As the populations ages rapidly, injury in older people is increasingly becoming a major health problem. This thesis examines the epidemiology of common injuries in older people and how data linkage can improve injury surveillance as well as knowledge about the circumstances and outcomes of injury in older people. These issues are explored using data from New South Wales, Australia and the emphasis is on injuries resulting from falls and traffic crashes as the most common mechanisms of injury in older people. The epidemiology and trends of hospitalisations as a result of hip, pelvic and wrist fractures are examined using NSW hospitalisation data during the 1990's. Internal data linkage of the 2000-2001 NSW hospitalisation data is used to eliminate double counting of hospital admissions for injurious falls in older people and to assess the validity and estimate the effects of previously used approaches on the incidence of hospitalised falls. Probabilistic data linkage of hospital and police crash records for the same year is also used to examine data quality in both collections and to explore the relationship between the circumstances and outcomes of injury in older drivers injured in a traffic crash. The findings of the epidemiological profile of hip, pelvic and wrist fractures in older people indicate that they are likely to continue to impose a considerable burden on acute health care services. The internal linkage of hospital data shows that data linkage techniques allow the identification of incident cases of hospitalised falls and point to the low validity of previously used approaches to estimate the incidence of these cases. Record linkage of hospital and police records demonstrates the limitations of using the datasets separately to examine the burden of traffic injuries in older people and shows the importance of environmental factors, complex road intersections in particular, in high injury severity in older drivers. The thesis also discusses some of the challenges of using record linkage for injury research and highlights the importance of including the date of injury and a unique personal identifier to improve the surveillance and reporting of injuries, including those in older people.

Yes / Congenital anomalies (CAs) are a common cause of infant death and disability. We linked children from a large birth cohort to a routine primary care database to detect CA diagnoses from birth to age 5 years. There could be evidence of underreporting by CA registries as they estimate that only 2% of CA registrations occur after age 1 year. Methods CA cases were identified by linking children from a prospective birth cohort to primary care records. CAs were classified according to the European Surveillance of CA guidelines. We calculated rates of CAs by using a bodily system group for children aged 0 to <5 years, together with risk ratios (RRs) with 95% CIs for maternal risk factors. Results Routinely collected primary care data increased the ascertainment of children with CAs from 432.9 per 10 000 live births under 1 year to 620.6 per 10 000 live births under 5 years. Consanguinity was a risk factor for Pakistani mothers (multivariable RR 1.87, 95% CI 1.46 to 2.83), and maternal age >34 years was a risk factor for mothers of other ethnicities (multivariable RR 2.19, 95% CI 1.36 to 3.54). Education was associated with a lower risk (multivariable RR 0.78, 95% CI 0.62 to 0.98). Conclusion 98% of UK CA registrations relate to diagnoses made in the first year of life. Our data suggest that this leads to incomplete case ascertainment with a further 30% identified after age 1 year in our study. Risk factors for CAs identified up to age 1 year persist up to 5 years. National registries should consider using routine data linkage to provide more complete case ascertainment after infancy. / Collaboration for Leadership in Applied Health Research and Care Yorkshire and Humber programme ‘Healthy Children Healthy Families Theme’ (IS-CLA-0113–10020).

To determine accurately the number of serious injuries at EU level and to compare serious injury rates between different countries it is essential to use a common definition. In January 2013, the High Level Group on Road Safety established the definition of serious injuries as patients with an injury level of MAIS3+(Maximum Abbreviated Injury Scale). Whatever the method used for estimating the number or serious injuries, at some point it is always necessary to use hospital records. The aim of this paper is to understand the implications for (1) in/exclusion criteria applied to case selection and (2) a methodological approach for converting ICD (International Classification of Diseases/Injuries) to MAIS codes, when estimating the number of road traffic serious injuries from hospital data. A descriptive analysis with hospital data from Spain and the Netherlands was carried out to examine the effect of certain choices concerning in- and exclusion criteria based on codes of the ICD9-CM and ICD10. The main parameters explored were: deaths before and after 30 days, readmissions, and external injury causes. Additionally, an analysis was done to explore the impact of using different conversion tools to derive MAIS3 + using data from Austria, Belgium, France, Germany, Netherlands, and Spain. Recommendations are given regarding the in/exclusion criteria and when there is incomplete data to ascertain a road injury, weighting factors could be used to correct data deviations and make more real estimations.

