Search

IntroductionTo prevent further vascular events, prescribing of multiple classes of medications (antihypertensive, antithrombotic and lipid-lowering) is recommended in national clinical guidelines following ischaemic stroke. Objectives and ApproachUsing real-world data, we determined the association between optimal combination pharmacotherapy (supply of all three classes, “optimal pharmacotherapy”) and survival after stroke. We linked a cohort of patients with first-ever ischaemic stroke from the Australian Stroke Clinical Registry (2010-2014) with national pharmaceutical dispensing and national mortality data. Cox regression was used to determine associations between pharmacotherapy in the first 30 days of stroke with 1-year (from day 31 to 395) all-cause mortality. All analyses were adjusted for socio-demographic (age, sex) and clinical characteristics (stroke severity, discharge destination). ResultsAmong 6684 patients discharged following first-ever ischaemic stroke (median length-of-stay 5 days), 6466 patients who survived to 30 days were included (44% female, median age 74 years). During the first 30 days from discharge, 71.4% received ≥1 medication class, and 32.9% (n=2125) received optimal pharmacotherapy. Patients with optimal pharmacotherapy were older (≥75 years 50.3% vs

IntroductionThe Dementias Platform UK (DPUK) Cohort Explorer is an interactive, online visualisation tool that allows users to explore data for a number of DPUK cohorts. Over 30 variables across cohorts have been harmonised, including information on demographics, lifestyle, cognition, health, and genetic biomarkers. Objectives and ApproachThe tool has been developed to complement existing DPUK cohort metadata to provide a visual representation of participant numbers and field-level information for a selection of cohorts. This enables users to determine a cohort’s eligibility before applying for access to a cohort’s data, and aid in shaping potential hypotheses. Developed using Microsoft PowerBI, the Explorer hosts a subset of the cohort’s baseline, harmonised data, allowing a user to interrogate the visualisations of the uploaded data in a secure manner on the DPUK Data Portal website. Visualisations are linked so that participant numbers and distributions can be explored interactively. ResultsThis approach allows the user to explore the harmonised data across a number of cohorts simultaneously whilst setting and adjusting filters that are of interest to the user’s search criteria. This provides a better understanding of the real-world data and enables the user to determine the feasibility of each cohort for potential studies, whilst facilitating meaningful comparisons across cohorts. The tool currently visualises five DPUK cohorts with a total of 82,391 participants, however it is being incrementally developed with more cohorts being added continually. Conclusion / ImplicationsBy combing an easy-to-use, interactive dashboard with harmonised sets of real-world data, the tool allows the user to explore, interrogate and better understand field-level information in a secure manner with zero data transfer. This provides more insight for the user when applying for access to a cohort dataset using the DPUK Data Portal and may help the user to make more informed decisions and/or hypotheses.

IntroductionData Linkage of hashed datasets is much more difficult than linking in-the-clear data. Hashing prevents the use of matching tools that overcome messy data such as ‘contained-within’ functions and edit distance metrics. Hashing sensitive data received from third parties is becoming more common due to increased Data Security concerns. Institutions need to be ready to link hashed data with high accuracy, otherwise the quality of outputs from these linked datasets will suffer. Objectives and ApproachWe designed an innovative matching method, Derive and Conquer (D&C). We derived variables containing substrings or patterns of the full variable (e.g. Soundex or first 4 characters of a string) to match on instead. However, using lots of combinations of these derived variables would require thousands of traditional match keys to be programmed, run, and reviewed. Instead, D&C runs matchkeys on a derived agreement variable which amalgamates information stored in multiple derived variables into one value, reducing the number of matchkeys to a manageable amount. D&C runs on distributing computing systems using PySpark to link datasets containing millions of records in a timely manner. ResultsD&C was developed using in-the-clear UK Census and health records with results comparable to the in-the-clear gold standard. It is currently being tested on hashed data to link UK tax and benefits data to UK health records. 66.4 million records were declared matched - a realistic match rate for the UK population. Research into the linkage quality is ongoing to produce estimates on the amount of bias in the linkage and the precision and recall. We will be excited to present these results at the Conference in October. These results will be used to improve D&C. Conclusion / ImplicationsUsing these derived variables, we have been able to overcome the challenge of matching massive hashed datasets with a realistic match rate and in a realistic time frame.

IntroductionThe Inverse Care Law (ICL) programme in Wales was setup to tackle health inequalities. Eligible populations from deprived communities, at higher risk of cardiovascular disease (CVD) were invited to a health-check and offered appropriate lifestyle and clinical interventions. Objectives and ApproachEvaluation of this programme is vital to ensure that targeted interventions have been received by those most in need, including referrals to lifestyle services and support. The use of longitudinal population-scale routine-data required the development of an approach which was both efficient and cost effective. To achieve this, the Welsh Longitudinal General Practice (WLGP) data held in SAIL Databank was utilised. A programme-specific methodology was agreed by the programme-board and developed so that data collected from GP records prior, during and post health-check accurately identified the eligible population and allowed the effective assessment of lifestyle and clinical risk factors for CVD; poor diet, physical inactivity, smoking and high alcohol intake, so appropriate interventions could be offered. ResultsWe evaluated the programme from 2015 to 2019 in 70 GP’s across the participating Health-Boards, and identified 175,671 individuals eligible by the programme criteria. Substantial preliminary work has been carried out to ensure the specification of outcome measures are both clinically and epidemiologically accurate and relevant. The final report scheduled for release in August-2020, which will evaluate the impact of the programme. Conclusion / ImplicationsThis ambitious evaluation of a large-scale programme set in the community involving disparate systems and a range of stakeholders, has been both complex and challenging, requiring substantial effort to design and implement. We hope the outcomes and lessons learned from our experience will improve the design, implementation and evaluation of the programme and lead to improvements in services and the quality of life for people in Wales, and provide an exemplar for health care providers worldwide wishing to conduct similar programmes in the future.

Introduction“The Consensus Statement on Public Involvement and Engagement with Data-Intensive Health Research”, recent data breaches, and growing public awareness and controversy associated with secondary use of health data all highlight the need to understand what data sharing the public will support, under what circumstances, for what purposes and with whom. Objectives and ApproachThis symposium explores methods and findings from public engagement at all stages of data linkage research, beginning with short presentations (~6-8 minutes) on recent work: Mhairi Aitken: Consensus Statement - principles and an application using deliberative workshops to explore public expectations of public benefits from data-intensive health research Annette Braunack-Mayer/Felicity Flack: Surveys and citizens’ juries: Sharing government data with private industry Kim McGrail/Mike Burgess: Public deliberations on cross-sector data linkage, and combining public and private sources of data Alison Paprica: Plain language communication informed by Health Data Research Network Canada’s Public Advisory Council. Half the session will be spent interacting with the audience through live polling. The moderator will post a series of poll question such as “What is the most important thing for meaningful public engagement?” to prompt audience thinking on the topic. After the audience responses are revealed, panelists will share their own views about what they think is the best answer, and the main reason(s) behind their choice. The last 10-15 minutes of the session will be reserved for Q&A and dialogue with the audience. ResultsWe anticipate that this approach will surface emerging and tacit knowledge from presenters and the audience, and augment that through generative discussion. Conclusion / ImplicationsSession attendees will leave with a better understanding of the current state of knowledge and ways to talk about that understanding with other researchers, policy makers and the public.

IntroductionStudies of select groups (e.g. pre-industrial or subsistence farming populations) suggest that children had improved survival if their grandparents were alive, particularly maternal grandmothers (MGMs). This question is unanswered for Aboriginal Australian communities, where care of children is commonly shared by extended family members, and it is not known if this association holds for physical health in addition to mortality. Objectives and ApproachWe examined relationships between the health of Aboriginal children born in Western Australia from 2000 to 2013 and the health and vital status of their grandparents. Birth, death, inpatient hospital and emergency department records of all Aboriginal children and their grandparents were linked. Grandparents were categorised as ‘healthy’ (alive with Charlson comorbidity index score of 0 or 1), ‘unhealthy’ (alive with a score of 2 or more), or dead when the child was born. Results27,425 Aboriginal children linked to their MGM (at a minimum). Mortality up to age 2 was lower with healthy grandparents than unhealthy or dead grandparents (e.g. 11 deaths per 1000 live births with healthy MGM; 22 with unhealthy MGMs and 16 with dead MGMs). Children also had fewer acute healthcare contacts (e.g. 13% with healthy MGMs spent at least 7 days in hospital in their first 2 years compared with 19% with unhealthy or dead MGMs). However, healthcare contacts was largely unrelated to grandfathers. Outcomes were also associated with the number of living grandmothers. (e.g. 1.5% of children with 2 grandmothers were discharged against medical advice in their first 2 years, compared to 2.7% with one healthy grandmother and 3.7% with no grandmothers). Conclusion / ImplicationsChildren with healthy grandmothers have lower mortality and morbidity, possibly because of their care giving. These associations are unlikely to be due to genetic or environmental factors, as they are weaker or missing for grandfathers.

