Factors associated with cardiac implantable electronic device-related infections, New South Wales, 2016-21: a retrospective cohort study

OBJECTIVES

To quantify the rate of cardiac implantable electronic device (CIED)-related infections and to identify risk factors for such infections.

DESIGN

Retrospective cohort study; analysis of linked hospital admissions and mortality data.

SETTING, PARTICIPANTS

All adults who underwent CIED procedures in New South Wales between 1 January 2016 and 30 June 2021 (public hospitals) or 30 June 2020 (private hospitals).

MAIN OUTCOME MEASURES

Proportions of patients hospitalised with CIED-related infections (identified by hospital record diagnosis codes); risk of CIED-related infection by patient, device, and procedural factors.

RESULTS

Of 37 675 CIED procedures (23 194 men, 63.5%), 500 were followed by CIED-related infections (median follow-up, 24.9 months; interquartile range, 11.2-40.8 months), including 397 people (1.1%) within twelve months of their procedures, and 186 of 10 540 people (2.5%) at high risk of such infections (replacement or upgrade procedures; new cardiac resynchronisation therapy with defibrillator, CRT-D). The overall infection rate was 0.50 (95% confidence interval [CI], 0.45-0.54) per 1000 person-months; it was highest during the first month after the procedure (5.60 [95% CI, 4.89-6.42] per 1000 person-months). The risk of CIED-related infection was greater for people under 65 years of age than for those aged 65-74 years (adjusted hazard ratio [aHR], 1.71; 95% CI, 1.32-2.23), for people with CRT-D devices than for those with permanent pacemakers (aHR, 1.46; 95% CI, 1.02-2.08), for people who had previously undergone CIED procedures (two or more v none: aHR, 1.51; 95% CI, 1.02-2.25) or had CIED-related infections (aHR, 11.4; 95% CI, 8.34-15.7), or had undergone concomitant cardiac surgery (aHR, 1.62; 95% CI, 1.10-2.39), and for people with atrial fibrillation (aHR, 1.33; 95% CI, 1.11-1.60), chronic kidney disease (aHR, 1.54; 95% CI, 1.27-1.87), chronic obstructive pulmonary disease (aHR, 1.37; 95% CI, 1.10-1.69), or cardiomyopathy (aHR 1.60; 95% CI, 1.25-2.05).

CONCLUSIONS

Knowledge of risk factors for CIED-related infections can help clinicians discuss them with their patients, identify people at particular risk, and inform decisions about device type, upgrades and replacements, and prophylactic interventions.

Predictive analytics for cardiovascular patient readmission and mortality: An explainable approach

BACKGROUND

Cardiovascular patients experience high rates of adverse outcomes following discharge from hospital, which may be preventable through early identification and targeted action. This study aimed to investigate the effectiveness and explainability of machine learning algorithms in predicting unplanned readmission and death in cardiovascular patients at 30 days and 180 days from discharge.

METHODS

Gradient boosting machines were trained and evaluated using data from hospital electronic medical records linked to hospital administrative and mortality data for 39,255 patients admitted to four hospitals in New South Wales, Australia between 2017 and 2021. Sociodemographic variables, admission history, and clinical information were used as potential predictors. The performance was compared to LASSO regression, as well as the HOSPITAL and LACE risk score indices. Important risk factors identified by the gradient-boosting machine model were explored using Shapley values.

RESULTS

The models performed well, especially for the mortality outcomes. Area under the receiver operating characteristic curve values were 0.70 for readmission and 0.87-0.90 for mortality using the full gradient boosting machine algorithms. Among the top predictors for 30-day and 180-day readmission were increased red cell distribution width, old age (especially above 80 years), high measured troponin and urea levels, not being married or in a relationship, and low albumin levels. For mortality, these included increased red cell distribution width, old age (especially older than 70 years), high measured troponin and urea levels, high neutrophil and monocyte counts, and low eosinophil and lymphocyte counts. The Shapley values gave clear insight into the dynamics of decision-tree-based models.

CONCLUSIONS

We demonstrated an explainable predictive algorithm to identify cardiovascular patients who are at high risk of readmission or death at discharge from the hospital and identified key risk factors.

The Australian Health Care Homes trial: quality of care and patient outcomes. A propensity score-matched cohort study

OBJECTIVE

To assess the impact of the Health Care Homes (HCH) primary health care initiative on quality of care and patient outcomes.

DESIGN, SETTING

Quasi-experimental, matched cohort study; analysis of general practice data extracts and linked administrative data from ten Australian primary health networks, 1 October 2017 - 30 June 2021.

PARTICIPANTS

People with chronic health conditions (practice data extracts: 9811; linked administrative data: 10 682) enrolled in the HCH 1 October 2017 - 30 June 2019; comparison groups of patients receiving usual care (1:1 propensity score-matched).

INTERVENTION

Participants were involved in shared care planning, provided enhanced access to team care, and encouraged to seek chronic condition care at the HCH practice where they were enrolled. Participating practices received bundled payments based on clinical risk tier.

MAIN OUTCOME MEASURES

Access to care, processes of care, diabetes-related outcomes, hospital service use, risk of death.

RESULTS

During the first twelve months after enrolment, the mean numbers of general practitioner encounters (rate ratio, 1.14; 95% confidence interval [CI], 1.11-1.17) and Medicare Benefits Schedule claims for allied health services (rate ratio, 1.28; 95% CI, 1.24-1.33) were higher for the HCH than the usual care group. Annual influenza vaccinations (relative risk, 1.20; 95% CI, 1.17-1.22) and measurements of blood pressure (relative risk, 1.09; 95% CI, 1.08-1.11), blood lipids (relative risk, 1.19; 95% CI, 1.16-1.21), glycated haemoglobin (relative risk, 1.06; 95% CI, 1.03-1.08), and kidney function (relative risk, 1.13; 95% CI, 1.11-1.15) were more likely in the HCH than the usual care group during the twelve months after enrolment. Similar rate ratios and relative risks applied in the second year. The numbers of emergency department presentations (rate ratio, 1.09; 95% CI, 1.02-1.18) and emergency admissions (rate ratio, 1.13; 95% CI, 1.04-1.22) were higher for the HCH group during the first year; other differences in hospital use were not statistically significant. Differences in glycaemic and blood pressure control in people with diabetes in the second year were not statistically significant. By 30 June 2021, 689 people in the HCH group (6.5%) and 646 in the usual care group (6.1%) had died (hazard ratio, 1.07; 95% CI, 0.96-1.20).

CONCLUSIONS

The HCH program was associated with greater access to care and improved processes of care for people with chronic diseases, but not changes in diabetes-related outcomes, most measures of hospital use, or risk of death.

Incidence and Predictors of Readmissions to Non-Index Hospitals After Transcatheter Aortic Valve Implantation in the Contemporary Era in New South Wales, Australia

BACKGROUND

In Australia, transcatheter aortic valve implantation (TAVI) is only performed in a limited number of specialised metropolitan centres, many of which are private hospitals, making it likely that TAVI patients who require readmission will present to another (non-index) hospital. It is important to understand the impact of non-index readmission on patient outcomes and healthcare resource utilisation.

METHOD

We analysed linked hospital and death records for residents of New South Wales, Australia, aged ≥18 years, who had an emergency readmission within 90 days following a TAVI procedure in 2013-2022. Mixed-effect, multi-level logistic regression models were used to evaluate predictors of non-index readmission, and associations between non-index readmission and readmission length of stay, 90-day mortality, and 1-year mortality.

RESULTS

Of 4,198 patients (mean age, 82.7 years; 40.6% female) discharged alive following TAVI, 933 (22.2%) were readmitted within 90 days of discharge. Over three-quarters (76.0%) of those readmitted returned to a non-index hospital, with no significant difference in readmission principal diagnosis between index hospital and non-index hospital readmissions. Among readmitted patients, independent predictors of non-index readmission included: residence in regional or remote areas, lower socio-economic status, having a pre-procedure transfer, and a private index hospital. Readmission length of stay (median, 4 days), 90-day mortality (adjusted odds ratio [OR] 1.04, 95% confidence interval [CI] 0.56-1.96) and 1-year mortality (adjusted OR 1.01, 95% CI 0.64-1.58) were similar between index and non-index readmissions.

CONCLUSIONS

Non-index readmission following TAVI was highly prevalent but not associated with increased mortality or healthcare utilisation. Our results are reassuring for TAVI patients in regional and remote areas with limited access to return to index TAVI hospitals.

Overcoming silos in health care systems through meso-level organisations - a case study of health reforms in New South Wales, Australia

Fragmented care delivery is a barrier to improving health system performance worldwide. Investment in meso-level organisations is a potential strategy to improve health system integration, however, its effectiveness remains unclear. In this paper, we provide an overview of key international and Australian integrated care policies. We then describe Collaborative Commissioning - a novel health reform policy to integrate primary and hospital care sectors in New South Wales (NSW), Australia and provide a case study of a model focussed on older person's care. The policy is theorised to achieve greater integration through improved governance (local stakeholders identifying as part of one health system), service delivery (communities perceive new services as preferable to status quo) and incentives (efficiency gains are reinvested locally with progressively higher value care achieved). If effectively implemented at scale, Collaborative Commissioning has potential to improve health system performance in Australia and will be of relevance to similar reform initiatives in other countries.

