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.

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.

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.

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.

Complex early childhood experiences: Characteristics of Northern Territory children across health, education and child protection data.

Early identification of vulnerable children to protect them from harm and support them in achieving their long-term potential is a community priority. This is particularly important in the Northern Territory (NT) of Australia, where Aboriginal children are about 40% of all children, and for whom the trauma and disadvantage experienced by Aboriginal Australians has ongoing intergenerational impacts. Given that shared social determinants influence child outcomes across the domains of health, education and welfare, there is growing interest in collaborative interventions that simultaneously respond to outcomes in all domains. There is increasing recognition that many children receive services from multiple NT government agencies, however there is limited understanding of the pattern and scale of overlap of these services. In this paper, NT health, education, child protection and perinatal datasets have been linked for the first time. The records of 8,267 children born in the NT in 2006-2009 were analysed using a person-centred analytic approach. Unsupervised machine learning techniques were used to discover clusters of NT children who experience different patterns of risk. Modelling revealed four or five distinct clusters including a cluster of children who are predominantly ill and experience some neglect, a cluster who predominantly experience abuse and a cluster who predominantly experience neglect. These three, high risk clusters all have low school attendance and together comprise 10-15% of the population. There is a large group of thriving children, with low health needs, high school attendance and low CPS contact. Finally, an unexpected cluster is a modestly sized group of non-attendees, mostly Aboriginal children, who have low school attendance but are otherwise thriving. The high risk groups experience vulnerability in all three domains of health, education and child protection, supporting the need for a flexible, rather than strictly differentiated response. Interagency cooperation would be valuable to provide a suitably collective and coordinated response for the most vulnerable children.

Uptake of Team Care Arrangements for adults newly diagnosed with cancer.

BACKGROUND: Medicare-subsidised Team Care Arrangements (TCAs) support Australian general practitioners to implement shared care between collaborating health professionals for patients with chronic medical conditions and complex needs. We assessed the prevalence of TCAs, factors associated with TCA uptake and visits to TCA-subsidised allied health practitioners, for adults newly diagnosed with cancer in New South Wales, Australia. METHODS: We carried out a retrospective individual patient data linkage study with 13 951 45 and Up Study participants diagnosed with incident cancer during 2006-16. We used a proportional hazards model to estimate the factors associated with receipt of a TCA after cancer diagnosis. RESULTS: In total, 6630 patients had a TCA plan initiated (47.5%). A TCA was more likely for patients aged ≥65years, those with higher service utilisation 4-15months prior to cancer diagnosis, a higher number of comorbidities, lower self-rated overall health status, living in areas of greater socio-economic disadvantage, lower educational attainment and those with no private health insurance. A total of 4084 (61.6%) patients with a TCA had at least one TCA-subsidised allied health visit within 24months of the TCA. CONCLUSIONS: TCAs appear to be well targeted at cancer patients with chronic health conditions and lower socioeconomic status. Nevertheless, not all patients with a TCA subsequently attended a TCA-subsidised allied healthcare professional. This suggests either a misunderstanding of the plan, the receipt of allied health via other public schemes, a low prioritisation of the plan compared to other health care, or suboptimal availability of these services.

De-identifying Australian hospital discharge summaries: An end-to-end framework using ensemble of deep learning models.

Electronic Medical Records (EMRs) contain clinical narrative text that is of great potential value to medical researchers. However, this information is mixed with Personally Identifiable Information (PII) that presents risks to patient and clinician confidentiality. This paper presents an end-to-end de-identification framework to automatically remove PII from Australian hospital discharge summaries. Our corpus included 600 hospital discharge summaries which were extracted from the EMRs of two principal referral hospitals in Sydney, Australia. Our end-to-end de-identification framework consists of three components: (1) Annotation: labelling of PII in the 600 hospital discharge summaries using five pre-defined categories: person, address, date of birth, individual identification number, phone/fax number; (2) Modelling: training six named entity recognition (NER) deep learning base-models on balanced and imbalanced datasets; and evaluating ensembles that combine all six base-models, the three base-models with the best F1 scores and the three base-models with the best recall scores respectively, using token-level majority voting and stacking methods; and (3) De-identification: removing PII from the hospital discharge summaries. Our results showed that the ensemble model combined using the stacking Support Vector Machine (SVM) method on the three base-models with the best F1 scores achieved excellent results with a F1 score of 99.16% on the test set of our corpus. We also evaluated the robustness of our modelling component on the 2014 i2b2 de-identification dataset. Our ensemble model, which uses the token-level majority voting method on all six base-models, achieved the highest F1 score of 96.24% at strict entity matching and the highest F1 score of 98.64% at binary token-level matching compared to two state-of-the-art methods. The end-to-end framework provides a robust solution to de-identifying clinical narrative corpuses safely. It can easily be applied to any kind of clinical narrative documents.