IntroductionRegular contact with a general practitioner (GP) has been shown to lower the risk of potentially avoidable hospitalisations (PAHs) independently of continuity of provider and frequency of contact. Multimorbidity affects between 55 and 98% people aged 65+ years and continues to place pressure on healthcare systems globally. However, little is known about its impact on the relationship between continuity of primary care and PAHs. Objectives and ApproachA retrospective, longitudinal cohort study using survey data linked to routinely-collected administrative health data from the 45 and Up Study conducted in New South Wales, Australia was used to investigate the effect measure modification by multimorbidity on the relationship between regularity of GP contact and PAHs. Multimorbidity was assessed using the Rx-Risk comorbidity score, which captures the number of condition groups, assigned based on medicine dispensing records, using a 5-year look-back period. PAHs were: (i) any unplanned hospitalisations, (ii) chronic ambulatory care sensitive conditions (ACSC) hospitalisations or (iii) unplanned ACSC hospitalisations. Multivariable logistic regression and population attributable fractions (PAF) were used to examine effect measure modification by multimorbidity. ResultsHigher GP regularity was significantly associated with a reduction in the probability of each PAH type. This reduction diminished with increasing multimorbidity with the effect measure modification most apparent for chronic ACSC and unplanned chronic ACSC hospitalisations. The PAF of moving to the highest quintile of regularity significantly reduced with increasing multimorbidity. For example, a reduction in the PAF of unplanned ACSC hospitalisations of 31.1% was observed in those with a RX Risk score of >10 (17.8%) compared with those with no multimorbidity (48.9%). Conclusion / ImplicationsWeakening of the relationship between GP visit regularity and PAHs with increasing levels of multimorbidity suggests a need to focus on improving primary care support to prevent PAHs for patients with multimorbidity.

IntroductionMonash University’s health researchers are at the forefront of Health, Epidemiological and Translational research. Powered by research funding of over $69 million per year, Monash encompasses 13+ registries, 44+ clinical trials and 47+ cohort studies underpinned by complex multimodal, large and often sensitive datasets. Many of these research activities are occurring in isolation. Opportunities exist to better deliver higher impact research at scale through capacity building in data sharing. Objectives and ApproachWe have designed this session to demonstrate how significant investment in health data platform infrastructure by a University, can support best practice in data linkage and health research. ResultsContent will include: An overview of Helix – people as a platform, research stakeholders, and key capabilities. Institutional data governance and sharing capabilities – linking two or more researcher held datasets exampled by a use-case of linking biobank data to a clinical registry dataset. Monash Helix data linkage working group – established to develop a community of practice to advance data linkage capabilities across Monash with a focus on: knowledge sharing; ethics, governance and data security; analytic training; and improved access to external administrative datasets. Secure eResearch Platform (SeRP) – developed by Swansea University who have partnered with Monash University to bring their best practice framework and technologies in managing sensitive health data to Monash researchers and their national and international collaborators. Future directions - participation in defining the strategic priorities of the Health Studies National Data Asset Program by creating distributed data infrastructure capability to develop, share and link FAIR data collections from the government, research and health care sectors. Conclusion/ImplicationsThe session will facilitate discussions on how we can build on these activities, both nationally and internationally, to develop a community of practice around knowledge sharing and best practice in data linkage.

IntroductionIn Australia, the notification rate for hepatitis C is 5 times greater among Aboriginal and Torres Strait Islander peoples than non‑Indigenous populations (164 vs. 35 per 100,000). Infection with Hepatitis C virus is associated with increased morbidity, mortality and health-related costs. Recently, the use of simple, tolerable and short-duration HCV therapies with extremely high efficacy (cure rates >90%) became available through the Pharmaceutical Benefits Scheme (PBS) on 1 March 2016. Objectives and ApproachOverall, this project aims to identify inequities in hepatitis C healthcare for Aboriginal and Torres Strait Islander peoples across SA and NT. By using 19 routinely collected health administration datasets, we will evaluate hepatitis C diagnoses, treatment uptake, and treatment outcomes, with focus on identifying patterns in healthcare access among Aboriginal and Torres Strait Islander people. The study links state-based and federal datasets, including Medicare, Pharmaceutical Benefits Scheme, and National Mortality Database, to hepatitis C notifications reported in SA since 1992 or in NT since 1999 using probabilistic data linkage methodology in a secured online environment. ResultsThe study population includes all hepatitis C notifications recorded in SA and NT, totalling more than 25,000 cases. Preliminary findings will be reported including incidence of hospitalisation, mortality, and morbidity by Aboriginal and Torres Strait Islander status. Conclusion / ImplicationsThis project is of national importance to address hepatitis C related morbidity and mortality in Australia, particularly among Aboriginal and Torres Strait Islander peoples. Now that highly effective direct acting anti-viral treatments are available, it is crucial to ensure key affected populations, including Aboriginal and Torres Strait Islander people, have access to these novel biomedical approaches. Findings will underpin future policy to reduce the burden of hepatitis C across SA and NT populations.

IntroductionStats NZ’s Integrated Data Infrastructure (IDI) is a linked longitudinal database combining administrative and survey data. Previously, false positive linkages (FP) in the IDI were assessed by clerical review of a sample of linked records, which was time consuming and subject to inconsistency. Objectives and ApproachA modelled approach, ‘SoLinks’ has been developed in order to automate the FP estimation process for the IDI. It uses a logistic regression model to calculate the probability that a given link is a true match. The model is based on the agreement types defined for four key linking variables – first name, last name, sex, and date of birth. Exemptions have been given to some specific types of links that we believe to be high quality true matches. The training data used to estimate the model parameters was based on the outcomes of the clerical review process over several years. ResultsWe have compared the FP rates estimated through clerical review to the ones estimated through the SoLinks model. Some SoLinks estimates fall outside the 95% confidence intervals of the clerically reviewed ones. This may be the result of the pre-defined probabilities for the specific types of links are too high. ConclusionThe automation of FP checking has saved analyst time and resource. The modelled FP estimates have been more stable across time than the previous clerical reviews. As this model estimates the probability of a true match at the individual link level, we may provide this probability to researchers so that they can calculate linked quality indicators for their research populations.

IntroductionThere are few examples of functional health information exchange environments or data linkage centres in the African settings. New opportunities are emerging as unique health identifiers and patient registration systems are being established nationally in many countries, and increasingly individuated health data are available for linkage, often linked to priority global health initiatives such as to support HIV and tuberculosis services. Objectives and ApproachWe sought to establish a province-wide health information exchange and data centre for individuated health data, leveraging a unique identifier and available individuated data. The intention with the Provincial Health Data Centre (PHDC) was to create a single well-governed environment which could simultaneous fulfil the functions of an exchange directly supporting care, as well as support research requests. Often the demands of academics, funders and global agencies for reporting result in data consolidation for research and reporting taking precedence over service delivery. ResultsThrough pragmatic use of data from all sources, the PHDC is able to usefully enumerate many health conditions of interest with sufficient fidelity for both service and research purposes. For research data requests, there has been a huge improvement in data governance alongside increased data availability as a result of the single environment with clear procedures for patient protection, and the benefits of data linkage prior to anonymisation. Many of the inference approaches have benefited hugely from interactions with researchers, which has in turn improved the quality of outputs for routine care. Conclusion / ImplicationsIn maturing digital health environments which are establishing consolidated data environments for the first time, aspiring from the start to a single well-governed environment for both patient care and research, is a virtuous model with many benefits over fragmented data linkage efforts

IntroductionAboriginal Community-Controlled Health Services (ACCHSs) routinely collect sexually transmissible infection (STI) and blood-borne virus (BBV) clinical data from individuals through electronic medical records (EMRs). At a population level these data are an underutilised resource and can be used to help drive clinical and public health interventions. Accordingly, we have established a national STI and BBV sentinel surveillance network for ACCHSs —the ATLAS network. Objectives and ApproachIn establishing the ATLAS network, we standardised and automated extraction of deidentified clinical data from multiple EMRs using The University of Melbourne’s GRHANITE™ software. Regular reports are provided to each ACCHS, addressing 12 clinical indicators considered best practice in STI and BBV control and management. Data are used to drive Continuous Quality Improvement strategies, can be aggregated at regional and national levels, and have the ability to link to other surveillance networks also using GRHANITE™. ResultsThe ATLAS network currently includes more than 30 ACCHSs, including urban, regional and remote areas. To date, data have been analysed for >50,000 individuals aged ≥15 years attending medical consultations at ACCHS between January 2016 and December 2018. For clients in the key age-range for STI testing of 15–29 years, the mean annual rate of chlamydia/gonorrhoea testing was 25% (range: 1–59%). The mean annual rate of chlamydia positivity for clients 15–29 years was 3% (range: 0–19%) and the mean annual rate of gonorrhoea positivity was 2% (range: 0–18%). Conclusion / ImplicationsAn integrated network of STI and BBV testing and management data has been successfully established amongst ACCHSs nationally. The ATLAS network can be readily expanded and has the potential to link to other datasets using the GRHANITE™ tool. ATLAS provides a unique data infrastructure with capacity to support novel research programs as well as underpin quality improvement activities within participating sites.