Readmission to a non-index hospital following total joint replacement

Aims

It is unclear whether mortality outcomes differ for patients undergoing total hip arthroplasty (THA) or total knee arthroplasty (TKA) surgery who are readmitted to the index hospital where their surgery was performed, or to another hospital.

Methods

We analyzed linked hospital and death records for residents of New South Wales, Australia, aged ≥ 18 years who had an emergency readmission within 90 days following THA or TKA surgery between 2003 and 2022. Multivariable modelling was used to identify factors associated with non-index readmission and to evaluate associations of readmission destination (non-index vs index) with 90-day and one-year mortality.

Results

Of 394,248 joint arthroplasty patients (THA = 149,456; TKA = 244,792), 9.5% (n = 37,431) were readmitted within 90 days, and 53.7% of these were admitted to a non-index hospital. Non-index readmission was more prevalent among patients who underwent surgery in private hospitals (60%). Patients who were readmitted for non-orthopaedic conditions (62.8%), were more likely to return to a non-index hospital compared to those readmitted for orthopaedic complications (39.5%). Factors associated with non-index readmission included older age, higher socioeconomic status, private health insurance, and residence in a rural or remote area. Non-index readmission was significantly associated with 90-day (adjusted odds ratio (aOR) 1.69; 95% confidence interval (CI) 1.39 to 2.05) and one-year mortality (aOR 1.31; 95% CI 1.16 to 1.47). Associations between non-index readmission and mortality were similar for patients readmitted with orthopaedic and non-orthopaedic complications (90-day mortality aOR 1.61; 95% CI 0.98 to 2.64, and aOR 1.67; 95% CI 1.35 to 2.06, respectively).

Conclusion

Non-index readmission was associated with increased mortality, irrespective of whether the readmission was for orthopaedic complications or other conditions.

Enriching Data Science and Health Care Education: Application and Impact of Synthetic Data Sets Through the Health Gym Project

Large-scale medical data sets are vital for hands-on education in health data science but are often inaccessible due to privacy concerns. Addressing this gap, we developed the Health Gym project, a free and open-source platform designed to generate synthetic health data sets applicable to various areas of data science education, including machine learning, data visualization, and traditional statistical models. Initially, we generated 3 synthetic data sets for sepsis, acute hypotension, and antiretroviral therapy for HIV infection. This paper discusses the educational applications of Health Gym's synthetic data sets. We illustrate this through their use in postgraduate health data science courses delivered by the University of New South Wales, Australia, and a Datathon event, involving academics, students, clinicians, and local health district professionals. We also include adaptable worked examples using our synthetic data sets, designed to enrich hands-on tutorial and workshop experiences. Although we highlight the potential of these data sets in advancing data science education and health care artificial intelligence, we also emphasize the need for continued research into the inherent limitations of synthetic data.

Does use of GP and specialist services vary across areas and according to individual socioeconomic position? A multilevel analysis using linked data in Australia

OBJECTIVE

Timely access to primary care and supporting specialist care relative to need is essential for health equity. However, use of services can vary according to an individual's socioeconomic circumstances or where they live. This study aimed to quantify individual socioeconomic variation in general practitioner (GP) and specialist use in New South Wales (NSW), accounting for area-level variation in use.

DESIGN

Outcomes were GP use and quality-of-care and specialist use. Multilevel logistic regression was used to estimate: (1) median ORs (MORs) to quantify small area variation in outcomes, which gives the median increased risk of moving to an area of higher risk of an outcome, and (2) ORs to quantify associations between outcomes and individual education level, our main exposure variable. Analyses were adjusted for individual sociodemographic and health characteristics and performed separately by remoteness categories.

SETTING

Baseline data (2006-2009) from the 45 and Up Study, NSW, Australia, linked to Medicare Benefits Schedule and death data (to December 2012).

PARTICIPANTS

267 153 adults aged 45 years and older.

RESULTS

GP (MOR=1.32-1.35) and specialist use (1.16-1.18) varied between areas, accounting for individual characteristics. For a given level of need and accounting for area variation, low education-level individuals were more likely to be frequent users of GP services (no school certificate vs university, OR=1.63-1.91, depending on remoteness category) and have continuity of care (OR=1.14-1.24), but were less likely to see a specialist (OR=0.85-0.95).

CONCLUSION

GP and specialist use varied across small areas in NSW, independent of individual characteristics. Use of GP care was equitable, but specialist care was not. Failure to address inequitable specialist use may undermine equity gains within the primary care system. Policies should also focus on local variation.

Software Application Profile: The daggle app-a tool to support learning and teaching the graphical rules of selecting adjustment variables using directed acyclic graphs

MOTIVATION

Directed acyclic graphs (DAGs) are used in epidemiological research to communicate causal assumptions and guide the selection of covariate adjustment sets when estimating causal effects. For any given DAG, a set of graphical rules can be applied to identify minimally sufficient adjustment sets that can be used to adjust for bias due to confounding when estimating the causal effect of an exposure on an outcome. The daggle app is a web-based application that aims to assist in the learning and teaching of adjustment set identification using DAGs.

GENERAL FEATURES

The application offers two modes: tutorial and random. The tutorial mode presents a guided introduction to how common causal structures can be presented using DAGs and how graphical rules can be used to identify minimally sufficient adjustment sets for causal estimation. The random mode tests this understanding by presenting the user with a randomly generated DAG-a daggle. To solve the daggle, users must correctly identify a valid minimally sufficient adjustment set.

IMPLEMENTATION

The daggle app is implemented as an R shiny application using the golem framework. The application builds upon existing R libraries including pcalg to generate reproducible random DAGs, dagitty to identify all valid minimal adjustment sets and ggdag to visualize DAGs.

AVAILABILITY

The daggle app can be accessed online at [http://cbdrh.shinyapps.io/daggle]. The source code is available on GitHub [https://github.com/CBDRH/daggle] and is released under a Creative Commons CC BY-NC-SA 4.0 licence.

Elective spinal surgery in New South Wales adults, 2001-20, by procedure funding type: a cross-sectional study

OBJECTIVE

To investigate elective rates of spinal fusion, decompression, and disc replacement procedures for people with degenerative conditions, by funding type (public, private, workers' compensation).

DESIGN, SETTING

Cross-sectional study; analysis of hospitals admissions data extracted from the New South Wales Admitted Patient Data Collection.

PARTICIPANTS

All adults who underwent elective spinal surgery (spinal fusion, decompression, disc replacement) in NSW, 1 July 2001 - 30 June 2020.

MAIN OUTCOME MEASURES

Crude and age- and sex-adjusted procedure rates, by procedure, funding type, and year; annual change in rates, 2001-20, expressed as incidence rate ratios (IRRs).

RESULTS

During 2001-20, 155 088 procedures in 129 525 adults were eligible for our analysis: 53 606 fusion, 100 225 decompression, and 1257 disc replacement procedures. The privately funded fusion procedure rate increased from 26.6 to 109.5 per 100 000 insured adults (per year: IRR, 1.06; 95% confidence interval [CI], 1.05-1.07); the workers' compensation procedure rate increased from 6.1 to 15.8 per 100 000 covered adults (IRR, 1.04; 95% CI, 1.01-1.06); the publicly funded procedure rate increased from 5.6 to 12.4 per 100 000 adults (IRR, 1.03; 95% CI, 1.01-1.06), and from 10.5 to 22.1 per 100 000 adults without hospital cover private health insurance (IRR, 1.03; 95% CI, 1.01-1.05). The privately funded decompression procedure rate increased from 93.4 to 153.6 per 100 000 people (IRR, 1.02; 95% CI, 1.01-1.03); the workers' compensation procedure rate declined from 19.7 to 16.7 per 100 000 people (IRR, 0.98; 95% CI, 0.96-0.99), and the publicly funded procedure rate did not change significantly. The privately funded disc replacement procedure rate increased from 6.2 per million in 2010-11 to 38.4 per million people in 2019-20, but did not significantly change for the other two funding groups. The age- and sex-adjusted rates for privately and publicly funded fusion and decompression procedures were similar to the crude rates.

CONCLUSIONS

Privately funded spinal surgery rates continue to be larger than for publicly funded procedures, and they have also increased more rapidly. These differences may indicate that some privately funded procedures are unnecessary, or that the number of publicly funded procedures does not reflect clinical need.

Risk for Congenital Anomalies in Children Conceived With Medically Assisted Fertility Treatment : A Population-Based Cohort Study

BACKGROUND

More than 2 million children are conceived annually using assisted reproductive technologies (ARTs), with a similar number conceived using ovulation induction and intrauterine insemination (OI/IUI). Previous studies suggest that ART-conceived children are at increased risk for congenital anomalies (CAs). However, the role of underlying infertility in this risk remains unclear, and ART clinical and laboratory practices have changed drastically over time, particularly there has been an increase in intracytoplasmic sperm injection (ICSI) and cryopreservation.

OBJECTIVE

To investigate the role of underlying infertility and fertility treatment on CA risks in the first 2 years of life.

DESIGN

Propensity score-weighted population-based cohort study.

SETTING

New South Wales, Australia.

PARTICIPANTS

851 984 infants (828 099 singletons and 23 885 plural children) delivered between 2009 and 2017.

MEASUREMENTS

Adjusted risk difference (aRD) in CAs of infants conceived through fertility treatment compared with 2 naturally conceived (NC) control groups-those with and without a parental history of infertility (NC-infertile and NC-fertile).

