Scholars studying organizations often work with multiple datasets lacking shared identifiers or covariates. In such situations, researchers usually use approximate string (“fuzzy”) matching methods to combine datasets. String matching, although useful, faces fundamental challenges. Even where two strings appear similar to humans, fuzzy matching often struggles because it fails to adapt to the informativeness of the character combinations. In response, a number of machine learning methods have been developed to refine string matching. Yet, the effectiveness of these methods is limited by the size and diversity of training data. This paper introduces data from a prominent employment networking site (LinkedIn) as a massive training corpus to address these limitations. By leveraging information from the LinkedIn corpus regarding organizational name-to-name links, we incorporate trillions of name pair examples into various methods to enhance existing matching benchmarks and performance by explicitly maximizing match probabilities. We also show how relationships between organization names can be modeled using a network representation of the LinkedIn data. In illustrative merging tasks involving lobbying firms, we document improvements when using the LinkedIn corpus in matching calibration and make all data and methods open source.

A single dataset is rarely sufficient to address a question of substantive interest. Instead, most applied data analysis combines data from multiple sources. Very rarely do two datasets contain the same identifiers with which to merge datasets; fields like name, address, and phone number may be entered incorrectly, missing, or in dissimilar formats. Combining multiple datasets absent a unique identifier that unambiguously connects entries is called the record linkage problem. While recent work has made great progress in the case where there are many possible fields on which to match, the much more uncertain case of only one identifying field remains unsolved: this fuzzy string matching problem, both its own problem and a component of standard record linkage problems, is our focus. We design and validate an algorithmic solution called Adaptive Fuzzy String Matching rooted in adaptive learning, and show that our tool identifies more matches, with higher precision, than existing solutions. Finally, we illustrate its validity and practical value through applications to matching organizations, places, and individuals.

There are research questions whose answers require record linkage of multiple databases that may be characterized by limited options for full data sharing. For this purpose, the Open Data Infrastructure for Social Science and Economic Innovations (ODISSEI) consortium has supported the development of the ODISSEI Secure Supercomputer (OSSC) platform that allows researchers to link cohort data to data from Statistics Netherlands and run large-scale analyses in a high-performance computing (HPC) environment. Here, we report a successful record linkage genomewide association (GWA) study on expenditure for total health, mental health, primary and hospital care, and medication. Record linkage for genotype data from 16,726 participants from the Netherlands Twin Register (NTR) with data from Statistics Netherlands was accomplished in the secure OSSC platform, followed by gene-based tests and estimation of total and single nucleotide polymorphism (SNP)-based heritability. The total heritability of expenditure ranged between 29.4% (SE 0.8) and 37.5% (SE 0.8), but GWA analyses did not identify SNPs or genes that were genomewide significantly associated with health care expenditure. SNP-based heritability was between 0.0% (SE 3.5) and 5.4% (SE 4.0) and was different from zero for mental health care and primary care expenditure. We conclude that successfully linking genotype data to administrative health care expenditure data from Statistics Netherlands is feasible and demonstrates a series of analyses on health care expenditure. The OSSC platform offers secure possibilities for analyzing linked data in large scale and realizing sample sizes required for GWA studies, providing invaluable opportunities to answer many new research questions.

The Nordic countries have comprehensive, population-based health and medical registries linkable on individually unique personal identity codes, enabling complete long-term follow-up. The aims of this study were to describe the NorTwinCan cohort established in 2010 and assess whether the cancer mortality and incidence rates among Nordic twins are similar to those in the general population. We analyzed approximately 260,000 same-sexed twins in the nationwide twin registers in Denmark, Finland, Norway and Sweden. Cancer incidence was determined using follow-up through the national cancer registries. We estimated standardized incidence (SIR) and mortality (SMR) ratios with 95% confidence intervals (CI) across country, age, period, follow-up time, sex and zygosity. More than 30,000 malignant neoplasms have occurred among the twins through 2010. Mortality rates among twins were slightly lower than in the general population (SMR 0.96; CI 95% [0.95, 0.97]), but this depends on information about zygosity. Twins have slightly lower cancer incidence rates than the general population, with SIRs of 0.97 (95% CI [0.96, 0.99]) in men and 0.96 (95% CI [0.94, 0.97]) in women. Testicular cancer occurs more often among male twins than singletons (SIR 1.15; 95% CI [1.02, 1.30]), while cancers of the kidney (SIR 0.82; 95% CI [0.76, 0.89]), lung (SIR 0.89; 95% CI [0.85, 0.92]) and colon (SIR 0.90; 95% CI [0.87, 0.94]) occur less often in twins than in the background population. Our findings indicate that the risk of cancer among twins is so similar to the general population that cancer risk factors and estimates of heritability derived from the Nordic twin registers are generalizable to the background populations.