Background/ObjectivePatients with IgA vasculitis (IgAV aka Henoch Schonlein) may require aggressive initial treatment, are prone to disease relapses and may have sustained abnormal IgA responsiveness. We investigated whether the relatively rare diagnosis of childhood IgAV is associated with subsequent comorbidity, hospitalisation and interventions. MethodsUsing linked administrative health data from the Western Australian Rheumatic Disease Registry we performed an observational cohort study comparing hospitalisation and ED visit rates(per 1000 person years), comorbidity accrual (Charlson comorbidity index; CCI) and procedures between 494 IgAV patients < 20 years and 1385 matched controls over a twenty-year period. ResultsHospitalisation was proportionally (73.5 vs 51.5%) and by rate (21.7 vs 18.9; rate ratio 1.15, p

IntroductionLinked health record collections, when combined with large longitudinal surveys, are a rich research resource to inform policy development and clinical practice across multiple sectors. Objectives and ApproachThe Australian Longitudinal Study on Women’s Health (ALSWH) is a national study of over 57,000 women in four cohorts. Survey data collection commenced in 1996. Over the past 20 years, ALSWH has also established an extensive data linkage program. The aim of this poster is to provide an overview of ALSWH’s program of regularly up-dated linked data collections for use in parallel with on-going surveys, and to demonstrate how data are made widely available to research collaborators. ResultsALSWH surveys collect information on health conditions, ageing, reproductive characteristics, access to health services, lifestyle, and socio-demographic factors. Regularly updated linked national and state administrative data collections add information on health events, health outcomes, diagnoses, treatments, and patterns of service use. ALSWH’s national linked data collections, include Medicare Benefits Schedule, Pharmaceutical Benefits Scheme, the National Death Index, the Australian Cancer Database, and the National Aged Care Data Collection. State and Territory hospital collections include Admitted Patients, Emergency Department and Perinatal Data. There are also substudies, such as the Mothers and their Children’s Health Study (MatCH), which involves linkage to children’s educational records. ALSWH has an internal Data Access Committee along with systems and protocols to facilitate collaborative multi-sectoral research using de-identified linked data. Conclusion / ImplicationsAs a large scale Australian longitudinal multi-jurisdictional data linkage and sharing program, ALSWH is a useful model for anyone planning similar research.

IntroductionAround the world, the need for mental health, addictions and home care services is growing. Government commitment and collective efforts to bridge data gaps, develop indicators and publicly report results are key elements in Canada’s efforts to improve access. Objectives and ApproachThis symposium will demonstrate how a coalition of stakeholders united to use real-world data to measure progress and drive change. Each presentation highlights a different aspect of the project with participant interaction, aiming for the Canadian context to spark knowledge exchange across sectors and countries: Presentation 1 - Coalitions and consensus (10 min.): processes and engagement for successful collaboration between governments, providers, measurement experts and people with lived experiences to select and develop indicators Facilitated Q&A (5 min.) Presentation 2 - Standards and data infrastructure (10 min.): new standards to enhance data comparability and strengthened data infrastructure to support measurement and reporting Facilitated Q&A (5 min.) Presentation 3 - Indicator development (10 min.): the indicator development cycle, methodological approaches using linked and partial data, and development strategies for new concepts Facilitated Q&A (5 min.) Presentation 4 - Public reporting and policy impact (10 min.): describes how public reporting supports sustained commitments and energizes change using targeted tools and messages Facilitated Q&A (15 min.) ResultsPublic reporting began in 2019, with 3 new indicators released annually over 4 years. Initial reporting provides a baseline to track improvements, and a starting point for health system planners to learn from peers across Canada. The indicators have been a catalyst to fill important data gaps in emergency and home care services. Conclusion / ImplicationsThrough shared priorities, coalitions and linked data, information gaps are being filled to drive advancements in access to mental health and addictions services, and home care. Lessons learned in Canada can be adapted internationally to galvanize needed improvements in these sectors.

IntroductionWhile the quantity and type of datasets used by data linkage projects is growing, there remain some datasets that are ‘not available’ or ‘hard to access’ by researchers and linkers, either due to legal/regulatory constraints restricting the release of personally identifying information or because of privacy or reputational concerns. Advances in privacy-preserving record linkage methods (e.g. PPRL-Bloom) have made it possible to overcome this impasse. These techniques aim to provide strong privacy protection while still maintaining high linkage quality. PPRL-Bloom methods are being used in practice. The Centre for Data Linkage (CDL) at Curtin University has been involved in several PPRL linkage and evaluation projects using real-world data. As the methods are relatively new, published information on achievable linkage quality in real-world scenarios is limited. Objectives and ApproachWe present and describe several real-world applications of privacy preserving record linkage (PPRL-Bloom) where the quality of the linkage could be ascertained. In each case, data was linked ‘blind’; that is, without linkers having access to the original personal identifiers at any stage, or having any additional information about the records. Evaluations include a linkage of state-based morbidity and mortality records, a linkage of a number of general practice datasets to morbidity and emergency records, and a linkage of a range of state-based non-health administrative data, including education, police, housing, birth and child protection records. ResultsThe privacy preserving record linkage performed admirably, with very high-quality results across all evaluations. Conclusion / ImplicationsPrivacy preserving linkage is a useful and innovative methodology that is currently being used in real world projects. The results of these evaluation suggest it can be an appropriate linkage tool when legal or other constraints block release of personally identifying information to third party linkage units.

IntroductionIn 2016, the Provincial Overdose Cohort (ODC) was created following the declaration of the public health emergency in British Columbia (BC), Canada. The ODC is a set of longitudinal and linked administrative data which identifies illicit drug-related overdose events, including death, ambulance, emergency room, hospital, physician, and prescription drug records. Objectives and ApproachThe ODC was developed to better understand factors associated with overdose in order to support response activities, prevent overdose deaths, and identify trends and opportunities for interventions. Person-level linkages were conducted using provincial health insurance and health history data; socio-economic information, mental and physical illness diagnoses, and corrections history were also appended. The ODC currently includes people who have had a drug-related overdose between January 1 st 2015 and December 31 st 2017 as well as a 20% random sample of the general population. ResultsThe ODC contains 36,576 overdose episodes and 23,161 people who have a drug-related overdose between January 1 st , 2015 and December 31 st 2017. Of the 23,161 people, 3,604 (15.6%) had a fatal overdose and 19,557 (84.4%) had a non-fatal overdose. 49.9% of people in BC who had an overdose were 20-39 years of age and 67.4% were males. From 2015 to 2017, the proportion of people experiencing 3 or more overdoses a year increased from 3.6% to 8.7%, respectively. There were an increasing number of fatal and non-fatal drug-related overdoses in BC during this time period. Conclusion / ImplicationsLarge population data linkages can be invaluable tools during a public health emergency. Collaborative partnerships and a shared data governance across jurisdictions was central in building the ODC and understanding how various social determinants of health impact risk of overdose among people who use drugs.

The Journey to Social Inclusion (J2SI) Social Impact Investment (SII) is Victoria’s first social impact investment. It has two outcome payments, for stable housing and reduction in hospital bed days. Linked administrative data is being used to calculate the performance levels and outcome payments from the Victorian Government to J2SI. This paper discusses the methodologies and challenges encountered in the use of the linked administrative data for the calculation of the outcome payments for the J2SI SII. IntroductionJ2SI will receive half of the program costs for service delivery based on the level of performance of two outcome payments. The outcome payments are the difference between the results for the Intervention Group and the Control Group. The results are being calculated using linked administrative datasets for health, housing and justice, by the Centre for Victorian Data Linkage (CVDL) and the Centre for Social Impact, University of Western Australia (CSI UWA) respectively. Objectives and ApproachThe paper will outline the approach taken to calculate the results for the Intervention Group and the Control Group, explore the benefits and challenges of arising from different parties being involved in the calculations and present the lessons learned for future social impact bonds in Victoria and other States/Territories. ResultsThe work has highlighted the importance of clearly understanding and documenting all of the datasets and how they are to be used prior to the commencement of the calculations. High level descriptions are not adequate preparation for the precision required for the actual calculations. The work has also highlighted the differences is using linked data for research and service implementation. Conclusion / ImplicationsLinked administrative data is a robust source of data for the calculation of outcome payments for social impact investments. However, with the timeliness of payments important for service delivery and financing, thorough preparation for the calculation of the results and outcome payments is essential.

IntroductionLittle is known of the potential impacts of introducing data linkage processes on response rates and biases in school-based surveys. Objectives and ApproachThis paper assessed: i) the impact on parental consent rates for student participation in a school survey; ii) sample representativeness; and iii) the quality of identifiable data provided to facilitate linkage. An option for data linkage was piloted in a sub-sample of schools participating in the Student Health and Wellbeing survey, a national survey of adolescents in Wales, UK. Schools agreeing to participate were randomized 2:1 to receive the data linkage question. Survey responses from consenting students were anonymised and linked to routine datasets (e.g. general practice, inpatient, and outpatient records). Parental withdrawal rates were calculated for linkage and non-linkage samples. Multilevel logistic regression models compared characteristics between: i) consenters and non-consenters; ii) successfully and unsuccessfully linked students; and iii) linked cohort and peers within the general population, with additional comparisons of mental health diagnoses and health service contacts. ResultsThe sub-sample comprised 64 eligible schools (out of 193), with data linkage piloted in 39. Parental consent was comparable across linkage and non-linkageschools. 48.7% (n=9,232) of students consented to data linkage. Consenting students were more likely to be younger, more affluent, have higher positive mental wellbeing, and report fewer risk-related behaviours compared to non-consenters. Overall, 69.8% of consenting students were successfully linked, with higher rates of success among younger students. The linked cohort had lower rates of mental health diagnoses (5.8% vs. 8.8%) and specialist contacts (5.2% vs. 7.7%) than general population peers. ConclusionIntroducing data linkage within a national survey of adolescents had no impact on study completion rates. Students consenting to data linkage, and those successfully linked, differed from non-consenting students on numerous key characteristics, raising questions concerning the representativeness of linked cohorts.