RESULTS

The overall incidence of CAs was 459 per 10 000 singleton births and 757 per 10 000 plural births. Compared with NC-fertile singleton control infants (n = 747 018), ART-conceived singleton infants (n = 31 256) had an elevated risk for major genitourinary abnormalities (aRD, 19.0 cases per 10 000 births [95% CI, 2.3 to 35.6]); the risk remained unchanged (aRD, 22 cases per 10 000 births [CI, 4.6 to 39.4]) when compared with NC-infertile singleton control infants (n = 36 251) (that is, after accounting for parental infertility), indicating that ART remained an independent risk. After accounting for parental infertility, ICSI in couples without male infertility was associated with an increased risk for major genitourinary abnormalities (aRD, 47.8 cases per 10 000 singleton births [CI, 12.6 to 83.1]). There was some suggestion of increased risk for CAs after fresh embryo transfer, although estimates were imprecise and inconsistent. There were no increased risks for CAs among OI/IUI-conceived infants (n = 13 574).

LIMITATIONS

This study measured the risk for CAs only in those children who were born at or after 20 weeks' gestation. Observational study design precludes causal inference. Many estimates were imprecise.

CONCLUSION

Patients should be counseled on the small increased risk for genitourinary abnormalities after ART, particularly after ICSI, which should be avoided in couples without problems of male infertility.

PRIMARY FUNDING SOURCE

Australian National Health and Medical Research Council.

Automated ICD coding using extreme multi-label long text transformer-based models

Encouraged by the success of pretrained Transformer models in many natural language processing tasks, their use for International Classification of Diseases (ICD) coding tasks is now actively being explored. In this study, we investigated two existing Transformer-based models (PLM-ICD and XR-Transformer) and proposed a novel Transformer-based model (XR-LAT), aiming to address the extreme label set and long text classification challenges that are posed by automated ICD coding tasks. The Transformer-based model PLM-ICD, which currently holds the state-of-the-art (SOTA) performance on the ICD coding benchmark datasets MIMIC-III and MIMIC-II, was selected as our baseline model for further optimisation on both datasets. In addition, we extended the capabilities of the leading model in the general extreme multi-label text classification domain, XR-Transformer, to support longer sequences and trained it on both datasets. Moreover, we proposed a novel model, XR-LAT, which was also trained on both datasets. XR-LAT is a recursively trained model chain on a predefined hierarchical code tree with label-wise attention, knowledge transferring and dynamic negative sampling mechanisms. Our optimised PLM-ICD models, which were trained with longer total and chunk sequence lengths, significantly outperformed the current SOTA PLM-ICD models, and achieved the highest micro-F1 scores of 60.8 % and 50.9 % on MIMIC-III and MIMIC-II, respectively. The XR-Transformer model, although SOTA in the general domain, did not perform well across all metrics. The best XR-LAT based models obtained results that were competitive with the current SOTA PLM-ICD models, including improving the macro-AUC by 2.1 % and 5.1 % on MIMIC-III and MIMIC-II, respectively. Our optimised PLM-ICD models are the new SOTA models for automated ICD coding on both datasets, while our novel XR-LAT models perform competitively with the previous SOTA PLM-ICD models.

Continuous time recurrent neural networks: Overview and benchmarking at forecasting blood glucose in the intensive care unit

OBJECTIVE

Blood glucose measurements in the intensive care unit (ICU) are typically made at irregular intervals. This presents a challenge in choice of forecasting model. This article gives an overview of continuous time autoregressive recurrent neural networks (CTRNNs) and evaluates how they compare to autoregressive gradient boosted trees (GBT) in forecasting blood glucose in the ICU.

METHODS

Continuous time autoregressive recurrent neural networks (CTRNNs) are a deep learning model that account for irregular observations through incorporating continuous evolution of the hidden states between observations. This is achieved using a neural ordinary differential equation (ODE) or neural flow layer. In this manuscript, we give an overview of these models, including the varying architectures that have been proposed to account for issues such as ongoing medical interventions. Further, we demonstrate the application of these models to probabilistic forecasting of blood glucose in a critical care setting using electronic medical record and simulated data and compare with GBT and linear models.

RESULTS

The experiments confirm that addition of a neural ODE or neural flow layer generally improves the performance of autoregressive recurrent neural networks in the irregular measurement setting. However, several CTRNN architecture are outperformed by a GBT model (Catboost), with only a long short-term memory (LSTM) and neural ODE based architecture (ODE-LSTM) achieving comparable performance on probabilistic forecasting metrics such as the continuous ranked probability score (ODE-LSTM: 0.118 ± 0.001; Catboost: 0.118 ± 0.001), ignorance score (0.152 ± 0.008; 0.149 ± 0.002) and interval score (175 ± 1; 176 ± 1).

CONCLUSION

The application of deep learning methods for forecasting in situations with irregularly measured time series such as blood glucose shows promise. However, appropriate benchmarking by methods such as GBT approaches (plus feature transformation) are key in highlighting whether novel methodologies are truly state of the art in tabular data settings.

Study protocol for development and validation of a single tool to assess risks of stroke, diabetes mellitus, myocardial infarction and dementia: DemNCD-Risk

INTRODUCTION

Current efforts to reduce dementia focus on prevention and risk reduction by targeting modifiable risk factors. As dementia and cardiometabolic non-communicable diseases (NCDs) share risk factors, a single risk-estimating tool for dementia and multiple NCDs could be cost-effective and facilitate concurrent assessments as compared with a conventional single approach. The aim of this study is to develop and validate a new risk tool that estimates an individual's risk of developing dementia and other NCDs including diabetes mellitus, stroke and myocardial infarction. Once validated, it could be used by the public and general practitioners.

METHODS AND ANALYSIS

Ten high-quality cohort studies from multiple countries were identified, which met eligibility criteria, including large representative samples, long-term follow-up, data on clinical diagnoses of dementia and NCDs, recognised modifiable risk factors for the four NCDs and mortality data. Pooled harmonised data from the cohorts will be used, with 65% randomly allocated for development of the predictive model and 35% for testing. Predictors include sociodemographic characteristics, general health risk factors and lifestyle/behavioural risk factors. A subdistribution hazard model will assess the risk factors' contribution to the outcome, adjusting for competing mortality risks. Point-based scoring algorithms will be built using predictor weights, internally validated and the discriminative ability and calibration of the model will be assessed for the outcomes. Sensitivity analyses will include recalculating risk scores using logistic regression.

ETHICS AND DISSEMINATION

Ethics approval is provided by the University of New South Wales Human Research Ethics Committee (UNSW HREC; protocol numbers HC200515, HC3413). All data are deidentified and securely stored on servers at Neuroscience Research Australia. Study findings will be presented at conferences and published in peer-reviewed journals. The tool will be accessible as a public health resource. Knowledge translation and implementation work will explore strategies to apply the tool in clinical practice.

The relationship between hyperglycaemia on admission and patient outcome is modified by hyperlactatemia and diabetic status: a retrospective analysis of the eICU collaborative research database

Both blood glucose and lactate are well-known predictors of organ dysfunction and mortality in critically ill patients. Previous research has shown that concurrent adjustment for glucose and lactate modifies the relationship between these variables and patient outcomes, including blunting of the association between blood glucose and patient outcome. We aim to investigate the relationship between ICU admission blood glucose and hospital mortality while accounting for lactate and diabetic status. Across 43,250 ICU admissions, weighted to account for missing data, we assessed the predictive ability of several logistic regression and generalised additive models that included blood glucose, blood lactate and diabetic status. We found that inclusion of blood glucose marginally improved predictive performance in all patients: AUC-ROC 0.665 versus 0.659 (p = 0.005), with a greater degree of improvement seen in non-diabetics: AUC-ROC 0.675 versus 0.663 (p < 0.001). Inspection of the estimated risk profiles revealed the standard U-shaped risk profile for blood glucose was only present in non-diabetic patients after controlling for blood lactate levels. Future research should aim to utilise observational data to estimate whether interventions such as insulin further modify this effect, with the goal of informing future RCTs of interventions targeting glycaemic control in the ICU.

Long-term Mortality and Reintervention After Endovascular and Open Abdominal Aortic Aneurysm Repairs in Australia, Germany, and the United States

OBJECTIVE

To examine long-term outcomes after endovascular (EVAR) and open repairs (OAR) for intact abdominal aortic aneurysms in Australia, Germany, and the United States, using a unified study design.

BACKGROUND

Similarities and differences in long-term outcomes after EVAR versus OAR across countries remained unclear, given differences in designs across existing studies.

METHODS

We identified patients aged >65 years undergoing intact abdominal aortic aneurysm repairs during 2010-2017/2018. We compared long-term patient mortality and reintervention after EVAR and OAR using Kaplan-Meier analyses and Cox regressions. Propensity score matching was performed within each country to adjust for differences in baseline patient characteristics between procedure groups.

RESULTS

We included 3311, 4909, and 145363 patients from Australia, Germany, and the United States, respectively. The median patient age was 76 to 77 years, and most patients were males (77%-84%). Patient mortality was lower after EVAR than OAR within the first 60 days and became similar at 3-year follow-up (Australia 14.7% vs 16.5%, Germany 18.2% vs 19.7%, United States: 24.4% vs 24.4%). At the end of follow-up, patient mortality after EVAR was higher than OAR in Australia [ hazard ratio (HR) 95% CI: 1.21 (0.96-1.54)] but similar to OAR in Germany [HR 95% CI: 0.92 (0.80-1.07)] and the United States [HR 95% CI: 1.02 (0.99-1.05)]. The risk of reintervention after EVAR was more than twice that after OAR in Australia [HR 95% CI: 2.60 (1.09-6.15)], Germany [HR 95% CI: 4.79 (2.56-8.98)], and the United States [HR 95% CI: 2.67 (2.38-3.00)]. The difference in reintervention risk appeared early in German and United States patients.