Changes in respiratory pathogen testing can affect disease burden estimates. Using linked data, we describe changes in respiratory virus testing among children born in Western Australia in 1996–2012. We extracted data on respiratory specimens from these children from birth to age 9 years. We estimated testing rates by age, year, Aboriginal status and geographical location. Predictors of testing among children hospitalised at least once before their 10th birthday were identified using logistic regression. We compared detection methods for respiratory viruses from nasal/nasopharyngeal (NP) specimens by age and year. Of 83 199 virology testing records in 2000–2012, 80% were nasal/NP specimens. Infants aged <1 month had the highest testing rates. Testing rates in all children increased over the study period with considerable yearly fluctuations. Among hospitalised children, premature children <32 weeks gestation had over three times the odds of being tested (95% CI 3·47–4·13) than those born at term. Testing using molecular methods increased from 5% to 87% over the study period. Proportion of positive samples increased from 36·3% to 44·4% (P < 0·01); this change was greatest in children aged 2–9 years. These findings will assist in interpreting results from future epidemiology studies assessing the pathogen-specific burden of disease.

A population-based record linkage case cohort of 239,995 births, to 119,214 women, born in Western Australia from 1980 to 2001 inclusive, was used to measure the recording of selected indicators of maternal health (current and prior) during pregnancy. We compared records of women with singleton pregnancies with that in twin pregnancies Mothers of first- and second-born singletons (n = 117,647) were compared with women with a first-born singleton followed by twins (n = 1,567). Binary indicators were used to calculate population prevalence of medical conditions, pregnancy complications and birth outcomes. Infant outcomes included stillbirth, low birthweight, preterm birth and birth defects. Women with twins were significantly older and taller, with similar rates of medical conditions and pregnancy complications during first singleton pregnancies compared with women with two consecutive singletons. However, during their second pregnancy, women with twins had significantly higher rates of essential hypertension, pre-eclampsia, threatened abortion, premature rupture of the membranes and ante partum hemorrhage with abruption than women with singletons. For both groups, maternal conditions in the first pregnancy were underreported in the second pregnancy, including diabetes, epilepsy, asthma, chronic renal dysfunction and essential hypertension. At the second birth, twins were 3 times more likely to be stillborn, 17 times more likely to be low birthweight and 4 times more likely to be delivered preterm compared with singletons. This research demonstrates the importance for epidemiologists and others, of having access to a complete maternal medical history for analyses of risks associated with maternal, infant and childhood morbidity.

As in other ageing populations, dementia, musculoskeletal conditions and cardiovascular disease affect a high proportion of Australians aged over 65 years, and the prevalence of these conditions increases significantly with age. People with these conditions may need to access a range of care services over time to enable them to remain living in their homes. Many eventually need to move into a nursing home.

In contrast to the considerable recent literature on the funding of long-term care systems for population ageing, studies on the care pathways followed by individuals are much less common. This paper explores the effect of disease on use of community care services and nursing homes over time, focusing on people with dementia, cardiovascular disease and musculoskeletal conditions. Care-use transitions are identified using linked administrative client data for a cohort of 33,300 community-living Australians who had an aged care assessment in 2003-04 and who had not previously used aged care services.

The different symptoms and courses of diseases meant that the patterns of aged care service use, both in terms of care services accessed and the timing of this access, varied considerably for people with different health conditions. These differences persisted across a range of client characteristics. In particular, people with dementia or cerebrovascular disease as their main health condition were more likely to enter nursing home care than those with heart disease or musculoskeletal conditions.

The variation in use of aged care services according to disease group need to be taken into account in any projections of demand for aged care. Such projections must allow for predictions of disease prevalence, or else they will yield inaccurate predictions of demand for both community and residential care.