IntroductionCanadians report persistent problems accessing primary care despite an increasing per-capita supply of primary care physicians (PCPs). There is speculation that PCPs, especially those early in their careers, may be working less and/or choosing to practice in focused clinical areas rather than comprehensive family medicine. Objectives and ApproachOur objectives were to compare practice patterns among early-career (

IntroductionDespite recommendations, most studies examining health inequalities fail to report both absolute and relative summary measures. We examine colorectal cancer (CRC) survival for patients with and without severe psychiatric illness (SPI) to demonstrate the use and importance of relative and absolute effects. Objectives and ApproachWe conducted a retrospective cohort study of CRC patients diagnosed between 01/04/2007 and 31/12/2012, using linked administrative databases. SPI was defined as diagnoses of major depression, bipolar disorder, schizophrenia, and other psychotic illnesses six months to five years preceding cancer diagnosis and categorized as inpatient, outpatient or none. Associations between SPI history and risk of death were examined using Cox Proportional Hazards regression to obtain hazard ratios and Aalen’s semi-parametric additive hazards regression to obtain absolute differences. Both models controlled for age, sex, primary tumour location, and rurality. ResultsThe final cohort included 24,507 CRC patients, 482 patients had an outpatient SPI history and 258 patients had an inpatient SPI history. 58.1% of patients with inpatient SPI history died, and 47.1% of patients with outpatient SPI history died. Patients with an outpatient SPI history had a 40% (HR 1.40, 95% CI: 1.22-1.59) increased risk of death and patients with an inpatient SPI history had a 91% increased risk of death (HR 1.91, 95% CI: 1.63-2.25), relative to no history of a mental illness. An outpatient SPI history was associated with an additional 33 deaths per 1000 person years, and an inpatient SPI was associated with an additional 82 deaths per 1000 person years after controlling for confounders. Conclusion / ImplicationsWe demonstrated that reporting of both relative and absolute effects is possible and calculating risk difference is relatively simple using Aalen models. We encourage future studies examining inequalities with time-to-event data to use this method and report both relative and absolute effect measures.

IntroductionMusculoskeletal problems, including conditions such as back pain, neck pain, rheumatoid arthritis, gout and osteoarthritis are common in the population and significant contributors to global disease burden. Age is one of the most common risk factors for musculoskeletal conditions and over 40% of older people accessing residential aged care have a musculoskeletal condition. It is not known whether individuals living in the community with musculoskeletal conditions have similar needs to those in permanent care and this is important to know in order to provide appropriate care. Objectives and ApproachThe objective of this study was to profile individuals with musculoskeletal conditions in different aged care service settings (i.e. permanent care, community care only, transition/ respite care, or no services). Specifically, we examined the concurrent chronic conditions, health risk factors and functional limitations of individuals by service setting. A cross-sectional evaluation of individuals in the National Historical Cohort of the Registry of Senior Australians (ROSA) between 2004 and 2014 was conducted. Multivariable logistic regression models estimated the factors associated with being in different aged care settings. Odds ratios (OR) and 95% confidence intervals (CI) were determined. Results401,026 (42.5%) individuals with musculoskeletal conditions were assessed for aged care service eligibility during the study period. Of these 197,181 (49.2%) accessed permanent care, 37,003 (9.2%) accessed home care, 54,826 (13.7%) transition/respite, and 112,016 (27.9%) - no care. Individuals accessing community care compared to residential care were more likely to be female, have pain and have difficulty maintaining their home, as were individuals accessing no services compared to residential care. Conclusion / ImplicationsCompared to those in residential care, individuals with musculoskeletal conditions in the community with or without assistance had few differences related to other chronic conditions and functional limitations. But the reasons why some had support, while others did not, are unclear.

IntroductionLarge volumes of health data are generated through the interaction of individuals with hospitals, government agencies and health care providers. There is potential in the linkage and sharing of administrative data with private industry to support improved drug and device provision but data sharing is highly contentious. Objectives and ApproachWe conducted a scoping review of quantitative and qualitative studies examining public attitudes towards the sharing of health data, held by government, with private industry for research and development. We searched four data bases, PubMed, Scopus, Cinahl and Web of Science as well as Google Scholar and Google Advanced. The search was confined to English-only publications since January 2014 but was not geographically limited. We thematically coded included papers. ResultsWe screened 6788 articles. Thirty-six studies were included primarily from UK and North America. No Australian studies were identified. Across studies, willingness to share non-identified data was generally high with the participant’s own health provider (84-91%) and academic researchers (64-93%) but fell if the data was to be shared with private industry (14-53%). There was widespread misunderstanding of the benefits of sharing data for health research. Publics expressed concern about a range of issues including data security, misuse of data and use of data to generate profit. Conditions which would increase public confidence in sharing of data included: strict safeguards on data collection and use including secure storage, opt-in or opt-out consent mechanisms, and good communication through trusted agents. Conclusion / ImplicationsWe identified a research gap: Australian views on sharing government health data with private industry. The international experience suggests that public scepticism about data sharing with private industry will need to be addressed by good communication about public benefit of data sharing, a strong program of public engagement and information sharing conducted through trusted entities.

IntroductionChildren with special educational needs (SEN) often struggle academically. Previous studies found that children’s abilities in kindergarten are predictive of their future SEN status. It is currently unknown whether these predictors differ in children with and without an early identification of a special need in kindergarten (SN-K). Objectives and ApproachWe investigated early predictors of SEN in Grade 3, in a cohort of Ontario children, with and without SN-K (1,824 and 62,842, respectively), who attended kindergarten between 2003/04 and 2005/06. Early Development Instrument data, a teacher-completed checklist of children’s development, were linked to Grade 3 standardized reading, writing, and mathematics test scores. Controlling for children’s demographics, multivariate binary logistic regressions were conducted examining the association between children’s developmental outcomes, their functional impairments, the necessity for further assessment (all reported by their kindergarten teacher) and their SEN status in Grade 3. ResultsOverall, 69.8% of children with SN-K had SEN in Grade 3, while 11.6% of children without SN-K had SEN. Our analyses revealed that, for children with SN-K, having a functional impairment was the most significant predictor of having SEN in Grade 3 (Odds Ratio=3.61, 2.59-5.02 95% confidence interval). For children without SN-K, teachers reporting the need for further assessment was the strongest predictor of having SEN in Grade 3 in children without SN-K (Odds Ratio=2.70, 2.49-2.93). Conclusion / ImplicationsEarly predictors of SEN in Grade 3 differ for children who receive an early identification (SN-K) compared to those who don’t. How children with SN-K function in a classroom is the best predictor of SEN in Grade 3, while teachers’ observation that a child needs further assessment is the strongest predictor of SEN in Grade 3 for those without SN-K. Addressing these areas early on may help reduce the number of children with SEN in later grades and may positively impact their future academic success.

IntroductionResilience is a key factor in healthy development of children who have experienced adversity in early life. Current methods of assessment involve using questionnaires - few of which are appropriate for young children, most are time consuming and rely on parent recall, thereby introducing bias. Additionally, widespread implementation of these would be costly, making population-level assessment of resilience impractical. The current study will leverage multi-sector, linkable, whole-population data from health, education, justice and social services to explore alternative ways to assess resilience in children. Objectives and ApproachThe purpose of this study is to identify factors in administrative data that emerge as significant determinants of resilience, demonstrated by children who experience adversity in early life but develop normally. Children born in Manitoba between 2000-2012 will be included. Adversity will be identified as families receiving income assistance, and/or the presence of adverse childhood experiences (ACEs) in linked databases such as Justice (incarcerated parent) and Health (parent with mental or substance use disorder). Development will be measured using the Early Development Instrument (EDI). EDI outcomes will be assessed according to severity and frequency of adversity. Pre-identified covariates that map onto sub-constructs of resilience will be assessed using multivariable linear regression to determine whether they are associated with higher EDI scores in the context of adversity. ResultsThe identification of administrative data variables associated with resilience in children will serve as a valuable tool for population-level assessment of resilience among children who experience adversity at young ages. Conclusion / ImplicationsWhile complete eradication of childhood adversity is unlikely, the continued development of programs and policies that build resilience in children is crucial. The results of this study will provide a means for population-level evaluation of such programs and policies, ultimately improving evidence for policy and decision-makers in the area of child development.