CONCLUSIONS

This multinational study demonstrated important similarities in long-term outcomes after EVAR versus OAR across 3 countries. Variation in long-term mortality and reintervention comparisons indicates possible differences in patient profiles, surveillance, and best medical therapy across countries.

Web-Based Application Based on Human-in-the-Loop Deep Learning for Deidentifying Free-Text Data in Electronic Medical Records: Development and Usability Study

BACKGROUND

The narrative free-text data in electronic medical records (EMRs) contain valuable clinical information for analysis and research to inform better patient care. However, the release of free text for secondary use is hindered by concerns surrounding personally identifiable information (PII), as protecting individuals' privacy is paramount. Therefore, it is necessary to deidentify free text to remove PII. Manual deidentification is a time-consuming and labor-intensive process. Numerous automated deidentification approaches and systems have been attempted to overcome this challenge over the past decade.

OBJECTIVE

We sought to develop an accurate, web-based system deidentifying free text (DEFT), which can be readily and easily adopted in real-world settings for deidentification of free text in EMRs. The system has several key features including a simple and task-focused web user interface, customized PII types, use of a state-of-the-art deep learning model for tagging PII from free text, preannotation by an interactive learning loop, rapid manual annotation with autosave, support for project management and team collaboration, user access control, and central data storage.

METHODS

DEFT comprises frontend and backend modules and communicates with central data storage through a filesystem path access. The frontend web user interface provides end users with a user-friendly workspace for managing and annotating free text. The backend module processes the requests from the frontend and performs relevant persistence operations. DEFT manages the deidentification workflow as a project, which can contain one or more data sets. Customized PII types and user access control can also be configured. The deep learning model is based on a Bidirectional Long Short-Term Memory-Conditional Random Field (BiLSTM-CRF) with RoBERTa as the word embedding layer. The interactive learning loop is further integrated into DEFT to speed up the deidentification process and increase its performance over time.

RESULTS

DEFT has many advantages over existing deidentification systems in terms of its support for project management, user access control, data management, and an interactive learning process. Experimental results from DEFT on the 2014 i2b2 data set obtained the highest performance compared to 5 benchmark models in terms of microaverage strict entity-level recall and F1-scores of 0.9563 and 0.9627, respectively. In a real-world use case of deidentifying clinical notes, extracted from 1 referral hospital in Sydney, New South Wales, Australia, DEFT achieved a high microaverage strict entity-level F1-score of 0.9507 on a corpus of 600 annotated clinical notes. Moreover, the manual annotation process with preannotation demonstrated a 43% increase in work efficiency compared to the process without preannotation.

CONCLUSIONS

DEFT is designed for health domain researchers and data custodians to easily deidentify free text in EMRs. DEFT supports an interactive learning loop and end users with minimal technical knowledge can perform the deidentification work with only a shallow learning curve.

Generating synthetic clinical data that capture class imbalanced distributions with generative adversarial networks: Example using antiretroviral therapy for HIV

OBJECTIVE

Clinical data's confidential nature often limits the development of machine learning models in healthcare. Generative adversarial networks (GANs) can synthesise realistic datasets, but suffer from mode collapse, resulting in low diversity and bias towards majority demographics and common clinical practices. This work proposes an extension to the classic GAN framework that includes a variational autoencoder (VAE) and an external memory mechanism to overcome these limitations and generate synthetic data accurately describing imbalanced class distributions commonly found in clinical variables.

METHODS

The proposed method generated a synthetic dataset related to antiretroviral therapy for human immunodeficiency virus (ART for HIV). We evaluated it based on five metrics: (1) accurately representing imbalanced class distribution; (2) the realism of the individual variables; (3) the realism among variables; (4) patient disclosure risk; and (5) the utility of the generated dataset for developing downstream machine learning models.

RESULTS

The proposed method overcomes the issue of mode collapse and generates a synthetic dataset that accurately describes imbalanced class distributions commonly found in clinical variables. The generated data has a patient disclosure risk of 0.095%, lower than the 9% threshold stated by Health Canada and the European Medicines Agency, making it suitable for distribution to the research community with high security. The generated data also has high utility, indicating the potential of the proposed method to enable the development of downstream machine learning algorithms for healthcare applications using synthetic data.

CONCLUSION

Our proposed extension to the classic GAN framework, which includes a VAE and an external memory mechanism, represents a promising approach towards generating synthetic data that accurately describe imbalanced class distributions commonly found in clinical variables. This method overcomes the limitations of GANs and creates more realistic datasets with higher patient cohort diversity, facilitating the development of downstream machine learning algorithms for healthcare applications.

Comparative effect of varenicline and nicotine patches on preventing repeat cardiovascular events

OBJECTIVE

To determine the comparative effectiveness of postdischarge use of varenicline versus prescription nicotine replacement therapy (NRT) patches for the prevention of recurrent cardiovascular events and mortality and whether this association differs by sex.

METHODS

Our cohort study used routinely collected hospital, pharmaceutical dispensing and mortality data for residents of New South Wales, Australia. We included patients hospitalised for a major cardiovascular event or procedure 2011-2017, who were dispensed varenicline or prescription NRT patches within 90day postdischarge. Exposure was defined using an approach analogous to intention to treat. Using inverse probability of treatment weighting with propensity scores to account for confounding, we estimated adjusted HRs for major cardiovascular events (MACEs), overall and by sex. We fitted an additional model with a sex-treatment interaction term to determine if treatment effects differed between males and females.

RESULTS

Our cohort of 844 varenicline users (72% male, 75% <65 years) and 2446 prescription NRT patch users (67% male, 65% <65 years) were followed for a median of 2.93 years and 2.34 years, respectively. After weighting, there was no difference in risk of MACE for varenicline relative to prescription NRT patches (aHR 0.99, 95% CI 0.82 to 1.19). We found no difference (interaction p=0.098) between males (aHR 0.92, 95% CI 0.73 to 1.16) and females (aHR 1.30, 95% CI 0.92 to 1.84), although the effect among females deviated from the null.

CONCLUSION

We found no difference between varenicline and prescription NRT patches in the risk of recurrent MACE. These results should be considered when determining the most appropriate choice of smoking cessation pharmacotherapy.

Associations of Breast Arterial Calcifications with Cardiovascular Disease

Cardiovascular diseases (CVD), including coronary artery disease (CAD), continue to be the leading cause of global mortality among women. While traditional CVD/CAD prevention tools play a significant role in reducing morbidity and mortality among both men and women, current tools for preventing CVD/CAD rely on traditional risk factor-based algorithms that often underestimate CVD/CAD risk in women compared with men. In recent years, some studies have suggested that breast arterial calcifications (BAC), which are benign calcifications seen in mammograms, may be linked to CVD/CAD. Considering that millions of women older than 40 years undergo annual screening mammography for breast cancer as a regular activity, innovative risk prediction factors for CVD/CAD involving mammographic data could offer a gender-specific and convenient solution. Such factors that may be independent of, or complementary to, current risk models without extra cost or radiation exposure are worthy of detailed investigation. This review aims to discuss relevant studies examining the association between BAC and CVD/CAD and highlights some of the issues related to previous studies' design such as sample size, population types, method of assessing BAC and CVD/CAD, definition of cardiovascular events, and other confounding factors. The work may also offer insights for future CVD risk prediction research directions using routine mammograms and radiomic features other than BAC such as breast density and macrocalcifications.

Machine-learning versus traditional approaches for atherosclerotic cardiovascular risk prognostication in primary prevention cohorts: a systematic review and meta-analysis

BACKGROUND

Cardiovascular disease (CVD) risk prediction is important in guiding the intensity of therapy in CVD prevention. Whilst current risk prediction algorithms use traditional statistical approaches, machine learning (ML) presents an alternative method that may improve risk prediction accuracy. This systematic review and meta-analysis aimed to investigate whether ML algorithms demonstrate greater performance compared to traditional risk scores in CVD risk prognostication.

METHODS

MEDLINE, EMBASE, CENTRAL and SCOPUS Web of Science Core collection were searched for studies comparing ML models to traditional risk scores for CV risk prediction between the years 2000 and 2021. We included studies which assessed both ML and traditional risk scores in adult (>18 years old) primary prevention populations. We assessed risk of bias using the Prediction model Risk of Bias Assessment Tool (PROBAST) tool. Only studies which provided a measure of discrimination (i.e. C-statistics with 95% confidence intervals) were included in the meta-analysis.

RESULTS

Sixteen studies were included in the review and meta-analysis (3 302 515 individuals). All study designs were retrospective cohort studies. Three of 16 studies externally validated their models, and 11 reported calibration metrics. Eleven studies demonstrated a high risk of bias. The summary c-statistics (95% CI) of the top performing ML models and traditional risk scores were 0.773 (95%CI: 0.740-0.806) and 0.759 (95%CI: 0.726-0.792) respectively. The difference in c-statistic was 0.0139 (95%CI 0.0139-0.140), P < 0.0001.