Routinely collected and reported indicators for health service utilization have traditionally been event/episode related and hospital centered. This is also the case for service utilization by persons with mental disorders, for whom national and international databases usually report rates of hospital discharges, mean length of stay for hospital episode and the like. Such event/episode-related indicators are of limited use for planning and improving services for persons with mental disorders. It is argued that new reporting systems are needed that allow the monitoring of the pathways of persons with mental disorders through the service system. It is shown how – owing to recent developments in techniques of ‘pseudonymization’ and the ever-increasing computer power for dealing with large volumes of patient data – such a system can be established and how it can contribute to analyzing empirically such mental health-care issues as ‘heavy utilizers’, ‘revolving door psychiatry’, ‘continuity of care’, ‘de-institutionalization’ and the like. Results of a record linkage study for the total population of a federal state of Austria monitoring both psychiatric and non-psychiatric health service utilization are reported. Some unexpected findings include the high utilization of non-psychiatric services by patients discharged from a psychiatric hospital bed, results which could not have been found by psychiatric case registers which usually only monitor utilization of psychiatric services.

To determine the extent age, sex and co-infection affect morbidity in people infected with hepatitis C virus (HCV), we performed a population-based study linking HCV notifications in New South Wales, Australia with their hospital (July 2000 to June 2006), hepatitis B virus (HBV) and HIV notification, and death records. Poisson models were used to calculate hospitalization rate ratios (RRs) for all-cause, illicit drug and liver-related admissions. Co-infection RRs were used to estimate attributable risk (AR). The 86 501 people notified with HCV contributed 422 761 person-years of observation; 0·8% had HIV, 3·7% HBV, and 0·04% had both. RRs for males were equal to or lower than for females in younger ages, but higher in older ages (P for interaction ⩽0·013). HBV/HIV co-infection resulted in ARs of over 70% for liver disease and 30–60% otherwise. However, at the cohort level the impact was minimal (population ARs 1·3–8·7%). Our findings highlight the importance and success of public health measures, such as needle and syringe exchange programmes, which have helped to minimize the prevalence of co-infection in Australia. The findings also suggest that the age of study participants needs to be considered whenever the burden of HCV-related morbidity is reported by sex. The results are likely to be representative of patterns in hospital-related morbidity for the entire HCV-infected population in Australia and the ARs generalizable to other developed countries.

The mechanisms that cause susceptibility to invasive meningococcal disease are largely unknown, but are likely to have important genetic and immunological components. We postulated that susceptibility to meningococcal disease might be associated with altered risks of development of other clinical disease. We studied cancer and immune-mediated disease in people who have been hospitalized with meningococcal disease. In cohorts of people who had invasive meningococcal disease, compared with reference cohorts, the rate ratio for cancer in an Oxford dataset studied from 1963 to 1998 was 0·88 [95% confidence interval (CI) 0·42–1·61] and in an all-England dataset studied from 1999 to 2005 it was 1·02 (95% CI 0·80–1·27). The respective rate ratios for immune-mediated disease were 1·49 (95% CI 0·81–2·50) and 0·69 (95% CI 0·53–0·89). Susceptibility to meningococcal disease was not associated with an altered risk of cancer. Occurrence of immune-mediated disease was, if anything, low in the large all-England cohort of people who had meningococcal disease.

Infection with Epstein–Barr virus (EBV) followed by infectious mononucleosis (IM) is now considered to be a risk factor for Hodgkin's disease (HD). It is less clear whether EBV infection and IM are associated with an increased risk of cancer generally. We used a longstanding record-linkage dataset in Oxford (years 1963–1998), and a more recent record-linkage dataset covering England (1999–2005), to compare rate ratios for cancer between people admitted to hospital for IM and a reference cohort. In the Oxford cohort, there was an increased risk of subsequent HD [rate ratio (RR) 6·0, 95% confidence interval (CI) 2·4–12·5] but not of other cancers combined (RR 0·85, 95% CI 0·57–1·23). In the England cohort, there were increased risks of HD (RR 3·2, 95% CI 1·2–7·0), non-Hodgkin's lymphoma (RR 5·6, 95% CI 2·9–9·8), and oropharyngeal cancer (RR 5·4, 95% CI 1·1–16·2), but no significant overall risk of cancer when lymphomas were excluded (RR 1·01, 95% CI 0·71–1·41). We confirm an association between IM and lymphoma; but the risk, if any, of cancer more generally is likely to be small.