? -Can population health survey datasets such as BHPS or SHeS help with characterisation of this group? -Both BHPS and SHeS use the General Health Questionnaire (GHQ-12); score of 4 or more indicating psychosocial distress, symptoms consistent with possible psychiatric disorder. Can we recode shared variables (SV) in both surveys and use to: -impute data related to GHQ-12 and the Client Interview Schedule-Revised (CIS-R) collected on SHeS into the BHPS? -conversely, impute data e.g. related to risk of unemployment from the BHPS into the SHeS?
DAMES eHealth and population surveys Example schema of exploratory analysis of matching criteria in BHPS & SHeS: Shared variables in both surveys recoded for equivalence
BHPS and SHeS imputation of data Schema: imputation of data from one survey on basis of the other BHPS variablesSVSHeS variables BHPS records (n=229k) ………………………………………………………………………………………. D=B[SV] + e (1) …. PredC=S[SV] (2) SHS records (n = 3k) PredD=B[SV] (3)…. …………………… C=S[SV] + e (4) (1)regression model for outcome variable D predicted by a set of coefficients B applied to shared variables within BHPS (2)calculation of predicted values for this variable for BHPS using shared variables. (3)calculation of predicted values for this variable using shared variables. (4)regression model for outcome variable C predicted by a set of coefficients S applied to shared variables within the SHeS
BHPS and SHeS imputation of data Example 1: Logit model using SHeS to predict anxiety/ depression (PD), predicted values of factors imported to BHPS
BHPS and SHeS imputation of data Example 1: Logit model using SHeS to predict anxiety/ depression, predicted values of factors imported to BHPS The disaggregation could be extended to further explore or model the characteristics of, and the past and/ or future behaviour over time, of those with predicted depression or anxiety and who do not visit the doctor.
BHPS and SHeS imputation of data Example 2: Logit model using BPHS to predict risk of being unemployed next year; predicted unemployment correlated with current health outcomes in SHeS Although the numbers involved are small, we identify small numbers of cases which might be thought high risk – both having poor mental health at present, and having socio-economic/socio-demographic profiles which suggest they are at risk future socio-economic misfortune.
Summary Clear potential for linked health datasets Potential for linking variables from different social surveys What next? Future analysis using as many shared variables as possible in Logit models What next for record linkage? –Research e.g SHeS-SMR, SMR-Census –Safe havens and organisational infrastructure e.g. SHIP, SAIL, ADLS, SDS…
THANK YOU UoS: Paul Lambert, Margaret Maxwell, Alison Dawson NeSC: Richard Sinnot, Susan McCafferty, John Watt NHS ISD: Anthea Springbett, Carole Morris, David Clarke ScotCen: Catherine Bromley, Joan Corbett, Lisa Given Nadine Dougall Senior Research Fellow NMAHP Research Unit Iris Murdoch Building University of Stirling Delivering, supporting and promoting high quality research to improve health Nursing, Midwifery and Allied Health Professions Research Unit