CONCLUSION

ML models outperformed traditional risk scores in discrimination of CVD risk prognostication. Integration of ML algorithms into electronic healthcare systems in primary care could improve identification of patients at high risk of subsequent CV events and hence increase opportunities for CVD prevention. It is uncertain whether they can be implemented in clinical settings. Future implementation research is needed to examine how ML models may be utilised for primary prevention.This review was registered with PROSPERO (CRD42020220811).

Maternal asthma in Australian indigenous women and perinatal outcomes: A whole population-linked study

OBJECTIVE

To assess the association between maternal asthma and adverse perinatal outcomes in an Australian Indigenous population.

METHODS

This prospective cohort study included all Indigenous mother and baby dyads for births from 2001 to 2013 in Western Australia (n = 25 484). Data were linked from Western Australia Births, Deaths, Midwives, Hospital, and Emergency Department collections. Maternal asthma was defined as a self-reported diagnosis at an antenatal visit or hospitalization or emergency visit for asthma during pregnancy or less than 3 years before pregnancy. Associations with birth, labor, and pregnancy outcomes were assessed using generalized estimating equations. Asthma exacerbation during pregnancy and stratification by remoteness was also assessed.

RESULTS

Maternal asthma was associated with placental abruption (adjusted odds ratio [aOR], 1.59 [95% confidence interval (CI), 1.07-2.35]), threatened preterm labor (aOR, 1.58 [95% CI, 1.39-1.79]), and emergency cesarean sections (aOR, 1.27 [95% CI, 1.13-1.44]). These risks increased further with an asthma exacerbation during pregnancy or if the mother was from a remote area. No associations were found for low birth weight, preterm birth, small for gestational age, or perinatal mortality.

CONCLUSION

Maternal asthma in Indigenous women is associated with an increased risk of emergency cesarean sections, placental abruption, and threatened preterm labor. These risks may be mitigated by improved management of asthma exacerbations during pregnancy.

Trends and Outcomes for Percutaneous Coronary Intervention and Coronary Artery Bypass Graft Surgery in New South Wales from 2008 to 2019

Risk profiles are changing for patients who undergo percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG). In Australia, little is known of the nature of these changes in contemporary practice and of the impact on patient outcomes. We identified all CABG (n = 40,805) and PCI (n = 142,399) procedures in patients aged ≥18 years in New South Wales, Australia, during 2008 to 2019. Between 2008 and 2019, the age- and gender-standardized revascularization rate increased by 20% (from 267/100,000 to 320/100,000 population) for all revascularizations. The increase in revascularization was particularly driven by a 35% increase (from 194/100,000 to 261/100,000) in PCI, whereas the rate of CABG decreased by 20% (from 73/100,000 to 59/100,000). Mean age and the prevalence of co-morbidities (especially diabetes and atrial fibrillation) increased for patients with PCI in more recent years but remained consistently lower than for patients with CABG. CABGs performed in patients presenting with a non-ST-segment-elevation acute coronary syndrome halved from 34.3% to 18.7% during the study period, whereas PCIs in this group decreased from 36.5% to 29.6%. Risk-adjusted in-hospital mortality decreased by 7.5 deaths/1,000 procedures per month for CABG but remained unchanged for PCI. Risk-adjusted readmission rates were consistently higher for CABG than for PCI and did not change significantly over time. In conclusion, we observed a dramatic shift over time from CABG to PCI as the revascularization procedure of choice, with the patient base for PCI extending to older and sicker patients. There was a large decrease in mortality after CABG, whereas mortality after PCI remained unchanged.

Uptake of prescription smoking cessation pharmacotherapies after hospitalization for major cardiovascular disease

AIMS

We determined the prevalence of prescription smoking cessation pharmacotherapy (SCP) use after hospitalization for major cardiovascular disease (MCD) among people who smoke and whether this varies by sex.

METHODS AND RESULTS

We conducted a population-based cohort study including all people hospitalized in New South Wales, Australia, between July 2013 and December 2018 (2017 for private hospitals) with an MCD diagnosis. For patients who also had a diagnosis of current tobacco use, we used linked pharmaceutical dispensing records to identify prescription SCP dispensings within 90 days post-discharge. We determined the proportion who were dispensed an SCP within 90 days, overall and by type of SCP. We used logistic regression to estimate the odds of females being dispensed an SCP relative to males. Of the 150 758 patients hospitalized for an MCD, 20 162 (13.4%) had a current tobacco use diagnosis, 31% of whom were female. Of these, 11.3% (12.4% of females, 10.9% of males) received prescription SCP within 90 days post-discharge; 3.0% were dispensed varenicline, and 8.3% were dispensed nicotine replacement therapy patches. Females were more likely than males to be dispensed a prescription SCP [odds ratio (OR) 1.16, 95% confidence interval (CI) 1.06-1.27)]; however, this was not maintained after adjusting for potential confounders (adjusted OR 1.04, 95% CI 0.94-1.15).

CONCLUSION

Very few females and males who smoke use prescription SCPs after hospitalization for an MCD. The use of varenicline, the SCP with the highest efficacy, was particularly low. This represents a missed opportunity to increase smoking cessation in this high-risk population, thereby reducing their risk of recurrent cardiovascular events.

The Health Gym: synthetic health-related datasets for the development of reinforcement learning algorithms

In recent years, the machine learning research community has benefited tremendously from the availability of openly accessible benchmark datasets. Clinical data are usually not openly available due to their confidential nature. This has hampered the development of reproducible and generalisable machine learning applications in health care. Here we introduce the Health Gym - a growing collection of highly realistic synthetic medical datasets that can be freely accessed to prototype, evaluate, and compare machine learning algorithms, with a specific focus on reinforcement learning. The three synthetic datasets described in this paper present patient cohorts with acute hypotension and sepsis in the intensive care unit, and people with human immunodeficiency virus (HIV) receiving antiretroviral therapy. The datasets were created using a novel generative adversarial network (GAN). The distributions of variables, and correlations between variables and trends in variables over time in the synthetic datasets mirror those in the real datasets. Furthermore, the risk of sensitive information disclosure associated with the public distribution of the synthetic datasets is estimated to be very low.

Estimating incidence rates of periprosthetic joint infection after hip and knee arthroplasty for osteoarthritis using linked registry and administrative health data

AIMS

The aim of this study was to estimate the 90-day periprosthetic joint infection (PJI) rates following total knee arthroplasty (TKA) and total hip arthroplasty (THA) for osteoarthritis (OA).

METHODS

This was a data linkage study using the New South Wales (NSW) Admitted Patient Data Collection (APDC) and the Australian Orthopaedic Association National Joint Replacement Registry (AOANJRR), which collect data from all public and private hospitals in NSW, Australia. Patients who underwent a TKA or THA for OA between 1 January 2002 and 31 December 2017 were included. The main outcome measures were 90-day incidence rates of hospital readmission for: revision arthroplasty for PJI as recorded in the AOANJRR; conservative definition of PJI, defined by T84.5, the PJI diagnosis code in the APDC; and extended definition of PJI, defined by the presence of either T84.5, or combinations of diagnosis and procedure code groups derived from recursive binary partitioning in the APDC.

RESULTS

The mean 90-day revision rate for infection was 0.1% (0.1% to 0.2%) for TKA and 0.3% (0.1% to 0.5%) for THA. The mean 90-day PJI rates defined by T84.5 were 1.3% (1.1% to 1.7%) for TKA and 1.1% (0.8% to 1.3%) for THA. The mean 90-day PJI rates using the extended definition were 1.9% (1.5% to 2.2%) and 1.5% (1.3% to 1.7%) following TKA and THA, respectively.

CONCLUSION

When reporting the revision arthroplasty for infection, the AOANJRR substantially underestimates the rate of PJI at 90 days. Using combinations of infection codes and PJI-related surgical procedure codes in linked hospital administrative databases could be an alternative way to monitor PJI rates.Cite this article: Bone Joint J 2022;104-B(9):1060-1066.

Hierarchical label-wise attention transformer model for explainable ICD coding

International Classification of Diseases (ICD) coding plays an important role in systematically classifying morbidity and mortality data. In this study, we propose a hierarchical label-wise attention Transformer model (HiLAT) for the explainable prediction of ICD codes from clinical documents. HiLAT firstly fine-tunes a pretrained Transformer model to represent the tokens of clinical documents. We subsequently employ a two-level hierarchical label-wise attention mechanism that creates label-specific document representations. These representations are in turn used by a feed-forward neural network to predict whether a specific ICD code is assigned to the input clinical document of interest. We evaluate HiLAT using hospital discharge summaries and their corresponding ICD-9 codes from the MIMIC-III database. To investigate the performance of different types of Transformer models, we develop ClinicalplusXLNet, which conducts continual pretraining from XLNet-Base using all the MIMIC-III clinical notes. The experiment results show that the F1 scores of the HiLAT + ClinicalplusXLNet outperform the previous state-of-the-art models for the top-50 most frequent ICD-9 codes from MIMIC-III. Visualisations of attention weights present a potential explainability tool for checking the face validity of ICD code predictions.

Differences in the pre-hospital management of women and men with stroke by emergency medical services in New South Wales

OBJECTIVES

To examine whether pre-hospital emergency medical service care differs for women and men subsequently admitted to hospital with stroke.

DESIGN, SETTING, PARTICIPANTS

Population-based cohort study; analysis of linked Admitted Patient Data Collection and NSW Ambulance data for people admitted to New South Wales hospitals with a principal diagnosis of stroke at separation, 1 July 2005 - 31 December 2018.