IntroductionRecent advances in Australia mean that it is possible to link national clinical registries with government held administrative datasets. However, formal evaluations of such activities and the potential impact for research are lacking. Objectives and ApproachWe aimed to assess the feasibility and accuracy of linking registrants from the Australian Stroke Clinical Registry (AuSCR) with the Medicare enrolment file. Following data custodian and ethics approvals, personal linkage identifiers were submitted to the Australian Institute of Health and Welfare (AIHW). De-identified data from AuSCR and the AIHW were submitted into the Secure Unified Research Environment and merged using project specific person-based IDs. We calculated the proportion of patients linked with the Medicare enrolment file that were present in the associated Medicare and medication dispensing datasets and the proportion with claims after their date of death. Logistic regression was used to identify factors associated with a non-merged patient. Results17,980 AuSCR registrants (January 2010-July 2014) were submitted for linkage (median age 76 years; 46% female; 67% ischaemic stroke; 16% TIA). Of these, 93% were merged with Medicare (N=16,648) and 95% with subsidised medication dispensing claims data (N=17,079). In those who died, 127 (0.8%) had one or more Medicare claim and 411 (2.4%) had one or more medications dispensed after their death date. Asian born registrants were less likely to be merged with Medicare (adjusted Odds Ratio [aOR]: 0.54; 95% Confidence Interval [CI]: 0.40, 0.72) than Australian born registrants. Those aged ≥85 years were less likely to be merged with Medicare data than those aged

IntroductionThe Census is the largest statistical collection undertaken by the Australian Bureau of Statistics (ABS), with its data critical to informing the planning and delivery of Government and community services. While the Census measure of income supports a wide range of analysis, demand exists for additional income topics to complement and extend the range of socio economic analysis that can be undertaken. The ABS has recently developed three experimental income topics for the 2016 Census using linked administrative data: main source of income; main source of government payments; and previous financial year income. Objectives and ApproachThis research utilised administrative data integrated by the ABS for the Multi-Agency Data Integration Project (MADIP). Taxation data from the Australian Taxation Office (ATO) and social security data from the Department of Social Services (DSS) were used in conjunction with the 2016 Census data to derive the additional topics. ResultsOverall, the three measures compare relatively closely to similar measures from the ABS Survey of Income and Housing (SIH). The ‘Main source of income’ and ‘Main source of government payments’ measures exhibit similar distributions to those from the SIH. The ‘Previous financial year income’ measures compare more closely with Census and SIH at the higher end of the income distribution, with some differences apparent at the lower end of the income distribution. Conclusion / ImplicationsThis work demonstrates the potential to supplement and enhance existing Census topics with linked administrative data. Further research, development and consultation with data users and the Australian community is needed.

IntroductionApplications in domains ranging from healthcare to national security increasingly require records about individuals in sensitive databases to be linked in privacy-preserving ways. Missing values make the linkage process challenging because they can affect the encoding of attribute values. No study has systematically investigated how missing values affect the outcomes of different encoding techniques used in privacy-preserving linkage applications. Objectives and ApproachBinary encodings, such as Bloom filters, are popular for linking sensitive databases. They are now employed in real-world linkage applications. However, existing encoding techniques assume the quasi-identifying attributes used for encoding to be complete. Missing values can lead to incomplete encodings which can result in decreased or increased similarities and therefore to false non-matches or false matches. In this study we empirically evaluate three binary encoding techniques using real voter databases, where pairs of records that correspond to the same voter (with name or address changes) resulted in files of 100,000 and 500,000 records containing from 0% to 50% missing values. ResultsWe encoded between two and four of the attributes first and last name, street, and city into three record-level binary encodings: Cryptographic long-term key (CLK) [Schnell et al. 2009], record-level Bloom filter (RBF) [Durham et al. 2014], and tabulation Min-hashing (TBH) [Smith 2017]. Experiments showed a 10% to 25% drop on average in both precision and recall for all encoding techniques when missing values are increasing. CLK resulted in the highest decrease in precision, while TBH resulted in the highest decrease in recall compared to the other encoding techniques. ConclusionBinary encodings such as Bloom filters are now used in practical applications for linking sensitive databases. Our evaluation shows that such encoding techniques can result in lower linkage quality if there are missing values in quasi-identifying attributes. This highlights the need for novel encoding techniques that can overcome the challenge of missing values.

IntroductionDuring 2019, the Western Australian (WA) government and Curtin University’s Centre for Data Linkage (CDL) created a large, de-identified researchable database – the Social Investment Data Resource (SIDR) – to support government in delivering targeted early interventions to young offenders and their families to reduce the likelihood of re-offending (the Target 120 program). Objectives and ApproachSIDR brings together administrative data from health, education, justice, child protection, disability and housing sectors. The linked, de-identified data provides an invaluable resource for actuarial assessment and social investment analytics to assess long-term costs and benefits of the Target 120 program. SIDR also provides an invaluable tool for academic research. SIDR adopted a distributed linkage model where linkage workload was shared between the Department of Health Data Linkage Branch who create and maintain the WA Data Linkage System (WADLS) and the CDL. Design elements of the model included a common spine (embedded into the infrastructure of both groups), methods for leveraging quality from WADLS, and inclusion of family relationships data from the WA Family Connections database. The linkage model within SIDR uses a combination of traditional and privacy-preserving record linkage (PPRL) methods. PPRL does not require release of personal identifiers; instead, data is irreversibly hashed prior to release for probabilistic linkage. ResultsThrough cooperation (distributed linkage) and innovation (a mix of traditional and PPRL linkage), the project has delivered a large, linked, cross-sectoral data resource for policymakers and researchers. Sharing of the linkage workload maximised the capacity and unique capabilities of each linkage unit. PPRL enabled ‘hard to get’ datasets from justice to be included. SIDR is being updated in 2020. Conclusion / ImplicationsSIDR provides a resource for whole-of-government policy development, service evaluation, academic research and social investment analytics for T120 and beyond. The SIDR linkage model has potential for adaptation and use elsewhere.

IntroductionIn their 2019 homelessness strategy, Welsh Government make the commitment for a Wales where homelessness is a rare experience, and if it does occur is brief and non-recurrent. However, measuring repeated homelessness/housing issues (or ‘non-recurrence’) has thus far not been possible in Wales, with only aggregate data collected nationally on interactions between people and local government housing-services. Drawing on recently available individual level data from a single local government area in Wales, we set out to explore repeated use of housing/homelessness services. Objectives and ApproachThe sample used in this analysis relates to approximately 6,000 individuals who sought assistance from Swansea local government housing-service during a four-year period, from January 2012 to December 2015. In order to group people with similar epeated interactions with the housing service, we use analyses that firstly aggregate the total time spent interacting with the service and over how many distinct periods (Cluster analysis), and secondly explore patterns in sequences of month-by-month service use histories (Sequence Analysis). ResultsWe describe the patterns of repeat housing-service use identified through the Cluster Analysis and the Sequence Analysis, and the characteristics of each group. We also directly compare the ‘time-aggregating’ and ‘time-patterned’ based approaches to classifying people, in order to illustrate the differences that methodology can bring to our conceptualisation of repeat housing service use. Conclusion / ImplicationsThe findings from this study have the potential to inform Wales’ move away from aggregate to individual level national homelessness data collection, by illustrating how individual level data can be used to provide new insights into homelessness. Exploring different conceptualisations of repeated housing-service use can also inform future developments in the measurement of repeat housing problems, and therefore monitor progress in achieving a Wales where homelessness is rare, brief, and non-recurrent.

IntroductionPatient comobidity at time of primary joint replacement (JR) impacts on outcomes including revision and mortality. Understanding changes in comorbidity profiles is important when assessing change in outcomes over time. Most arthroplasty registries have limited comorbidity information due to their minimum dataset. One approach to obtaining additional comorbidity data is linking registry data with national administrative data. Objectives and ApproachObjectives were to quantify pre-operative comorbidity profile of patients undergoing primary total hip replacement (THR) and total knee replacement (TKR) for osteoarthritis. Also, to examine temporal trends in individual comorbidities for THR and TKR patients. National pharmaceutical dispensing data were linked with THR and TKR arthroplasty patients. Medication dispensing histories in 12-months preceding JR (2003-2017) for 237,333 THR and 394,965 TKR patients, were mapped to 47 comorbidity classes using the Rx-Risk-V measure - a pharmacy-based measure of comorbidity. Comorbidity scores were calculated by summing comorbidity categories for individual patients. Trends in comorbidity scores/categories were described, with comorbidity information presented by PBS beneficiary category (concessional/general), stratified by age (

IntroductionThe Applied Health Research Question (AHRQ) portfolio is an initiative funded by the Ontario Ministry of Health. Knowledge users submit AHRQ requests to use administrative health data to inform their planning, policy and program development. Objectives and ApproachA request to estimate prevalence and incidence of blood-borne viral infections (BBVI) among individuals seeking fertility treatment services was approved by the ICES AHRQ team. To determine whether current BBVI testing guidelines are effective for timely identification of BBVI among individuals undergoing fertility treatment, this AHRQ sought to estimate: a) the testing prevalence for BBVI (using viral hepatitis as a proxy for all BBVI including HIV), and b) BBVI incidence over a five-year follow-up period. We used infertility codes billed by gynecologists and urologists to identify individuals seeking fertility treatment between April 1, 2002 and March 31, 2013. We looked forward five years from a first BBVI test billed by a physician and assessed annual and cumulative testing prevalence and incidence of viral hepatitis and HIV using rule-based administrative algorithms. ResultsIn total, 184,328 individuals (61.6% female) were included. The five-year BBVI testing prevalence was 45.5% and the median number of tests was 2 (interquartile range: 2-3). An estimated 949 new cases (48% diagnosed within 12 months of first BBVI test) of viral hepatitis were identified in the five-year follow-up with a cumulative incidence of 522 (95%CI 489-556) per 100,000 population. There were 57 new HIV cases (68% diagnosed within 12 months of first BBVI test) with a cumulative incidence of 31.4 (95%CI 23.7-40.6) per 100,000. The median times from first BBVI test to hepatitis and HIV diagnoses were 13.0 months (IQR 2.33-32.2) and 3.0 months (IQR 1.34-20.0), respectively. Conclusion / ImplicationsThese findings can be used to develop evidence-based BBVI testing guidelines for individuals seeking fertility treatment services.

