DEMoS: a deep learning-based ensemble approach for predicting the molecular subtypes of gastric adenocarcinomas from histopathological images.

MOTIVATION: The molecular subtyping of gastric cancer (adenocarcinoma) into four main subtypes based on integrated multiomics profiles, as proposed by The Cancer Genome Atlas (TCGA) initiative, represents an effective strategy for patient stratification. However, this approach requires the use of multiple technological platforms, and is quite expensive and time-consuming to perform. A computational approach that uses histopathological image data to infer molecular subtypes could be a practical, cost- and time-efficient complementary tool for prognostic and clinical management purposes. RESULTS: Here, we propose a deep learning ensemble approach (called DEMoS) capable of predicting the four recognized molecular subtypes of gastric cancer directly from histopathological images. DEMoS achieved tile-level area under the receiver-operating characteristic curve (AUROC) values of 0.785, 0.668, 0.762 and 0.811 for the prediction of these four subtypes of gastric cancer [i.e. (i) Epstein-Barr (EBV)-infected, (ii) microsatellite instability (MSI), (iii) genomically stable (GS) and (iv) chromosomally unstable tumors (CIN)] using an independent test dataset, respectively. At the patient-level, it achieved AUROC values of 0.897, 0.764, 0.890 and 0.898, respectively. Thus, these four subtypes are well-predicted by DEMoS. Benchmarking experiments further suggest that DEMoS is able to achieve an improved classification performance for image-based subtyping and prevent model overfitting. This study highlights the feasibility of using a deep learning ensemble-based method to rapidly and reliably subtype gastric cancer (adenocarcinoma) solely using features from histopathological images. AVAILABILITY AND IMPLEMENTATION: All whole slide images used in this study was collected from the TCGA database. This study builds upon our previously published HEAL framework, with related documentation and tutorials available at http://heal.erc.monash.edu.au. The source code and related models are freely accessible at https://github.com/Docurdt/DEMoS.git. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Spectrum of illness among returned Australian travellers from Bali, Indonesia: a 5-year retrospective observational study.

BACKGROUND: Bali, Indonesia, presents significant infectious and non-infectious health risks for Australian travellers. Understanding this spectrum of illnesses has the potential to assist clinicians in evaluating unwell returning travellers and guide provision of pre-travel advice. AIM: To describe the spectrum of illnesses seen in returned travellers from Bali. METHODS: Using a novel text mining approach, we performed a retrospective, observational study of all adult emergency department presentations to a metropolitan health service in Melbourne, from 2011 to 2015. Outcome measures included demographic, clinical and laboratory features of travel-related illnesses. RESULTS: A total of 464 patients met inclusion criteria. Gastroenteritis (119/464, 26%), systemic febrile illness (88/464, 19%) and respiratory tract infection (51/464, 11%) were the most common diagnoses. Dengue was the most common laboratory-confirmed diagnosis (25/464, 5%). No cases of malaria were identified. Common non-infectious presentations included traumatic injury (47/464, 10%) and animal bites requiring rabies post-exposure prophylaxis (29/464, 6%). A total of 110 patients (24%) was admitted to the hospital; those presenting with systemic febrile illness were more likely to be admitted compared to those presenting with other illnesses (odds ratio 3.42, 95% confidence interval 2.02-5.75, P < 0.001). CONCLUSION: This is the first study to use a text mining approach to identify and describe emergency department presentations related to diseases acquired in Bali by Australian travellers. Although infections are important causes of illness, trauma and animal bites account for a significant number of hospital presentations. Our findings contribute to the knowledge on the health risks for travellers to Bali, and will assist clinicians in relevant pre- and post-travel evaluations.

Evaluation of the impact of implementation of a Medical Assessment and Planning Unit on length of stay.

OBJECTIVE: The Medical Assessment and Planning Unit (MAPU) model provides a multidisciplinary and 'front end loading' approach to acute medical care. The objective of this study was to evaluate the impact of a 10-bed MAPU in Royal Melbourne Hospital (RMH) on hospital length of stay. METHODS: A pre-post study design was used. Cases were defined as all general medical patients admitted to the RMH between 1 August 2003 and 31 January 2004. MAPU patients were defined as general medical patients who had been discharged from RMH MAPU unit as part of their RMH inpatient admission. Historical controls were defined as all general medical patients admitted to the RMH between 1 August 2002 and 31 January 2003. RESULTS: There was a reduction in median length of stay that did not reach statistical significance. During the study period, median emergency department length of stay for MAPU patients was 10.3 h compared with 13.2 h for non-MAPU patients who were admitted directly to general wards. CONCLUSIONS: The reductions in length of stay are likely to be of clinical significance at the emergency department (ED) level. The MAPU model also contributes to providing care appropriate care for older admitted patients.

A case study of centralised monitoring of hospital access performance.