MAIN OUTCOME MEASURES

Emergency medical service assessments, protocols, and management for patients subsequently diagnosed with stroke, by sex.

RESULTS

Of 202 231 people hospitalised with stroke (mean age, 73 [SD, 14] years; 98 599 women [51.0%]), 101 357 were conveyed to hospital by ambulance (50.1%). A larger proportion of women than men travelled by ambulance (52.4% v 47.9%; odds ratio [OR], 1.09; 95% CI, 1.07-1.11), but time between the emergency call and emergency department admission was similar for both sexes. The likelihood of being assessed as having a stroke (adjusted OR [aOR], 0.97; 95% CI, 0.93-1.01) or subarachnoid haemorrhage (aOR, 1.22; 95% CI, 0.73-2.03) was similar for women and men, but women under 70 years of age were less likely than men to be assessed as having a stroke (aOR, 0.89; 95% CI, 0.82-0.97). Women were more likely than men to be assessed by paramedics as having migraine, other headache, anxiety, unconsciousness, hypertension, or nausea. Women were less likely than men to be managed according to the NSW Ambulance pre-hospital stroke care protocol (aOR, 0.95; 95% CI, 0.92-0.97), but the likelihood of basic pre-hospital care was similar for both sexes (aOR, 1.01; 95% CI, 0.99-1.04).

CONCLUSION

Our large population-based study identified sex differences in pre-hospital management by emergency medical services of women and men admitted to hospital with stroke. Paramedics should receive training that improves the recognition of stroke symptoms in women.

A new and automated risk prediction of coronary artery disease using clinical endpoints and medical imaging-derived patient-specific insights: protocol for the retrospective GeoCAD cohort study

INTRODUCTION

Coronary artery disease (CAD) is the leading cause of death worldwide. More than a quarter of cardiovascular events are unexplained by current absolute cardiovascular disease risk calculators, and individuals without clinical risk factors have been shown to have worse outcomes. The 'anatomy of risk' hypothesis recognises that adverse anatomical features of coronary arteries enhance atherogenic haemodynamics, which in turn mediate the localisation and progression of plaques. We propose a new risk prediction method predicated on CT coronary angiography (CTCA) data and state-of-the-art machine learning methods based on a better understanding of anatomical risk for CAD. This may open new pathways in the early implementation of personalised preventive therapies in susceptible individuals as a potential key in addressing the growing burden of CAD.

METHODS AND ANALYSIS

GeoCAD is a retrospective cohort study in 1000 adult patients who have undergone CTCA for investigation of suspected CAD. It is a proof-of-concept study to test the hypothesis that advanced image-derived patient-specific data can accurately predict long-term cardiovascular events. The objectives are to (1) profile CTCA images with respect to variations in anatomical shape and associated haemodynamic risk expressing, at least in part, an individual's CAD risk, (2) develop a machine-learning algorithm for the rapid assessment of anatomical risk directly from unprocessed CTCA images and (3) to build a novel CAD risk model combining traditional risk factors with these novel anatomical biomarkers to provide a higher accuracy CAD risk prediction tool.

ETHICS AND DISSEMINATION

The study protocol has been approved by the St Vincent's Hospital Human Research Ethics Committee, Sydney-2020/ETH02127 and the NSW Population and Health Service Research Ethics Committee-2021/ETH00990. The project outcomes will be published in peer-reviewed and biomedical journals, scientific conferences and as a higher degree research thesis.

The Medicines Intelligence Centre of Research Excellence: Co-creating real-world evidence to support the evidentiary needs of Australian regulators and payers

Regulators and payers play a pivotal role in facilitating timely and affordable access to safe and efficacious medicines. They use evidence generated from randomised clinical trials (RCTs) to support decisions to register and subsidise medicines. However, at the time of registration and subsidy approval, regulators and payers face uncertainty about how RCT outcomes will translate to real-world clinical practice. In response to this situation, medicines policy agencies worldwide have endorsed the use of real-world data (RWD) to derive novel insights on the use and outcomes of prescribed medicines. Recent reforms around data availability and use in Australia are creating unparalleled data access and opportunities for Australian researchers to undertake large-scale research to generate evidence on the safety and effectiveness of medicines in the real world. Highlighting the critical importance of research in this area, Quality Use of Medicines and Medicine Safety was announced as Australia's 10th National Health Priority in 2019. The National Health and Medical Research Council, Medicines Intelligence Centre of Research Excellence (MI-CRE) has been formed to take advantage of the renewed focus on quality use of medicines and the changing data landscape in Australia. It will generate timely research supporting the evidentiary needs of Australian medicines regulators and payers by accelerating the development and translation of real-world evidence on medicines use and outcomes. MI-CRE is developing a coordinated approach to identify, triage and respond to priority questions where there are significant uncertainties about medicines use, (cost)-effectiveness, and/or safety and creating a data ecosystem that will streamline access to Australian data to enable researchers to generate robust evidence in a timely manner. This paper outlines how MI-CRE will partner with policy makers, clinicians, and consumer advocates to leverage real-world data to co-create real-world evidence, to improve quality use of medicines and reduce medicine-related harm.

The Medicines Intelligence Centre of Research Excellence: Co-creating real-world evidence to support the evidentiary needs of Australian medicines regulators and payers

Regulators and payers play a pivotal role in facilitating timely and affordable access to safe and efficacious medicines. They use evidence generated from randomised clinical trials (RCTs) to support decisions to register and subsidise medicines. However, at the time of registration and subsidy approval, regulators and payers face uncertainty about how RCT outcomes will translate to real-world clinical practice. In response to this situation, medicines policy agencies worldwide have endorsed the use of real-world data (RWD) to derive novel insights on the use and outcomes of prescribed medicines. Recent reforms around data availability and use in Australia are creating unparalleled data access and opportunities for Australian researchers to undertake large-scale research to generate evidence on the safety and effectiveness of medicines in the real world. Highlighting the critical importance of research in this area, Quality Use of Medicines and Medicine Safety was announced as Australia's 10th National Health Priority in 2019. The National Health and Medical Research Council, Medicines Intelligence Centre of Research Excellence (MI-CRE) has been formed to take advantage of the renewed focus on quality use of medicines and the changing data landscape in Australia. It will generate timely research supporting the evidentiary needs of Australian medicines regulators and payers by accelerating the development and translation of real-world evidence on medicines use and outcomes. MI-CRE is developing a coordinated approach to identify, triage and respond to priority questions where there are significant uncertainties about medicines use, (cost)-effectiveness, and/or safety and creating a data ecosystem that will streamline access to Australian data to enable researchers to generate robust evidence in a timely manner. This paper outlines how MI-CRE will partner with policy makers, clinicians, and consumer advocates to leverage real-world data to co-create real-world evidence, to improve quality use of medicines and reduce medicine-related harm.

Improving ambulance dispatch triage to trauma: A scoping review using the framework of development and evaluation of clinical prediction rules

OBJECTIVE

Ambulance dispatch algorithms should function as clinical prediction rules, identifying high acuity patients for advanced life support, and low acuity patients for non-urgent transport. Systematic reviews of dispatch algorithms are rare and focus on study types specific to the final phases of rule development, such as impact studies, and may miss the complete value-added evidence chain. We sought to summarise the literature for studies seeking to improve dispatch in trauma by performing a scoping review according to standard frameworks for developing and evaluating clinical prediction rules.

METHODS

We performed a scoping review searching MEDLINE, EMBASE, CINAHL, the CENTRAL trials registry, and grey literature from January 2005 to October 2021. We included all study types investigating dispatch triage to injured patients in the English language. We reported the clinical prediction rule phase (derivation, validation, impact analysis, or user acceptance) and the performance and outcomes measured for high and low acuity trauma patients.

RESULTS

Of 2067 papers screened, we identified 12 low and 30 high acuity studies. Derivation studies were most common (52%) and rule-based computer-aided dispatch was the most frequently investigated (23 studies). Impact studies rarely reported a prior validation phase, and few validation studies had their impact investigated. Common outcome measures in each phase were infrequent (0 to 27%), making a comparison between protocols difficult. A series of papers for low acuity patients and another for pediatric trauma followed clinical prediction rule development. Some low acuity Medical Priority Dispatch System codes are associated with the infrequent requirement for advanced life support and clinician review of computer-aided dispatch may enhance dispatch triage accuracy in studies of helicopter emergency medical services.

CONCLUSIONS

Few derivation and validation studies were followed by an impact study, indicating important gaps in the value-added evidence chain. While impact studies suggest clinician oversight may enhance dispatch, the opportunity exists to standardize outcomes, identify trauma-specific low acuity codes, and develop intelligent dispatch systems.

Extract, transform, load framework for the conversion of health databases to OMOP

Common data models standardize the structures and semantics of health datasets, enabling reproducibility and large-scale studies that leverage the data from multiple locations and settings. The Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) is one of the leading common data models. While there is a strong incentive to convert datasets to OMOP, the conversion is time and resource-intensive, leaving the research community in need of tools for mapping data to OMOP. We propose an extract, transform, load (ETL) framework that is metadata-driven and generic across source datasets. The ETL framework uses a new data manipulation language (DML) that organizes SQL snippets in YAML. Our framework includes a compiler that converts YAML files with mapping logic into an ETL script. Access to the ETL framework is available via a web application, allowing users to upload and edit YAML files via web editor and obtain an ETL SQL script for use in development environments. The structure of the DML maximizes readability, refactoring, and maintainability, while minimizing technical debt and standardizing the writing of ETL operations for mapping to OMOP. Our framework also supports transparency of the mapping process and reuse by different institutions.