IntroductionAn estimated 23% of children in England are exposed to maternal mental health problems, with growing concern around how maternal mental health affects parenting capacity and entries into public care for their children. There is limited evidence on the prevalence of mental health service need among women whose children enter care and on whether these needs are sufficiently met. With England’s family justice system and children’s social care sector increasingly overstretched, there is a need to identify opportunities for an improved response to maternal mental health. Objectives and ApproachWe linked administrative family court data for women (n = 5463) involved in public family court proceedings (care proceedings) across eight London boroughs between 2007 and 2019, with patient records from South London and Maudsley NHS mental health trust (SLaM). Work is ongoing to (1) describe the frequency, type and timing of mental health service use among women who link and (2) compare these findings to general population trends. Results2843 (52%) women linked to a SLaM patient record via deterministic linkage. Matches had better recording of identifiers and were more likely to have returned to court than non-matches (36% vs 26%). The majority of women (75.0%) were referred for treatment before the start of their (first) care proceeding, while 34.8% had a diagnosis recorded in a structured field. To appraise linkage quality, we manually reviewed de-identified clinician notes for 100 randomly selected women who linked; we confirmed ‘true’ match status for 95 women and found no evidence contradicting match status for the remainder. Conclusion / ImplicationsThis novel linkage between mental health and family court data supports growing evidence that women whose children enter care in England have high prevalence of mental health problems. Research using this linkage will help to inform strategies to improve maternal mental health and outcomes for vulnerable women and their children.

IntroductionEvidence suggests that many women whose children enter care via family court proceedings (care proceedings) in England experience mental health and substance misuse problems. Further, an estimated 22% of women will return to court within 5 years for subsequent proceedings, often returning quickly with a new infant and with little time to address underlying health-related problems. With growing pressures on England’s family justice system and the wider children’s social care sector, there is a need to identify opportunities to engage with women in a healthcare setting to reduce the likelihood of (recurrent) care proceedings. Objectives and ApproachWe used linked family court and mental health service data (n = 2843) for mothers who were referred to mental health services and were involved in care proceedings in South London between 2007 and 2019. We (1) describe differences in the frequency, type and timing of mental health and addictions service use between mothers who return to court and those who do not and (2) evaluate risk/protective factors for returning to court. ResultsAfter resolving duplicate matches (n = 2808), 1028 women (36.6%) referred to mental health services over the study period subsequently returned to court. We will describe differences in characteristics of service use and care proceedings between women who return to court and those that do not. We will also explore a range of potential risk/protective factors for returning to court, including maternal age at first child, indicators for failure to engage with services, mental disorder type, substance misuse, and maternal neighbourhood deprivation status. Conclusion / ImplicationsBetter understanding of how women use mental health and addictions services before, during and after proceedings, and identification of key risk/protective factors for returning to court, would inform strategies to reduce recurrent involvement in care proceedings among this population group.

IntroductionLinking the existing longitudinal data assets with administrative datasets provide the opportunity to transform longitudinal data into valuable assets to inform research and policy development. Objectives and ApproachThis paper will focus on Growing Up in Australia: The Longitudinal Study of Australian Children (LSAC) data linkage landscape and consents which are invaluable for the development of evidence-based health-social-economic policies. ResultsLSAC is Australia’s first nationally-representative longitudinal study of child development. Since 2004, two cohorts of 5,000 children and their parents have been interviewed every two years (B (baby) cohort and K (kindergarten) cohort). Over the years, multiple data linkage has been undertaken based on either parental consent or study child consent. In 2004 parents were asked to consent on behalf of the study child to link Medicare Benefits Schedule (MBS), Pharmaceutical Benefits Scheme (PBS)/Repatriation Pharmaceutical Benefits Scheme (RPBS) and Australian Childhood Immunisation Register (ACIR) administrative data to LSAC. The consent rate was 93% was for MBS, PBS and ACIR. Nearly 90% of B cohort parents provided consent to link Australian Early Development Census (AEDC)/National Assessment Program – Literacy and Numeracy (NAPLAN) and 95.4% of K cohort parents provided consent to link NAPLAN.Then ten years later, children in the K cohort were asked to consent to MBS/PBS and income-support administrative data. The rates were 86.6% for MBS, 85.4% for PBS and 81.2% for income-support administrative data. Parental consent to link their MBS, PBS and income-support administrative data was also sought and these rates varied between 60% to 88%. Conclusion/ImplicationsThe discussion will focus on differences in consent rates by time of consent, consenting individual and type of administrative data to be linked. Challenges and considerations that researches should be aware of when designing the linkage consent methodology will also be discussed.

IntroductionPostsurgical sepsis is a common complication, accounting for one-third of all sepsis cases and associated with higher risk of mortality. Objectives and ApproachThe Bureau of Health Information (BHI) in New South Wales (NSW), Australia produces independent reports and information about the performance of the healthcare system. Our report investigates the rate of sepsis within 30-days following surgery, and explores variation in risk standardised postoperative sepsis across NSW public hospitals. The study cohort and outcome definitions were drawn from existing international patient safety indicators, modified for appropriate use with Australian linked hospital and mortality data to identify patients transferred, or readmitted to any NSW public or private hospital with postoperative sepsis. Fine and Gray competing risks regression models were used to calculate risk-adjusted rates, and funnel plots were used to identify outliers. ResultsRate of 30-day postoperative sepsis was 0.9% among 66,143 adult elective surgeries that met the inclusion criteria during January 2015 to December 2017. Results showed around 33% of post-operative sepsis were identified as occurring during either a transfer or readmission to hospital. Hospital-level unadjusted rates of postsurgical sepsis ranged from 0% to 2% across more than 80 NSW public hospitals. The majority of hospitals (94%) had postoperative sepsis results lower or no different than expected. Conclusion/ImplicationsWe explored postoperative sepsis to help understanding of sepsis risk during the follow up period. Results showed a considerable number of postoperative sepsis cases would not have been identified without the use of linked data. Results on hospital risk standardised 30-day postoperative sepsis may prove useful as a screening tool for quality improvement in this area.

IntroductionInfection is a major complication following joint replacement (JR) surgery. However, little data exist on baseline use of antibiotics following primary JR and how use changes with subsequent revision surgery. Objectives & ApproachOur study objectives were to describe community use of antibiotics before and after primary total hip replacement (THR) and change in use pre and post revision procedure. Registry data were linked with national medication dispensing data using probabilistic record linkage. Patients with THR for osteoarthritis in a private hospital between 1999 and 2017 were included. Three groups were analysed: patients with primary procedures revised for infection, revised for non-infection reasons and those not revised. Rate of antibiotic dispensing/month was calculated as number of patients dispensed at least one antibiotic in a given month divided by number of patients at-risk. ResultsThere were 102,577 patients included in the non-revised group, 3,156 revised for non-infection and 520 revised for infection. Prior to primary THR, baseline antibiotic dispensing rate was 9-11%/month in all groups. Post-primary rates were similar (10-11%) for non-revised and revised non-infection patients but higher (16-17%) for revised-infection patients. In 1, 6 and 12 months preceding revision for infection, antibiotic use was 55%, 27% and 22%, respectively. For patients revised for non-infection, antibiotic use was 21%, 14%, 13%, respectively. One-month following revision for infection, 82% of patients were dispensed antibiotics, remaining high (38%) at 6-months and 28% at 12-months. In the revision non-infection group, antibiotic use was 48% first month post-surgery, reducing rapidly to 15% at 6-months. Conclusion / ImplicationsNon-revision and revision non-infection patients had similar antibiotic dispensing before and after surgery. Revision infection patients however, maintained higher antibiotic dispensing post-primary, pre and post revision. This may reflect either ongoing infection, need for long-term suppressive therapy or reluctance of treating physicians to terminate treatment.

