Equitable Machine Learning for Hypoglycaemia Risk Management.

We developed a machine learning (ML) model for the detection of patients with high risk of hypoglycaemic events during their hospital stay to improve the detection and management of hypoglycaemia. Our model was trained on data from a regional local health care district in Australia. The model was found to have good predictive performance in the general case (AUC 0.837). We conducted subgroup analysis to ensure that the model performed in a way that did not disadvantage population subgroups, in this case based on gender or indigenous status. We found that our specific problem domain assisted us in reducing unwanted bias within the model, because it did not rely on practice patterns or subjective judgements for the outcome measure. With careful analysis for equity there is great potential for ML models to automate the detection of high-risk cohorts and automate mitigation strategies to reduce preventable errors.

How digital health translational research is prioritised: a qualitative stakeholder-driven approach to decision support evaluation.

OBJECTIVES: Digital health is now routinely being applied in clinical care, and with a variety of clinician-facing systems available, healthcare organisations are increasingly required to make decisions about technology implementation and evaluation. However, few studies have examined how digital health research is prioritised, particularly research focused on clinician-facing decision support systems. This study aimed to identify criteria for prioritising digital health research, examine how these differ from criteria for prioritising traditional health research and determine priority decision support use cases for a collaborative implementation research programme. METHODS: Drawing on an interpretive listening model for priority setting and a stakeholder-driven approach, our prioritisation process involved stakeholder identification, eliciting decision support use case priorities from stakeholders, generating initial use case priorities and finalising preferred use cases based on consultations. In this qualitative study, online focus group session(s) were held with stakeholders, audiorecorded, transcribed and analysed thematically. RESULTS: Fifteen participants attended the online priority setting sessions. Criteria for prioritising digital health research fell into three themes, namely: public health benefit, health system-level factors and research process and feasibility. We identified criteria unique to digital health research as the availability of suitable governance frameworks, candidate technology's alignment with other technologies in use,and the possibility of data-driven insights from health technology data. The final selected use cases were remote monitoring of patients with pulmonary conditions, sepsis detection and automated breast screening. CONCLUSION: The criteria for determining digital health research priority areas are more nuanced than that of traditional health condition focused research and can neither be viewed solely through a clinical lens nor technological lens. As digital health research relies heavily on health technology implementation, digital health prioritisation criteria comprised enablers of successful technology implementation. Our prioritisation process could be applied to other settings and collaborative projects where research institutions partner with healthcare delivery organisations.

Clinical decision support foundations.

This chapter gives an educational overview of: * The elements of a clinical decision; * The elements of decision making: prior probability, evidence (likelihood), posterior probability, actions, utility (value); * A framework for decision making, and support, encompassing validity, utility, importance and certainty; and * The required elements of a clinical decision support system. * The role of knowledge management in the construction and maintenance of clinical decision support.

Clinical decision support implementations.

This chapter gives an educational overview of: * Evidence for the benefits of CDSS; * Categories of CDSS including user and workflow requirements; * A framework for the implementation of CDSS, including human factors and the problem of free text; * a framework for the evaluation of CDSS.

Evidence-based risk factors for postoperative deep vein thrombosis.

BACKGROUND: Deep vein thrombosis (DVT) is a common postoperative complication that is associated with significant morbidity and mortality. Thromboprophylaxis has been shown to be underused. In the absence of prophylaxis, rates as high as 50% have been reported following orthopaedic surgery, and 25% following general surgery. Many risk factors have been suggested but there is often little evidence to support these claims. METHODS: A systematic review was performed to determine the evidence base behind each suggested risk factor, and, where sufficient data were available, a random-effects meta-analysis was performed. RESULTS: There is evidence to support a significant association between increased age, obesity, a past history of thromboembolism, varicose veins, the oral contraceptive pill, malignancy, Factor V Leiden gene mutation, general anaesthesia and orthopaedic surgery, with higher rates of postoperative DVT, although there remain some variables within the study designs that may lead to overestimation of effect. There is no evidence to support the suggested risk factors of hormone replacement therapy, gender, ethnicity or race, chemotherapy, other thrombophilias, cardiovascular factors, smoking and blood type. CONCLUSIONS: An accurate knowledge of evidence-based risk factors is important in predicting and preventing postoperative DVT, and can be incorporated into a decision support system for appropriate thromboprophylaxis use.

A decision aid for predicting non-sentinel node involvement in women with breast cancer and at least one positive sentinel node.

BACKGROUND: Several clinical trials are re-evaluating the management of the axilla after sentinel node (SN) biopsy. Approximately 50-70% of patients with positive SN have no further nodal involvement. Estimates of the risk of non-sentinel node (NSN) involvement would aid decisions regarding further axillary surgery. METHODS: Clinical and pathological variables for 82 breast cancer patients with metastasis to at least one SN, were used to find independent predictors of the status of NSNs. RESULTS: NSN metastases were found in 46.3% of patients. In a regression model patient age, proportion of SN replaced by metastasis and number of SNs were independent predictors of NSN status. CONCLUSION: Data available after SN biopsy allow estimation of the risk of NSN metastases among patients with positive SNs. Individualised estimates of the risk of NSN involvement may facilitate discussions regarding the trade off between the likely benefits of further axillary surgery and the morbidity of this procedure.

Leiomyosarcoma of somatic soft tissues: a tumor of vascular origin with multivariate analysis of outcome in 42 cases.

Leiomyosarcomas of the somatic soft tissues (SST) are rare compared with their retroperitoneal and cutaneous counterparts and, therefore, have not been extensively studied. We have analyzed the clinicopathologic features of 42 SST leiomyosarcomas referred in consultation to determine what factors affect outcome. Cutaneous, visceral, retroperitoneal, uterine, gastrointestinal, and major vessel leiomyosarcomas were excluded. By definition all lesions possessed at least focal cytologic atypia and mitotic activity, although the latter varied from <1/10 high power fields to 66/10 high power fields. The patients (21 females and 21 males) ranged in age from 26 to 86 years (mean 60 years); tumors developed in the lower (n = 28) or upper extremity (n = 11) and trunk (n = 3). Most arose in deep (n = 27) as opposed to superficial (n = 15) soft tissue; 39 arose from a small vein. During the follow-up period (mean 47 months, range 9-162 months), 3 of 38 (8%) patients developed local recurrence and 17 of 38 metastasized (45%) mostly to the lungs. In a univariate analysis age >62 years, size >4 cm, extensive necrosis, modified updated French Federation of Cancer Centers (FFCC) grade, and whether the tumor had been "disrupted" by a previous incisional biopsy or incomplete excision were significantly correlated with metastasis. AJCC stage also approached significance (p = 0.096) but could not be reliably tested because of the sparseness of the data. In multivariate analyses the logistic regression model that best predicted metastasis at 36 months incorporated the effects of age, FFCC grade, and disruption and had a sensitivity of 94.1% and a specificity of 95.2%. Disruption was the only significant risk factor for metastasis in a multivariate analysis (relative risk 2.70; p = 0.0001) but was strongly correlated with large size and deep location. Other parameters did not improve the predictive power of the model significantly. We concluded that the majority of SST leiomyosarcomas are actually of vascular origin, an observation that has clinical and possibly biologic ramifications. Our histologic definition of leiomyosarcoma to include atypia and any level of mitotic activity appears warranted by the biologic outcome in our cases. The risk of metastasis can be calculated from a model incorporating age, FFCC grade, and disruption. Because disruption correlates with size and depth, it could represent a surrogate as opposed to causal marker for metastasis. Nevertheless, in view of their vascular origin, the possibility that tumor disruption may facilitate or promote access to the bloodstream merits further study.