Living in the age of "Big Data" is both a blessing and a curse. On he one hand, the raw data can be analysed and then used for weather redictions, user recommendations, targeted advertising and more. On he other hand, when data is aggregated from multiple sources, there is no guarantee that each source has stored the data in a standardized or even compatible format to what is required by the application. So there is a need to parse the available data and convert it to the desired form. Here is where the problems start to arise: often the correspondences are not quite so straightforward between data instances that belong to the same domain, but come from different sources. For example, in the film industry, information about movies (cast, characters, ratings etc.) can be found on numerous websites such as IMDb or Rotten Tomatoes. Finding and matching all the data referring to the same movie is a challenge. The aim of this project is to select the most efficient algorithm to correlate movie related information gathered from various websites automatically. We have implemented a flexible application that allows us to make the performance comparison of multiple algorithms based on machine learning techniques. According to our experimental results, a well chosen set of rules is on par with the results from a neural network, these two proving to be the most effective classifiers for records with movie information as content.

Royal Perth Hospital is the largest hospital in Western Australia and also has the largest intensive care unit (ICU) in the State. It was the first public hospital to provide intensive care services in Western Australia. This thesis examines the intermediateand long-term outcomes of patients admitted to the Royal Perth Hospital ICU between 1987 and 2002. Intermediate-term survival, defined as survival after discharge from hospital to one year and long-term survival, that exceeding one year after discharge, are important outcomes. Information on outcomes can be used by ICU staff in discussions with patients and their families and to inform policy decision-making and future research. The aim of this research was to examine one-year and long-term outcomes of patients admitted to the ICU between 1987 and 2002 and explore the factors that might be associated with the outcomes for 22,298 patients admitted to the ICU. A clinical ICU database was linked to morbidity and mortality databases by Data Linkage WA. A wide range of demographic and clinical factors were examined for their effect on outcome. These included age, sex, comorbidity, severity of illness, organ failure, ICU diagnostic groups, type of admission (medical, elective surgical and non-elective surgical), length of stay in ICU and era of admission (1987-1990, 1991-1994, 1995-1998, 1999-2002). Patients were followed-up to study end, 31st December 2003 or death if it occurred before study end, that is, up to 17 years after the index ICU admission. Kaplan Meier survival curves and Cox regression models were used to examine intermediate and long-term survival for patients who survived to hospital discharge. A comparison of admissions to hospital before and after the index ICU admission was made using descriptive statistics and logistic regression. Throughout the study period survival for the ICU cohort was shorter when compared to the Australian population. This was consistent throughout the follow-up period. The most important determinants of long-term survival were age, comorbidity, severity of illness and diagnostic group but the strength of association varied with the duration of follow-up. Although age, comorbidity and severity of illness increased among the critically ill survival improved over time. Hospital admissions were more frequent after a discharge from hospital that required an admission to ICU than before the index admission, even after adjusting for the ageing of the cohort. This study provides unique information about the survival and other outcomes of patients discharged from a hospital admission that included an ICU stay. The strength of this study lies in the follow-up to 17 years and the more comprehensive range of explanatory factors than in previous studies. This thesis demonstrates that follow-up studies after intensive care should be of sufficient duration to account for the changes that occur in survival over time and indicates the range of factors that should be taken into account when making comparisons of long-term survival.