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.

A Machine Learning Approach to Predict the Added-Sugar Content of Packaged Foods.

BACKGROUND: Dietary guidelines recommend limiting the intake of added sugars. However, despite the public health importance, most countries have not mandated the labeling of added-sugar content on packaged foods and beverages, making it difficult for consumers to avoid products with added sugar, and limiting the ability of policymakers to identify priority products for intervention. OBJECTIVE: The aim was to develop a machine learning approach for the prediction of added-sugar content in packaged products using available nutrient, ingredient, and food category information. METHODS: The added-sugar prediction algorithm was developed using k-nearest neighbors (KNN) and packaged food information from the US Label Insight dataset (n = 70,522). A synthetic dataset of Australian packaged products (n = 500) was used to assess validity and generalization. Performance metrics included the coefficient of determination (R2), mean absolute error (MAE), and Spearman rank correlation (ρ). To benchmark the KNN approach, the KNN approach was compared with an existing added-sugar prediction approach that relies on a series of manual steps. RESULTS: Compared with the existing added-sugar prediction approach, the KNN approach was similarly apt at explaining variation in added-sugar content (R2 = 0.96 vs. 0.97, respectively) and ranking products from highest to lowest in added-sugar content (ρ = 0.91 vs. 0.93, respectively), while less apt at minimizing absolute deviations between predicted and true values (MAE = 1.68 g vs. 1.26 g per 100 g or 100 mL, respectively). CONCLUSIONS: KNN can be used to predict added-sugar content in packaged products with a high degree of validity. Being automated, KNN can easily be applied to large datasets. Such predicted added-sugar levels can be used to monitor the food supply and inform interventions aimed at reducing added-sugar intake.

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.

Medications used disproportionately during pregnancy: Priorities for research on the risks and benefits of medications when used during pregnancy.

PURPOSE: To identify medications used disproportionately more or less among pregnant women relative to women of childbearing age. METHODS: Medication use among pregnant women in New South Wales, Australia was identified using linked perinatal and pharmaceutical dispensing data from 2006 to 2012. Medication use in women of childbearing age (including pregnant women) was identified using pharmaceutical dispensing data for a 10% random sample of the Australian population. Pregnant social security beneficiaries (n = 111 612) were age-matched (1:3) to female social security beneficiaries in the 10% sample. For each medication, the risk it was dispensed during pregnancy relative to being dispensed during an equivalent time period among matched controls was computed. Medications were mapped to Australian pregnancy risk categories. RESULTS: Of the 181 included medications, 35 were statistically significantly more commonly dispensed to pregnant women than control women. Of these, 23 are categorised as posing no increased risk to the foetus. Among medications suspected of causing harm or having insufficient safety data, the strongest associations were observed for hydralazine, ondansetron, dalteparin sodium and ranitidine. Use was less likely during pregnancy than control periods for 127 medications, with the strongest associations observed for hormonal contraceptives and progestogens. CONCLUSIONS: Most medications found to be used disproportionately more by pregnant women are indicated for pregnancy-related problems. A large number of medications were used disproportionately less among pregnant women, where avoidance of some of these medications may pose a greater risk of harm. For many other medications avoided during pregnancy, current data are insufficient to inform this risk-benefit assessment.

Diagnostic and health service pathways to diagnosis of cancer-registry notified cancer of unknown primary site (CUP).