The impact of re-opening the international border on COVID-19 hospitalisations in Australia: a modelling study

OBJECTIVE

To estimate the numbers of COVID-19-related hospitalisations in Australia after re-opening the international border.

DESIGN

Population-level deterministic compartmental epidemic modelling of eight scenarios applying various assumptions regarding SARS-CoV-2 transmissibility (baseline R0 = 3.5 or 7.0), vaccine rollout speed (slow or fast), and scale of border re-opening (mean of 2500 or 13 000 overseas arrivals per day).

SETTING

Simulation population size, age structure, and age-based contact rates based on recent estimates for the Australian population. We assumed that 80% vaccination coverage of people aged 16 years or more was reached in mid-October 2021 (fast rollout) or early January 2022 (slow rollout).

MAIN OUTCOME MEASURES

Numbers of people admitted to hospital with COVID-19, December 2021 - December 2022.

RESULTS

In scenarios assuming a highly transmissible SARS-CoV-2 variant (R0  = 7.0), opening the international border on either scale was followed by surges in both infections and hospitalisations that would require public health measures beyond mask wearing and social distancing to avoid overwhelming the health system. Reducing the number of hospitalisations to manageable levels required several cycles of additional social and mobility restrictions.

CONCLUSIONS

If highly transmissible SARS-CoV-2 variants are circulating locally or overseas, large and disruptive COVID-19 outbreaks will still be possible in Australia after 80% of people aged 16 years or more have been vaccinated. Continuing public health measures to restrict the spread of disease are likely to be necessary throughout 2022.

Predicting cardiovascular risk from national administrative databases using a combined survival analysis and deep learning approach

Background

Machine learning-based risk prediction models may outperform traditional statistical models in large datasets with many variables, by identifying both novel predictors and the complex interactions between them. This study compared deep learning extensions of survival analysis models with Cox proportional hazards models for predicting cardiovascular disease (CVD) risk in national health administrative datasets.

Methods

Using individual person linkage of administrative datasets, we constructed a cohort of all New Zealanders aged 30–74 who interacted with public health services during 2012. After excluding people with prior CVD, we developed sex-specific deep learning and Cox proportional hazards models to estimate the risk of CVD events within 5 years. Models were compared based on the proportion of explained variance, model calibration and discrimination, and hazard ratios for predictor variables.

Results

First CVD events occurred in 61 927 of 2 164 872 people. Within the reference group, the largest hazard ratios estimated by the deep learning models were for tobacco use in women (2.04, 95% CI: 1.99, 2.10) and chronic obstructive pulmonary disease with acute lower respiratory infection in men (1.56, 95% CI: 1.50, 1.62). Other identified predictors (e.g. hypertension, chest pain, diabetes) aligned with current knowledge about CVD risk factors. Deep learning outperformed Cox proportional hazards models on the basis of proportion of explained variance (R²: 0.468 vs 0.425 in women and 0.383 vs 0.348 in men), calibration and discrimination (all P <0.0001).

Conclusions

Deep learning extensions of survival analysis models can be applied to large health administrative datasets to derive interpretable CVD risk prediction equations that are more accurate than traditional Cox proportional hazards models.

Indigenous and Tribal Peoples Data Governance in Health Research: A Systematic Review

There is increasing potential to improve the research and reporting on the health and wellbeing of Indigenous and Tribal peoples through the collection and (re)use of population-level data. As the data economy grows and the value of data increases, the optimization of data pertaining to Indigenous peoples requires governance that defines who makes decisions on behalf of whom and how these data can and should be used. An international a priori PROSPERO (#CRD42020170033) systematic review was undertaken to examine the health research literature to (1) identify, describe, and synthesize definitions and principles; (2) identify and describe data governance frameworks; and (3) identify, describe, and synthesize processes, policies and practices used in Indigenous Data Governance (ID-GOV). Sixty-eight articles were included in the review that found five components that require consideration in the governance of health research data pertaining to Indigenous people. This included (1) Indigenous governance; (2) institutional ethics; (3) socio-political dynamics; (4) data management and data stewardship; and (5) overarching influences. This review provides the first systematic international review of ID-GOV that could potentially be used in a range of governance strategies moving forward in health research.

Burden of cardiovascular diseases in older adults using aged care services

OBJECTIVE

To quantify the burden of cardiovascular diseases (CVD) in older adults using community and residential care services.

METHODS

The study population comprised people aged 45+ from the 45 and Up Study (2006-09, n = 266,942) in Australia linked with records for hospital stays, aged care service and deaths for the period 2006-14. Follow-up time for each person was allocated to three categories of service use: no aged care, community care and residential care, with censoring at date of death. We calculated the prevalence at baseline and entry to aged care, and incidence rates for major CVD and six cardiovascular diagnoses, seven cardiovascular interventions (collectively CV interventions), cardiovascular-related intensive care unit stays and cardiovascular death.

RESULTS

The prevalence of major CVD at entry into community care and residential care was 41% and 58% respectively. Incidence per 1,000 person-years of all major CVD hospitalisations and CV interventions, respectively, was 182.8 (95% CI: 180.0-185.8) and 37.0 (95% CI: 35.6-38.4) for people using community care, and 280.7 (95% CI: 272.2-289.4) and 11.7 (95% CI: 9.8-13.9) for people using residential care. Similar trends were observed for each of the CVD diagnoses and interventions. Crude incidence rates for cardiovascular deaths per 1,000 person-years were 1.4 (95% CI: 1.3-1.5) in no aged care, 13.3 (95% CI: 12.6-14.1) in community care, and 149.7 (95% CI: 144.4-155.2) in residential care.

CONCLUSION

Our findings demonstrate the significant burden of CVD in people using both community-based and residential aged care services and highlights the importance of optimising cardiovascular care for older adults.

942Global Indigenous and Tribal Peoples Data Governance in Health Research: A systematic review

Background

Despite a broad range of research using Indigenous specific data, there is limited research to describe what constitutes data governance. To identify priorities and monitor progress in health, data is critical. Accurate collection and appropriate analysis of population level data is necessary in understanding population distributions of health and wellbeing. The aim of this review is to provide a comprehensive understanding of the current literature describing Indigenous Data Governance (IDG) processes in health research.

Methods

A comprehensive a-priori search strategy has been developed and submitted for registration through PROSPERO(APP170033). Literature will be sourced from bibliographic databases, review articles, key journals, conference proceedings, grey literature and referrals by key experts up until 01/2020 and synthesised (through meta-study and meta-aggregation approaches) in accordance with PRISMA guidelines.

Results

Once completed, a descriptive overview along with discussion on IDG processes in decision making specific to health research will be synthesized. IDG 'interventions' may include, but are not limited to, community advisory committees, Indigenous leadership, or institutional and/or project policies pertaining to decision making processes in the use of data in health research.

Conclusions

This review will provide evidence of definitions and procedures specific to IDG in the literature and how IDG, as a process, is operationalised in Indigenous health research across the globe.

Key messages

Preliminary Key messages include the lack of IDG processes described in the literature and the need for the development of guidelines to support researchers in operationalising Indigenous Data Sovereignty.

541A Nationwide Evaluation of International Standards and Commonly-used Growth Charts for Predicting Adverse Perinatal Outcomes

Background

Abnormal fetal growth is a risk factor for perinatal mortality and morbidity. There is considerable debate about the choice and performance of growth charts to classify newborns as small or large for gestational age (SGA and LGA) as a proxy for the at-risk infants. Several international charts have been proposed to be adopted worldwide. We aim to evaluate the performance of commonly-used growth charts (including international INTERGROWTH-21st-standards) for predicting adverse outcomes among SGA and LGA babies.

Methods

A population cohort of 2.4 million singleton births (24+0–40+6 weeks) delivered in Australia, 2004–2013. Performance was evaluated by prevalence, relative risk and diagnostic accuracy for adverse outcome based on AUC.

Results

There was wide variation in SGA and LGA classification across charts. For example, compared to other charts, the INTERGROWTH-21st-standards classified half of the number of term-SGA babies (prevalence: 3-4% vs. 7-10%) (&lt;10th-centile) and double the number of LGA babies (prevalence: 24-25% vs. 8-18%) (&gt;90th-centile), resulting in a smaller cohort of term-SGA at higher-risk of adverse outcome, and a larger LGA cohort with lower-risk of adverse outcome. All charts performed poorly for detecting adverse outcomes (AUC range for a composite outcome: 0.49-0.68) and across birthweight centiles.

Conclusions

Significant differences in the classification of newborns and the chart performance raises concerns about whether the INTERGROWTH-21st-standards are applicable to a multi-ethnic population such as Australia.

Key messages

Significant differences in the classification of newborns and the relatively poor predictive ability of growth charts means that over reliance on infant size alone may misclassify, and thus miss at-risk infants.

1467Interaction effects of multimorbidity and frailty on adverse health outcomes in elderly hospitalised patients

Background

Multimorbidity and frailty are each known to be drivers of adverse outcomes, but their joint effects are less well known.