Orientador: Prof. Dr. Márcio Katsumi Oikawa / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Ciência da Computação, 2016. / Através do relacionamento probabilístico de bases de dados é possível obter informações que a análise individual ou manual de bases de dados não proporcionaria. Esse trabalho visa encontrar, através do relacionamento probabilístico de registros, doadores de sangue da base de dados da Fundação Pró-Sangue (FPS) no Sistema de Informações sobre Mortalidade (SIM), nos anos de 2001 a 2006, favorecendo assim a manutenção de hemoderivados da instituição, inferindo se determinado doador veio à óbito. Para tal, foram avaliadas a eficiência de diferentes chaves de blocking que foram aplicadas em um conjunto de softwares gratuitos de record linkage e no software implementado para uso específico do estudo, intitulado SortedLink. Nos estudos, os registros foram padronizados e apenas os que possuíam dados da mãe cadastrados foram utilizados. Para avaliar a eficiência das chaves de blocking, foram selecionados 100.000 registros aleatoriamente das bases de dados SIM e FPS, e adicionados 30 registros de validação para cada conjunto. Sendo que o software SortedLink, implementado no trabalho, foi o que apresentou os melhores resultados e foi utilizado para obter os resultados dos possíveis pares de registros na base total de dados, 1.709.819 de registros para o SIM e 334.077 para o FPS. Além disso, o estudo também avalia a eficiência dos algoritmos de codificação fonética SOUNDEX, tipicamente utilizado no processo de record linkage, e do BRSOUND, desenvolvido para codificação de nomes e sobrenomes oriundos da língua portuguesa do Brasil. / Through probabilistic record linkage of databases is possible to obtain information that the individual or manual analysis of databases do not provide. This work aims to find, through probabilistic record relationship, blood donors from the database of Fundação Pró-Sangue (FPS) in the Sistema de Informações sobre Mortalidade (SIM) from Brazil, in the year 2001 to 2006, thus favoring maintenance blood products of the institution, inferring whether a donor came to death. For this purpose, we evaluated the effectiveness of different blocking keys that were applied to a set of free software record linkage and a software implemented for specific use of the study, entitled SortedLink. In the studies, the records were standardized and only those who had registered mother information were used. To assess the effectiveness of blocking keys were selected randomly 100, 000 records of SIM and FPS databases, and added 30 validation records for each set. Since the SortedLink software, implemented in this work, showed the best results, it was used to obtain the results of the possible pairs of records in the total database, 1.709.819 records from SIM and 334.077 from FPS. In addition, the study also evaluated the efficiency of SOUNDEX phonetic encoding algorithms, typically used in the record linkage process and the BRSOUND, developed for encoding names and surnames derived from the Portuguese language of Brazil.

Background: The impact children with complex healthcare needs have on the healthcare system is significant and requires a multidisciplinary response. Congenital anomaly (CA) is a group of conditions requiring complex and variable input from primary and secondary healthcare. This thesis explores the literature on health system preparedness for children with complex healthcare needs and quantitatively describes healthcare use for a population of children with CA, an exemplar for children with complex healthcare needs. Methods: Routine health data from primary care was explored to identify children with CA and linked to secondary care data, outpatient records, and questionnaire data from a multi-ethnic prospective birth cohort over a five-year period. Rates of CA were calculated and healthcare use for children with and without CA was analysed. Results: Out of a birth cohort of 13,857 children, 860 had a CA. Using primary care data for children aged 0 to 5 years, the number of children with CA was found to be 620.6 per 10,000 live births, above the national rate of 226.5 per 10,000 live births. Healthcare use was higher for children with CA than those without CA. Demand for use of hospital services for children with CA was higher (Incident rate ratio (IRR) 4.38, 95% confidence interval (CI) 3.90 to 4.92) than demand for primary care services (IRR, 1.27, 95% CI 1.20 to 1.35). Conclusion: These results suggest that using primary care data as a source of CA case ascertainment reveals more children with CA than previously thought. These results have significant implications for commissioning healthcare services for children with complex healthcare needs.

[Truncated abstract] The primary aims of this thesis were to investigate health indicators of child maltreatment, as well as pathways into the child protection system using routinely collected government databases, enabling a preventative health approach to child abuse and neglect. This thesis aims to improve understanding of the trends in child maltreatment and the factors, at the child and family level, which increase or reduce vulnerability to child maltreatment so more effective prevention policies and practices can be developed. This project uses longitudinal de-identified population data from the Western Australian Government Departments of Child Protection, Health and Disability Services. These data contained information on demographic, clinical, social and child protection outcomes of children and their families. Record linkage of administrative data was undertaken to: investigate health indicators of abuse and neglect using Hospital Morbidity data to enable the monitoring of population trends in abuse and neglect; compare proportion of cases obtained using health indicators with the Department of Child Protection data, and describe the physical, psychological and social characteristics of abused and/or neglected children and families. Statistical techniques utilised include logistic and Cox regression to investigate risk of adverse child outcomes, taking into account potential confounding and time to event. The main findings include: There has been an increase in assault and maltreatment related hospital admissions over the last 25 years. ... There has been a marked increase in the birth prevalence of Neonatal Withdrawal Syndrome (NWS) in Western Australia over the last 25 years, from 1 per 10,000 live births in 1980, to 31 per 10,000 live births in 2005. Specific maternal characteristics associated with having a child with NWS are identified and these children have an increased risk of child protection involvement. A population level analysis of child and parental factors determined the estimated increase in risk of substantiated child maltreatment for child intellectual disability, parental admissions for mental health, substance use, and assault, as well as greater socio-economic disadvantage. Conclusions This is the first body of research which has extensively used longitudinal, population level linked health and child protection data to investigate health indicators of child abuse and neglect and antecedent causal pathways. Monitoring injuries and conditions associated with child abuse and neglect in routinely collected data and using multiple sources of ascertainment are important initiatives in child maltreatment surveillance. Health indicators of child abuse and neglect are not subject to the same definitional and policy issues as child protection data and therefore provide a more valid comparison over time and between jurisdictions. The identification of factors which increase vulnerability for children and families to child maltreatment is essential in the implementation of prevention strategies including universal public health approaches as well as the identification of at-risk families for targeted intervention.