BACKGROUND: Cancer of unknown primary (CUP) is a late-stage malignancy with poor prognosis, but we know little about what diagnostic tests and procedures people with CUP receive prior to diagnosis. The purpose of this study was to determine how health service utilisation prior to diagnosis for people with cancer-registry notified CUP differs from those notified with metastatic cancer of known primary. METHODS: We identified people with a cancer registry notification of CUP (n = 327) from the 45 and Up Study, a prospective cohort of 266,724 people ≥45 years in New South Wales, Australia, matched with up to three controls with a diagnosis of metastatic cancer of known primary (n = 977). Baseline questionnaire data were linked to population health data to identify all health service use, diagnostic tests, and procedures in the month of diagnosis and 3 months prior. We used conditional logistic regression to estimate adjusted odds ratios (ORs) and 95% confidence intervals (CIs). RESULTS: After adjusting for age and educational attainment, people with a cancer-registry notified CUP diagnosis were more likely to be an aged care resident (OR = 2.78, 95%CI 1.37-5.63), have an emergency department visit (OR = 1.65, 95%CI 1.23-2.21), serum tumor marker tests (OR = 1.51, 95%CI 1.12-2.04), or a cytology test without immunohistochemistry (OR = 2.01, 95%CI 1.47-2.76), and less likely to have a histopathology test without immunohistochemistry (OR = 0.43, 95%CI 0.31-0.59). Neither general practitioner, specialist, allied health practitioner or nurse consultations, hospitalisations, nor imaging procedures were associated with a CUP diagnosis. CONCLUSIONS: The health service and diagnostic pathway to diagnosis differs markedly for people notified with CUP compared to those with metastatic cancer of known primary. While these differences may indicate missed opportunities for earlier detection and appropriate management, for some patients they may be clinically appropriate.

Benchmarking Deep Learning Architectures for Predicting Readmission to the ICU and Describing Patients-at-Risk.

To compare different deep learning architectures for predicting the risk of readmission within 30 days of discharge from the intensive care unit (ICU). The interpretability of attention-based models is leveraged to describe patients-at-risk. Several deep learning architectures making use of attention mechanisms, recurrent layers, neural ordinary differential equations (ODEs), and medical concept embeddings with time-aware attention were trained using publicly available electronic medical record data (MIMIC-III) associated with 45,298 ICU stays for 33,150 patients. Bayesian inference was used to compute the posterior over weights of an attention-based model. Odds ratios associated with an increased risk of readmission were computed for static variables. Diagnoses, procedures, medications, and vital signs were ranked according to the associated risk of readmission. A recurrent neural network, with time dynamics of code embeddings computed by neural ODEs, achieved the highest average precision of 0.331 (AUROC: 0.739, F1-Score: 0.372). Predictive accuracy was comparable across neural network architectures. Groups of patients at risk included those suffering from infectious complications, with chronic or progressive conditions, and for whom standard medical care was not suitable. Attention-based networks may be preferable to recurrent networks if an interpretable model is required, at only marginal cost in predictive accuracy.

Predicting 7-day, 30-day and 60-day all-cause unplanned readmission: a case study of a Sydney hospital.

BACKGROUND: The identification of patients at high risk of unplanned readmission is an important component of discharge planning strategies aimed at preventing unwanted returns to hospital. The aim of this study was to investigate the factors associated with unplanned readmission in a Sydney hospital. We developed and compared validated readmission risk scores using routinely collected hospital data to predict 7-day, 30-day and 60-day all-cause unplanned readmission. METHODS: A combination of gradient boosted tree algorithms for variable selection and logistic regression models was used to build and validate readmission risk scores using medical records from 62,235 live discharges from a metropolitan hospital in Sydney, Australia. RESULTS: The scores had good calibration and fair discriminative performance with c-statistic of 0.71 for 7-day and for 30-day readmission, and 0.74 for 60-day. Previous history of healthcare utilization, urgency of the index admission, old age, comorbidities related to cancer, psychosis, and drug-abuse, abnormal pathology results at discharge, and being unmarried and a public patient were found to be important predictors in all models. Unplanned readmissions beyond 7 days were more strongly associated with longer hospital stays and older patients with higher number of comorbidities and higher use of acute care in the past year. CONCLUSIONS: This study demonstrates similar predictors and performance to previous risk scores of 30-day unplanned readmission. Shorter-term readmissions may have different causal pathways than 30-day readmission, and may, therefore, require different screening tools and interventions. This study also re-iterates the need to include more informative data elements to ensure the appropriateness of these risk scores in clinical practice.

Theoretical approaches of online social network interventions and implications for behavioral change: a systematic review.

RATIONAL, AIMS AND OBJECTIVES: The aim of this review was to identify general theoretical frameworks used in online social network interventions for behavioral change. To address this research question, a PRISMA-compliant systematic review was conducted. METHODS: A systematic review (PROSPERO registration number CRD42014007555) was conducted using 3 electronic databases (PsycINFO, Pubmed, and Embase). Four reviewers screened 1788 abstracts. RESULTS: 15 studies were selected according to the eligibility criteria. Randomized controlled trials and controlled studies were assessed using Cochrane Collaboration's "risk-of-bias" tool, and narrative synthesis. Five eligible articles used the social cognitive theory as a framework to develop interventions targeting behavioral change. Other theoretical frameworks were related to the dynamics of social networks, intention models, and community engagement theories. Only one of the studies selected in the review mentioned a well-known theory from the field of health psychology. CONCLUSION: Conclusions were that guidelines are lacking in the design of online social network interventions for behavioral change. Existing theories and models from health psychology that are traditionally used for in situ behavioral change should be considered when designing online social network interventions in a health care setting.