Methods

We used state-wide hospital and mortality data for persons aged ≥75 years, admitted during 2010-2012 in New South Wales, Australia, to ascertain multimorbidity, frailty risk and outcomes: prolonged length of stay (LOS), 30-day mortality and 30-day emergency readmission. We estimated the relative risk (RR) of each outcome using Poisson models with random intercept for hospital. Interactions on both additive and multiplicative scales were examined.

Results

Among 257,535 elderly inpatients, 33.6% had multimorbidity and elevated frailty risk, 14.7% had multimorbidity only, 19.9% had elevated frailty risk only and 31.8% had neither. Additive interactions were present for all outcomes, with a further multiplicative interaction for mortality and LOS. Mortality risk was 4.2 (95% CI 4.1 – 4.4), prolonged LOS 3.3 (95% CI 3.3 – 3.4) and readmission 1.8 (95% CI 1.7 – 1.9) times higher in patients with both factors present compared with patients with neither.

Conclusions

Multimorbidity and frailty coexist in older hospitalized patients and interact to increase the risk of adverse outcomes beyond the sum of their individual effects. Their joint effect should be considered in health outcomes research and when administering hospital resources.

Key messages

Multimorbidity and frailty coexist and interact to increase the risk of adverse outcomes in hospitalised patients. Research and health planners should consider their joint effects to ensure patients at the highest risk of adverse outcomes are identified.

Performance of six birth-weight and estimated-fetal-weight standards for predicting adverse perinatal outcome: a 10-year nationwide population-based study

OBJECTIVE

To evaluate three birth-weight (BW) standards (Australian population-based, Fenton and INTERGROWTH-21^st ) and three estimated-fetal-weight (EFW) standards (Hadlock, INTERGROWTH-21^st and WHO) for classifying small-for-gestational age (SGA) and large-for-gestational age (LGA) and predicting adverse perinatal outcomes in preterm and term babies.

METHODS

This was a nationwide population-based study conducted on a total of 2.4 million singleton births that occurred from 24 + 0 to 40 + 6 weeks' gestation between 2004 and 2013 in Australia. The performance of the growth charts was evaluated according to SGA and LGA classification, and relative risk (RR) and diagnostic accuracy based on the areas under the receiver-operating-characteristics curves (AUCs) for stillbirth, neonatal death, perinatal death, composite morbidity and a composite of perinatal death and morbidity outcomes. The analysis was stratified according to gestational age at delivery (< 37 + 0 vs ≥ 37 + 0 weeks).

RESULTS

Following exclusions, 2 392 782 singleton births were analyzed. There were significant differences in the SGA and LGA classification and risk of adverse outcomes between the six BW and EFW standards evaluated. For the term group, compared with the other standards, the INTERGROWTH-21^st BW and EFW standards classified half the number of SGA (< 10^th centile) babies (3-4% vs 7-11%) and twice the number of LGA (> 90^th centile) babies (24-25% vs 8-15%), resulting in a smaller cohort of term SGA at higher risk of adverse outcome and a larger LGA cohort at lower risk of adverse outcome. For term SGA (< 3^rd centile) babies, the RR of perinatal death using the two INTERGROWTH-21^st standards was up to 1.5-fold higher than those of the other standards (including the WHO-EFW and Hadlock-EFW), while the INTERGROWTH-21^st -EFW standard indicated a 12-26% reduced risk of perinatal death for LGA cases across centile thresholds. Conversely, for the preterm group, the WHO-EFW and Hadlock-EFW standards identified a higher SGA classification rate than did the other standards (18-19% vs 10-11%) and a 20-65% increased risk of perinatal death in term LGA babies. All BW and EFW charts had similarly poor performance in predicting adverse outcomes, including the composite outcome (AUC range, 0.49-0.62) for both preterm (AUC range, 0.58-0.62) and term (AUC range, 0.49-0.50) cases and across centiles. Furthermore, specific centile thresholds for identifying adverse outcomes varied markedly by chart between BW and EFW standards.

CONCLUSIONS

This study addresses the recurrent problem of identifying fetuses at risk of morbidity and perinatal mortality associated with growth disorders and provides new insights into the applicability of international growth standards. Our findings of marked variation in classification and the similarly poor performance of prescriptive international standards and the other commonly used standards raise questions about whether the prescriptive international standards that were constructed for universal adoption are indeed applicable to a multiethnic population such as that of Australia. Thus, caution is needed when adopting universal standards for clinical and epidemiological use. © 2020 International Society of Ultrasound in Obstetrics and Gynecology.

Vaccinating Australia: How long will it take?

The Australian Government began to roll out the national COVID-19 vaccination program in late February 2021, with the initial aim to vaccinate the Australian adult population by the end of October 2021. The task of vaccinating some 20 million people presents considerable logistic challenges, but a rapid rollout is essential to allow for the reopening of borders and is especially urgent as new more transmissible variants arise. Here, we run a series of projections to estimate how long it will take to vaccinate the Australian population under different assumptions about the rate of vaccine administration, the schedule for vaccine eligibility and prevalence of vaccine hesitancy. Our analysis highlights the number of vaccine doses that can be administered per day as the key factor determining the duration of the vaccine rollout. A rate of 200,000 doses per day would achieve 90% population coverage by the end of 2021; 80,000 doses a day would see the rollout extended until mid-2023. Vaccine hesitancy has the potential to greatly slow down the rollout and becomes the main limiting factor when the supply of vaccine doses is high. Speed is of the essence when it comes vaccinating populations against COVID-19: a rapid rollout will minimise the risk of sporadic and costly lockdowns and the potential for small, local clusters getting out of control and sparking new epidemic waves. In order to achieve rapid population coverage, the Australian government must ramp up vaccine administration to at least 200,000 doses per day as quickly as possible, while also promoting vaccine willingness in the community through clear public health messaging, especially to known hesitant demographics.

A Versatile Big Data Health System for Australia: Driving Improvements in Cardiovascular Health

Cardiovascular diseases (CVD) are leading causes of death and morbidity in Australia and worldwide. Despite improvements in treatment, there remain large gaps in our understanding to prevent, treat and manage CVD events and associated morbidities. This article lays out a vision for enhancing CVD research in Australia through the development of a Big Data system, bringing together the multitude of rich administrative and health datasets available. The article describes the different types of Big Data available for CVD research in Australia and presents an overview of the potential benefits of a Big Data system for CVD research and some of the major challenges in establishing the system for Australia. The steps for progressing this vision are outlined.

Incorporating real-world evidence into the development of patient blood glucose prediction algorithms for the ICU

Objective

Glycemic control is an important component of critical care. We present a data-driven method for predicting intensive care unit (ICU) patient response to glycemic control protocols while accounting for patient heterogeneity and variations in care.

Materials and Methods

Using electronic medical records (EMRs) of 18 961 ICU admissions from the MIMIC-III dataset, including 318 574 blood glucose measurements, we train and validate a gradient boosted tree machine learning (ML) algorithm to forecast patient blood glucose and a 95% prediction interval at 2-hour intervals. The model uses as inputs irregular multivariate time series data relating to recent in-patient medical history and glycemic control, including previous blood glucose, nutrition, and insulin dosing.

Results

Our forecasting model using routinely collected EMRs achieves performance comparable to previous models developed in planned research studies using continuous blood glucose monitoring. Model error, expressed as mean absolute percentage error is 16.5%–16.8%, with Clarke error grid analysis demonstrating that 97% of predictions would be clinically acceptable. The 95% prediction intervals achieve near intended coverage at 93%–94%.

Discussion

ML algorithms built on observational data sources, such as EMRs, present a promising approach for personalization and automation of glycemic control in critical care. Future research may benefit from applying a combination of methodologies and data sources to develop robust methodologies that account for the variations seen in ICU patients and difficultly in detecting the extremes of observed blood glucose values.

Conclusion

We demonstrate that EMRs can be used to train ML algorithms that may be suitable for incorporation into ICU decision support systems.

Identifying preventable risk factors for hospitalised asthma in young Aboriginal children: a whole-population cohort study

BACKGROUND

Australia has one of the highest rates of asthma worldwide. Indigenous children have a particularly high burden of risk determinants for asthma, yet little is known about the asthma risk profile in this population.

AIM

To identify and quantify potentially preventable risk factors for hospitalised asthma in Australian Aboriginal children (1-4 years of age).

METHODS

Birth, hospital and emergency data for all Aboriginal children born 2003-2012 in Western Australia were linked (n=32 333). Asthma was identified from hospitalisation codes. ORs and population attributable fractions were calculated for maternal age at birth, remoteness, area-level disadvantage, prematurity, low birth weight, maternal smoking in pregnancy, mode of delivery, maternal trauma and hospitalisations for acute respiratory tract infection (ARTI) in the first year of life.

RESULTS

There were 705 (2.7%) children hospitalised at least once for asthma. Risk factors associated with asthma included: being hospitalised for an ARTI (OR 4.06, 95% CI 3.44 to 4.78), area-level disadvantage (OR 1.58, 95% CI 1.28 to 1.94), being born at <33 weeks' gestation (OR 3.30, 95% CI 2.52 to 4.32) or birth weight <1500 g (OR 2.35, 95% CI 1.39 to 3.99). The proportion of asthma attributable to an ARTI was 31%, area-level disadvantage 18%, maternal smoking 5%, and low gestational age and birth weight were 3%-7%. We did not observe a higher risk of asthma in those children who were from remote areas.

CONCLUSION

Improving care for pregnant Aboriginal women as well as for Aboriginal infants with ARTI may help reduce the burden of asthma in the Indigenous population.