Pedestrian movement is unconstrained. For this reason it is not amenable to mathematical modelling in the same way as road trac. Individual pedestrians are notoriously difficult to monitor at a microscopic level. This has led to a lack of primary data that can be used to develop reliable models. Although video surveillance is cheap to install and operate, video data is extremely expensive to process for this purpose. An alternative approach is to use passive infrared detectors that are able to track individuals unobtrusively. This thesis describesthe use of a low cost infrared sensor for use in tracking pedestrians. The sensor itself, manufactured by a British company, is designed to count people crossing an arbitrary datum line. However, with the development of additional software, the functionality of these sensors can be extended beyond their original design specication. This allows the trajectories of individual pedestrians to be tracked. Although the field of view of each sensor is relatively small (44 m), five were deployed in a busy indoor corridor, covering most of its length. In this research, the technical challenges involved in using the sensors in this way are addressed. Statistics derived from the data collected are then compared to other studies at this scale.

McMaster University MASTER OF PUBLIC HEALTH (2020) Hamilton, Ontario (Health Research Methods, Evidence, and Impact) TITLE: Applying Machine Learning to Determine Nutrients Predictive of Cardiovascular Disease Using Canadian Linked Population-Based Data AUTHOR: Jason D. Morgenstern, B.Sc. (University of Guelph), M.D. (Western University) SUPERVISOR: Professor L.N. Anderson, NUMBER OF PAGES: xv, 121 / The use of big data and machine learning may help to address some challenges in nutritional epidemiology. The first objective of this thesis was to explore the use of machine learning prediction models in a hypothesis-generating approach to evaluate how detailed dietary features contribute to CVD risk prediction. The second objective was to assess the predictive performance of the models. A population-based retrospective cohort study was conducted using linked Canadian data from 2004 – 2018. Study participants were adults age 20 and older (n=12 130 ) who completed the 2004 Canadian Community Health Survey, Cycle 2.2, Nutrition (CCHS 2.2). Statistics Canada has linked the CCHS 2.2 data to the Discharge Abstracts Database and the Canadian Vital Statistics Death database, which were used to determine cardiovascular outcomes (stroke or ischemic heart disease events or deaths). Conditional inference forests were used to develop models. Then, permutation feature importance (PFI) and accumulated local effects (ALEs) were calculated to explore contributions of nutrients to predicted disease. Supplement-use (median PFI (M)=4.09 x 10-4, IQR=8.25 x 10-7 – 1.11 x 10-3) and caffeine (M=2.79 x 10-4, IQR= -9.11 x 10-5 – 5.86 x 10-4) had the highest median PFIs for nutrition-related features. Supplement-use was associated with decreased predicted risk of CVD (accumulated local effects range (ALER)= -3.02 x 10-4 – 2.76 x 10-4) and caffeine was associated with increased predicted risk (ALER= -9.96 x 10-4 – 0.035). The best-performing model had a logarithmic loss of 0.248. Overall, many non-linear relationships were observed, including threshold, j-shaped, and u-shaped. The results of this exploratory study suggest that applying machine learning to the nutritional epidemiology of CVD, particularly using big datasets, may help elucidate risks and improve predictive models. Given the limited application thus far, work such as this could lead to improvements in public health recommendations and policy related to dietary behaviours. / Thesis / Master of Public Health (MPH) / This work explores the potential for machine learning to improve the study of diet and disease. In chapter 2, opportunities are identified for big data to make diet easier to measure. Also, we highlight how machine learning could find new, complex relationships between diet and disease. In chapter 3, we apply a machine learning algorithm, called conditional inference forests, to a unique Canadian dataset to predict whether people developed strokes or heart attacks. This dataset included responses to a health survey conducted in 2004, where participants’ responses have been linked to administrative databases that record when people go to hospital or die up until 2017. Using these techniques, we identified aspects of nutrition that predicted disease, including caffeine, alcohol, and supplement-use. This work suggests that machine learning may be helpful in our attempts to understand the relationships between diet and health.