Access to care for patients remains a concern for all parties in the provision of hospital services. It is the subject of patient complaints, large investments of funds and vigorous debate in the community, hospitals and the political arena. This is a common problem in developed nations. There has been little achievement in information technology solutions to this significant problem in Australia. This paper presents a case study of the development and implementation of an organisational access display system intended to provide real-time, or near to real-time information and feedback on access for staff on the floor. This is believed to be one of the first times such a development has been reported in the Australian literature, albeit limited to the context of a single organisation.

A framework to provide clarity in the use of predictive technologies in health care management.

This paper reports on a survey of health care managers and other stakeholders which assesses the need for a framework regarding predictive technologies in health care management. In the context of this paper, predictive technologies are defined as those that enable an insight into, or measurement of, events yet to occur. A framework could include the ability to classify the problems confronting managers, and the range of possible tools and techniques that could be used to address those problems. This could be of mutual benefit to health care managers, technologists and modellers. The survey was intended to clarify the level of interest in such a framework, and also the possible dimensions that it ought to contain. Our results indicate that there is strong support for a proposed framework, with 97% of respondents indicating that a framework would be possibly or very useful. The results also show a low level of background knowledge in relation to existing tools, techniques and technologies. The draft framework is also presented. It includes dimensions relating to problem and tool definitions, scenarios to be investigated and the findings of those investigations.

Myths of ideal hospital occupancy.

Significant problems in health care, such as access block and long waiting lists for elective surgery, have led to calls for keeping hospital occupancy at no more than 85%. It is elementary queueing theory that a finite-capacity system with variable demand cannot sustain both full utilisation and full availability. However, the statement that there is a single level of ideal or safe occupancy suitable for all situations is a simplistic interpretation and application of the underlying science. We argue that specific study and action are necessary to understand and deal with the problems of long waiting lists and access block in any given health care facility.

A technology ecosystem perspective on hospital management information systems: lessons from the health literature.

Hospital managers have a large range of information needs including quality metrics, financial reports, access information needs, educational, resourcing and decision support needs. Currently these needs involve interactions by managers with numerous disparate systems, both electronic such as SAP, Oracle Financials, PAS' (patient administration systems) like HOMER, and relevant websites; and paper-based systems. Hospital management information systems (HMIS) can be thought of sitting within a Technology Ecosystem (TE). In addition, Hospital Management Information Systems (HMIS) could benefit from a broader and deeper TE model, and the HMIS environment may in fact represents its own TE (the HMTE). This research will examine lessons from the health literature in relation to some of these issues, and propose an extension to the base model of a TE.

Short-term forecasting of emergency inpatient flow.

Hospital managers have to manage resources effectively, while maintaining a high quality of care. For hospitals where admissions from the emergency department to the wards represent a large proportion of admissions, the ability to forecast these admissions and the resultant ward occupancy is especially useful for resource planning purposes. Since emergency admissions often compete with planned elective admissions, modeling emergency demand may result in improved elective planning as well. We compare several models for forecasting daily emergency inpatient admissions and occupancy. The models are applied to three years of daily data. By measuring their mean square error in a cross-validation framework, we find that emergency admissions are largely random, and hence, unpredictable, whereas emergency occupancy can be forecasted using a model combining regression and autoregressive integrated moving average (ARIMA) model, or a seasonal ARIMA model, for up to one week ahead. Faced with variable admissions and occupancy, hospitals must prepare a reserve capacity of beds and staff. Our approach allows estimation of the required reserve capacity.

Measurement for improvement: a survey of current practice in Australian public hospitals.

OBJECTIVE: To identify patient safety measurement tools in use in Australian public hospitals and to determine barriers to their use. DESIGN: Structured survey, conducted between 4 March and 19 May 2005, designed to identify tools, and to assess current use of, levels of satisfaction with, and barriers to use of tools for measuring the domains and subdomains of: organisational capacity to provide safe health care; patient safety incidents; and clinical performance. PARTICIPANTS AND SETTING: Hospital executives, managers and clinicians from a nationwide random sample of Australian public hospitals stratified by state and hospital peer grouping. MAIN OUTCOME MEASURES: Tools used by hospitals within the three domains and their subdomains; patient safety tools and processes identified by individuals at these hospitals; satisfaction with the tools; and barriers to their use. RESULTS: Eighty-two of 167 invited hospitals (49%) responded. The survey ascertained a comprehensive list of patient safety measurement tools that are in current use for measuring all patient safety domains. Overall, there was a focus on use of processes rather than quantitative measurement tools. Approximately half the 182 individual respondents from participating hospitals reported satisfaction with existing tools. The main reported barriers were lack of integrated supportive systems, resource constraints and inadequate access to robust measurement tools validated in the Australian context. Measurement of organisational capacity was reported by 50 (61%), of patient safety incidents by 81 (99%) and of clinical performance by 81 (99%). CONCLUSION: Australian public hospitals are measuring the safety of their health care, with some variation in measurement of patient safety domains and their subdomains. Improved access to robust tools may support future standardisation of measurement for improvement.