IntroductionThe risk of cardiovascular events amongst people with epilepsy who are receiving enzyme-inducing anti-epileptic drugs (EIAEDs) seems to be higher than those on other medications and the general population. National-level record linkage enables development of case-control studies at a wider scope accounting for multiple factors. Objectives and ApproachPeople with epilepsy were identified between 2003-01-01 and 2017-12-31 and were matched to a control group on: age, gender, deprivation quintile and year of diagnosis, accounting for any changes in clinical therapeutic guidelines. Primary and secondary care population records were linked to capture relevant comorbidities and major cardiovascular events. Annual district birth and death extract were used in combination with the Welsh Demographic Service (WDS) dataset to capture demographic and cardiovascular related death records. The WDS dataset was used to identify eligible control groups for each case and a linkage approach between the control and case database was developed for matching cases and controls with replacement and randomization. Survival analysis was conducted to evaluate the difference in time to first major cardiovascular event in patients receiving EIAED versus Non-EIAEDs and controls. Results10,241 cases (mean age 49.6 years, 52.2% male) with diagnosis of epilepsy were matched to 35,145 controls. 3,180 (31.1%) cases received EIAEDs and 7,061 (68.9%) received non-EIAEDs. The risk of experiencing a major cardiovascular event was higher in cases compared to controls (adjusted hazard ratio 1.52,95%CI[1.50–1.55];p

IntroductionStats NZ provides the world-leading Integrated Data Infrastructure (IDI) to help improve New Zealanders’ lives. Ngā Tikanga Paihere (the Tikanga Framework) is a tool designed to ensure data use is carefully considered, and practice occurs in good faith. It also aims to build and maintain public trust and confidence in the way Stats NZ manages administrative data in the IDI. It draws on general customary concepts from Te Ao Māori (Māori world) and sits in alignment with the current model of the Five Safes Framework for access. Objectives and ApproachTikanga are appropriate customary practices or ‘layers of the culture’ developed by Māori communities and individuals and informed by common cultural values and concepts. Stats NZ applies the framework when reviewing applications to use data in the IDI and Longitudinal Business Database (LBD). The framework is used with the original ‘Five Safes’ framework in the review process to ensure that data will be used in an appropriate and collaborative way, without marginalising any specific populations. In research proposals relating specifically to Māori, researchers are asked to demonstrate what value their work will bring to Māori communities and how the researchers will work collaboratively with those communities. ResultsUse of Ngā Tikanga Paihere has been well received. Stats NZ is now looking to broaden the use of Ngā Tikanga Paihere to assess all integrated data research proposals that focus on minority, identifiable, and marginalised populations. Conclusion / ImplicationsWhile Ngā Tikanga Paihere does not resolve data governance or cultural licence issues, it encourages researchers to actively engage and collaborate with Māori groups when these communities are potentially impacted, or when the project might use data about these communities. Stats NZ sees data governance and cultural licence matters as an ongoing conversation between agencies and Treaty/Te Tiriti partners. Ngā Tikanga Paihere should be a tool to help guide the government data system to work with Māori data ‘in good faith’.

IntroductionThe WA Child Development Atlas (CDA) is a unique resource that generates geographic profiles of the development, health and well-being of WA children. The CDA has been developed at the Telethon Kids Institute with funding from the Ian Potter and Minderoo Foundations and in partnership with government agencies and service providers. Objectives and ApproachVisualisation of data within geographic boundaries helps to build a picture of ‘place’. This informs better decision making and underpins the development of evidence-based policy and service planning that acknowledges the differing needs of communities. The CDA is an online, interactive mapping tool that utilises geographic information system (GIS) technologies to identify spatial patterns in population-level child and youth development indicators. Data are sourced from core health, social, and linked administrative datasets, from 1990 onwards. Aggregated, de-identified data on children and young people (0-24 years) and their parents are mapped within geographic boundaries across WA. The CDA was piloted to ensure it is useful and relevant to a diversity of stakeholders. Extensive community consultation has been undertaken at all stages of the project. The CDA has been received favourably by consumers, government organisations, communities and researchers, with significant support for public launch. ResultsThe CDA is a freely available, confidential and secure resource. Features of the CDA include: it is online and interactive (no software is required), area profiles, downloadable charts, tables and summary statistics, map panning and zooming. The Atlas has been used to generate community profiles and empowers all the people of WA with enhanced knowledge and understanding of the development of children. Conclusion / ImplicationsThe CDA has built capacity to use spatial information, assisting agencies who make decisions relevant to child development and to facilitate cross-agency collaboration. It is an important and valuable resource utilising disparate data sources with significant public benefit.

IntroductionEpidemiological studies of twin pairs provide researchers with the opportunity to better understand the roles of genetics and the environment on human traits and health conditions. Twin births are also of interest for public health, given they are five times more likely to be of low birth weight and preterm compared to singletons. Male twin newborns are at high risk of mortality, although the causes of such disadvantage are still largely unknown. Objectives and ApproachWe deterministically linked population birth to death records of twins born in Brazil from 2012–2016, and probabilistically linked twins within pairs through a bespoke algorithm. We studied male-female twin pairs to investigate sex differences in infant mortality stratified by early neonatal (0–6 days), late neonatal (7–27 days) and late infant (28–365 days) deaths. We used conditional logistic regression to estimate odds ratios (OR) for male sex, adjusting for birth weight and matching for familial factors by design. ResultsOur algorithm successfully matched 101,382 twin pairs, 28,558 were male-female pairs included in the study. Average birthweight was 100g greater for males compared with females. We found that males were at higher risk of infant mortality than their female co-twins even after adjusting for birth weight and familial factors (aOR = 1.42, 95% CI: 1.22–1.66). The aORs for neonatal death increased from those born at less than 28 weeks to those born at 28–31, 32–26 and 37+ weeks: 1.47 (1.02–2.13), 1.73 (1.17–2.57), 1.99 (1.17–3.38), and 3.35 (1.29–8.73), respectively. Conclusion / ImplicationsMale twins have greater risk of infant and neonatal mortality compared with female co-twins, more so the higher the gestational age. Unmeasured familial and maternal factors may influence the role of birth weight in the association between sex and poor early life outcomes.

IntroductionRepresenting patient-registered addresses as pseudonymised Unique Property Reference Numbers (UPRNs) enables linkage of environmental and household information to electronic health records (EHRs). However, the accuracy and potential biases in address-matching algorithm results applied to patient addresses is unknown. Objectives and ApproachTo investigate accuracy, match rate, and biases in assigning UPRNs to general practitioner (GP)-registered patient addresses for a geographically-defined UK population, using a bespoke deterministic address-matching algorithm comprising 213 rules applied in rank order of minimising false-positives, developed for the Discovery Data Service. We ran this algorithm to match 906,220 adult patient GP-registered addresses (48% female, 47% non-White, 89% 20-64) sampled in mid-2018 from 159 GP practices in four London boroughs to Ordnance Survey’s AddressBase Premium database. We evaluated the error rates using a gold-standard dataset. We used binary logistic regression to estimate the likelihood (Odds Ratio [OR]; 95% Confidence Intervals [CI]) of no UPRN match according to and adjusting for patient age, sex, ethnic background, deprivation, residential mobility and multiple GP registrations. Results96% of patient addresses were successfully assigned a UPRN. Algorithm sensitivity, specificity, positive and negative predictive-values and F-measure were, respectively: 0.993, 0.019, 0.914, 0.204, and 0.9516. After mutual adjustment, UPRN assignment was less likely for: men (OR: 0.87; 95%CI: 0.83,0.91); adolescents and the elderly (15-19 years: 0.57;0.43,0.77; ≥90 years: 0.39;0.18,0.84); those from Chinese ethnic backgrounds (0.87;0.8,0.91), living in the least deprived areas (0.25;0.21,0.31), or with two or more distinct UPRNs across multiple registrations (0.37;0.28,0.49); and more likely for: those from Bangladeshi ethnic backgrounds (1.79;1.61,2.00), registered before 2018 (5.10;4.42,5.87), or with multiple GP registrations (2.36;1.82,3.05). Conclusion / ImplicationsThe Discovery open-source algorithm achieves a high accurate match rate and quantifies the demographic groups that may be under-represented among those successfully matched. This is the first time that bias in matching rates for an address-matching algorithm has been evaluated using patient-registered addresses.

IntroductionIschaemic Heart Disease (IHD) is a leading cause of death in Western countries. Common occupational exposures such as loud noise, long working hours, and sedentary work have been associated with increased IHD risks, but inconsistently. Objectives and ApproachThis study examines associations between incident IHD and exposure to long working hours, sedentary work, and loud noise. Individual-level microdata from Statistics New Zealand Integrated Data Infrastructure (IDI) were extracted for adults (age 20-64 years) with occupation recorded on the 2013 Census. The number of working hours was extracted from the Census, and exposure to sedentary work and loud noise was assessed through job exposure matrices (JEMs). IHD events (from 2013 to end of 2018) were identified using hospitalisations, prescriptions and deaths. Hazard ratios (HRs) were calculated using cox regression adjusted for age, socioeconomic status, and smoking. Results were stratified by sex and ethnicity. ResultsA total of 20,610 IHD cases were identified from 1,594,680 individuals employed at time of Census. Both short (90% of the time compared to 90dBA) compared to no exposure (

IntroductionData from population health surveys, administrative health records and environmental monitoring are increasingly being linked at the individual level. As these data become available to health researchers, there is an increasing need for methods which can make sense of large, noisy and heterogeneous data and can model complex relationships. Using these data, machine learning methods have the potential to produce population health risk algorithms with better performance than those developed with traditional statistical approaches. Objectives and ApproachThe objective of this work is to explore the use of machine learning methods for the development, validation and implementation of predictive risk algorithms designed specifically for population health planning purposes. Algorithms to predict risk of dementia and avoidable hospitalizations are in development using the Canadian Community Health Survey, geographic sociodemographic information, administrative health care utilization data and vital statistics. Methods being explored include naïve Bayes, gradient boosting, support vector machines and neural networks. ResultsRisk algorithms for population health should generally prioritize calibration over discrimination due to implications for resource allocation decisions. Approaches to minimize the risk of overfitting should be used and reweighting of unbalanced data avoided as it distorts the population-level nature of the data. It is important to be aware of propagating underlying bias in the data or exacerbating existing health inequities, which can be evaluated in part through assessment of calibration across relevant population subgroups. Approaches that consider multi-level data structures are needed to appropriately incorporate neighbourhood-level measures with individual-level information. To maximize population health impact and acceptability, model transparency and interpretability should be prioritized. ConclusionThere is tremendous potential for machine learning approaches to leverage large volumes of linked population data to produce predictive risk algorithms that will inform population health decision-making. Future work will explore use of complex environmental remote sensing and built environment data.