Real-time prediction of mortality, readmission, and length of stay using electronic health record data.

OBJECTIVE: To develop a predictive model for real-time predictions of length of stay, mortality, and readmission for hospitalized patients using electronic health records (EHRs). MATERIALS AND METHODS: A Bayesian Network model was built to estimate the probability of a hospitalized patient being "at home," in the hospital, or dead for each of the next 7 days. The network utilizes patient-specific administrative and laboratory data and is updated each time a new pathology test result becomes available. Electronic health records from 32 634 patients admitted to a Sydney metropolitan hospital via the emergency department from July 2008 through December 2011 were used. The model was tested on 2011 data and trained on the data of earlier years. RESULTS: The model achieved an average daily accuracy of 80% and area under the receiving operating characteristic curve (AUROC) of 0.82. The model's predictive ability was highest within 24 hours from prediction (AUROC = 0.83) and decreased slightly with time. Death was the most predictable outcome with a daily average accuracy of 93% and AUROC of 0.84. DISCUSSION: We developed the first non-disease-specific model that simultaneously predicts remaining days of hospitalization, death, and readmission as part of the same outcome. By providing a future daily probability for each outcome class, we enable the visualization of future patient trajectories. Among these, it is possible to identify trajectories indicating expected discharge, expected continuing hospitalization, expected death, and possible readmission. CONCLUSIONS: Bayesian Networks can model EHRs to provide real-time forecasts for patient outcomes, which provide richer information than traditional independent point predictions of length of stay, death, or readmission, and can thus better support decision making.

The management of severe hypertension in Australian general practice.

BACKGROUND: Severe hypertension (SHT) (Blood Pressure, BP ≥ 180/110 mmHg) is associated with considerable morbidity and mortality, yet little is known about how it is managed. The purpose of this study is to examine the management of SHT by Australian general practitioners (GPs) and to explore its variance across patient characteristics and clinical practices. METHODS: Review of electronic medical records for a year before and after a recorded measure of SHT in 7,499 patients by 436 GPs in 167 clinics throughout Australia during 2008-2009. Outcome measures included follow-up, referral, changes to antihypertensive drug treatment, and BP control (normotensive reading, BP < 140/90 mmHg, and whether subsequent recorded measures were also in the normal range--sustained normotension). RESULTS: Of 7,499 patients with an electronic BP record of SHT, 94% were followed up (median time 14 days); 8% were referred to an appropriate specialist (median time 89 days--2% within 7 days) and 86% were managed by GPs. GPs initiated or changed antihypertensive drugs in 5,398 patients (72% of cohort); of these, 46% remained hypertensive (4% with SHT) and 7% achieved sustained normotension; 6% had no further electronic BP records. The remaining 14% had no medication changes; among these, 43% remained hypertensive (5% with SHT) and 3% achieved sustained normotension; 32% had no further electronic BP records. Some outcome measures displayed a variance across GP clinics that was mostly unexplained by patient or practice characteristics. CONCLUSIONS: Most patients with SHT had at least one follow-up visit and 72% had initiation of, or changes to, antihypertensive drug treatment. Although most of the patients experienced some improvement, blood pressure control was poor. Some clinics showed better performance. Suggestions are made for the development of clinical standards to facilitate appropriate management of this dangerous condition.

Exploring the role of pathology test results in the prediction of remaining days of hospitalisation.

Accurate prediction of discharge time and identification of patients at risk of extended length of stay (LOS) can facilitate discharge planning and positively impact both the patient and the hospital in a variety of ways. To date, however, most studies only focus on the prediction of the overall LOS, which is generally estimated at admission time to hospital, emergency department or intensive care unit. This paper explores whether individual laboratory results can improve predictions of time of discharge as the tests become available. This study suggests that there is a statistically significant relationship between individual test results and remaining days in hospital and that there is a trend towards better estimates as more consecutive tests are taken into consideration. Their effect on the estimate of discharge time is generally weak. Further work integrating groups of test results into a more sophisticated dynamical model is required.