IntroductionThe current data access model in BC involves project-specific applications and data provisioning. The timeline from application to provisioning is 6-8 months. Novel initiatives including Program of Research (POR), Core Data Sets (CORE), and Data Reuse are being explored and evaluated. Objectives and ApproachWe aim to develop data provisioning models that improve efficiency and access timelines by reducing process duplication and adopting open and flexible approaches to data use while ensuring data privacy. ResultsPOR allows researchers to access broad programmatic data that fulfills data requirements for multiple thematically-linked projects. While we provision the program data, a research team data manager extracts the project-specific data from the program dataset. A pilot program with two active projects is ongoing. The timeline from application to program data provisioning was 8 months. Project data was delivered in 2-3 months. CORE is a transformative data provisioning model that allows researchers to access entire data sets that contain a group of pre-approved and non-sensitive data variables for the BC population for all available years. This decreases the possibility of variable omission which is prevalent under the existing process. Additionally, this model allows researchers the flexibility to identify their cohort using their preferred methodology. Data Reuse allows re-use of data between similar projects conducted by the same investigator. Projects were surveyed for similar objectives, investigators and data requirements. Similar projects were grouped and analyzed to evaluate pre-implementation timelines. Application to provisioning timeline for one group of six projects ranged from 7-18 months. Post-implementation timelines will be evaluated once Data Reuse is implemented. Conclusion / ImplicationsThese new initiatives have shown promising results in access efficiency and data privacy in the pilot phase. Continuous process and privacy evaluations are involved and ongoing collaborations with the data providers and researchers are required prior to full implementation.

IntroductionDiagnostic codes, such as the ICD-10, may be considered as sensitive information. If such codes have to be encoded using current methods for data linkage, all hierarchical information given by the code positions will be lost. We present a technique (HPBFs) for preserving the hierarchical information of the codes while protecting privacy. The new method modifies a widely used Privacy-preserving Record Linkage (PPRL) technique based on Bloom filters for the use with hierarchical codes. Objectives and ApproachAssessing the similarities of hierarchical codes requires considering the code positions of two codes in a given diagnostic hierarchy. The hierarchical similarities of the original diagnostic code pairs should correspond closely to the similarity of the encoded pairs of the same code. Furthermore, to assess the hierarchy-preserving properties of an encoding, the impact on similarity measures from differing code positions at all levels of the code hierarchy can be evaluated. A full match of codes should yield a higher similarity than partial matches. Finally, the new method is tested against ad-hoc solutions as an addition to a standard PPRL setup. This is done using real-world mortality data with a known link status of two databases. ResultsIn all applications for encoded ICD codes where either categorical discrimination, relational similarity or linkage quality in a PPRL setting is required, HPBFs outperform other known methods. Lower mean differences and smaller confidence intervals between clear-text codes and encrypted code pairs were observed, indicating better preservation of hierarchical similarities. Finally, using these techniques allows for much better hierarchical discrimination for partial matches. ConclusionThe new technique yields better linkage results than all other known methods to encrypt hierarchical codes. In all tests, comparing categorical discrimination, relational similarity and PPRL linkage quality, HPBFs outperformed methods currently used.

Introduction and PurposeChronic Disease Management (CDM) plans are used by general practitioners to manage chronic diseases such as stroke. However, there is limited evidence that being on these plans improve adherence to secondary prevention medications after stroke. We aimed to assess the association of the duration on a CDM plan in improving adherence to secondary prevention medications following stroke. MethodsAustralian survivors of stroke or transient ischaemic attack were participants from the STAND FIRM trial. Patients were individually linked with claims for CDM plans from Medicare and dispensings of secondary prevention medications from the Pharmaceutical Benefits Scheme. We estimated (1) duration on a CDM plan based on the timing and Medicare items claimed and (2) the proportion of days that patients would have been covered by these medications (PDC), while accounting for deaths and instances of over-supply. Dosage for each quantity of medication was determined by the regularity in which patients returned for a refill. Logistic regression was used to evaluate factors associated with ≥80% adherence, up to 3 years after stroke, for each of antihypertensive, antithrombotic and lipid-modifying drugs. ResultsThe median PDC for 563 patients (median age 70yrs; 36% female) ranged from 92% to 95% among the three classes of medications. Approximately 27% did not take up a CDM plan, 33% were on plans for

IntroductionDonor suitability assessment is often time-sensitive, with imperfect information available at referral. Opportunities for donation may be missed if donors are rejected based on inaccurate information. Objectives and ApproachWe sought to compare assessment of cancer transmission risk based on information available at referral (perceived) versus subsequently obtained (verified) detailed medical history, and to identify any missed opportunities for donation. We used data from the NSW Biovigilance Register, a cohort study of deceased organ donor referrals in New South Wales (NSW) 2010-2015. Referrals from NSW who had consent and were otherwise suitable for donation (except for perceived or verified tumour) were analysed. Perceived tumour details were obtained from NSW Organ and Tissue Donation Service referral logs, while verified details were based on the Central Cancer Registry and Admitted Patient Data Collection. Cancer transmission risk was graded using Transplantation Society of Australia and New Zealand guidelines. ResultsAmong 772 referrals otherwise suitable for donation, 601 (78%) had accurately classified cancer transmission risk (κ=0.59). Overall, 53 suitable referrals were rejected due to cancer transmission risk (missed opportunities), including 49 (92%) with overestimated risk (perceived high, verified low/none), and 4 (8%) due to excessive risk aversion (perceived and verified low risk). Missed opportunities had overestimated transmission risk for tumours including leukaemia (8, 16%), lung (8, 16%), melanoma (7, 14%), brain (6, 12%), and breast (6, 12%). Conclusion / ImplicationsDespite time pressure and limited information availability, there is moderate agreement between perceived and verified cancer transmission risk. Nevertheless, improved information availability (e.g. via data linkage) could meaningfully increase the number of actual donors.

IntroductionDue to the ever-growing volume and complexity of clinical data, it has become a tedious task to extract information from data for secondary uses such as decision support, quality assurance, and outcome analysis. Recently, there have been great advances in Natural Language Processing (NLP) approaches that automate knowledge extraction from clinical reports in order to save costs and improve efficiency. Objectives/Approach​Our goal is the development of an NLP tool designed to automatically extract and encode clinical information from laboratory reports. This study describes and evaluates our NLP tool on provincial repositories of laboratory tests and results called Ontario Laboratory Information System (OLIS). OLIS is an electronic system that covers >200 labs and stores patients’ current and past test results as patients move through different areas of the healthcare system. Our NLP tool is a modular system of pipelined components including Named Entity Recognition module for extracting mentions of virus and test mentions and inference to combine extracted entities into a meaningful outcome. Results​Initial analyses were conducted on a segment of OLIS related to laboratory tests for respiratory viruses. This data included over a million observations corresponding to ~100 Logical Observation Identifiers Names and Codes (LOINC), with >40,000 unique strings. The clinical text was cleaned, tokenized, and parsed using an in-house text algorithm that was continually refined with manual review from clinical experts. This data was then encoded as virus and test types to be used as a ground truth. The NLP tool was built on ground truth data and achieved an accuracy greater than 95%. Conclusion/ImplicationsApproaches like these can be applied to many areas of health research that make use of clinical reports. Our methods, when optimized and validated, can be deployed into clinical systems to provide on-the-spot analysis of various laboratory reports.

IntroductionResidential fires remain a significant global public health problem. It is recognised that the reported number of residential fires, fire-related injuries and deaths significantly underestimate the true number. Australian population-based surveys show that around two-thirds of respondents who experience a residential fire are unwilling to call fire services, and studies from the US and New Zealand highlight that many individuals who access medical treatment for fire-related injuries do not have an associated fire incident report. Objectives and ApproachThis population-based study aimed to quantify the total number of residential fires, fire-related injuries and associated health service utilisation. The cohort included all persons residing at a residential address in New South Wales, Australia, which experienced a fire between 1 January 2005 - 31 December 2014. The cohort comprised linked person-level data from eight administrative datasets and includes information about nature of fire, first responder use (Fire and Rescue (FRNSW) and ambulance services), health service utilisation (emergency department, hospital and burns outpatient clinic) and health outcomes. ResultsOver the study period, FRNSW responded to 42,491 residential-fire incidents, involving 42,160 individuals with some individuals reporting multiple times. In total, 3,382 individuals used one or more health service and 154 individuals died. Of individuals who contacted FRNSW, 1,661 (3.9%) used health services;ambulance (n=1,101), emergency department (n=1,114), hospital admissions (n=168). There were 95 deaths. There were 1,721 (51%) additional individuals who used one or more health service as a result of a residential-fire that did not contact FRNSW and 59 additional deaths were identified. Conclusion / ImplicationsThis study found that more than half of individuals who used health services for residential fire-related injuries did not have an associated fire report, highlighting the importance of data linkage for accurate communication to policy makers and the public on the prevalence and impact of residential-fires.