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.

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.

Evaluating the Impact of the Grading and Assessment of Predictive Tools Framework on Clinicians and Health Care Professionals' Decisions in Selecting Clinical Predictive Tools: Randomized Controlled Trial.

BACKGROUND: While selecting predictive tools for implementation in clinical practice or for recommendation in clinical guidelines, clinicians and health care professionals are challenged with an overwhelming number of tools. Many of these tools have never been implemented or evaluated for comparative effectiveness. To overcome this challenge, the authors developed and validated an evidence-based framework for grading and assessment of predictive tools (the GRASP framework). This framework was based on the critical appraisal of the published evidence on such tools. OBJECTIVE: The aim of the study was to examine the impact of using the GRASP framework on clinicians' and health care professionals' decisions in selecting clinical predictive tools. METHODS: A controlled experiment was conducted through a web-based survey. Participants were randomized to either review the derivation publications, such as studies describing the development of the predictive tools, on common traumatic brain injury predictive tools (control group) or to review an evidence-based summary, where each tool had been graded and assessed using the GRASP framework (intervention group). Participants in both groups were asked to select the best tool based on the greatest validation or implementation. A wide group of international clinicians and health care professionals were invited to participate in the survey. Task completion time, rate of correct decisions, rate of objective versus subjective decisions, and level of decisional conflict were measured. RESULTS: We received a total of 194 valid responses. In comparison with not using GRASP, using the framework significantly increased correct decisions by 64%, from 53.7% to 88.1% (88.1/53.7=1.64; t193=8.53; P<.001); increased objective decision making by 32%, from 62% (3.11/5) to 82% (4.10/5; t189=9.24; P<.001); decreased subjective decision making based on guessing by 20%, from 49% (2.48/5) to 39% (1.98/5; t188=-5.47; P<.001); and decreased prior knowledge or experience by 8%, from 71% (3.55/5) to 65% (3.27/5; t187=-2.99; P=.003). Using GRASP significantly decreased decisional conflict and increased the confidence and satisfaction of participants with their decisions by 11%, from 71% (3.55/5) to 79% (3.96/5; t188=4.27; P<.001), and by 13%, from 70% (3.54/5) to 79% (3.99/5; t188=4.89; P<.001), respectively. Using GRASP decreased the task completion time, on the 90th percentile, by 52%, from 12.4 to 6.4 min (t193=-0.87; P=.38). The average System Usability Scale of the GRASP framework was very good: 72.5% and 88% (108/122) of the participants found the GRASP useful. CONCLUSIONS: Using GRASP has positively supported and significantly improved evidence-based decision making. It has increased the accuracy and efficiency of selecting predictive tools. GRASP is not meant to be prescriptive; it represents a high-level approach and an effective, evidence-based, and comprehensive yet simple and feasible method to evaluate, compare, and select clinical predictive tools.

Prospective Validation of a Checklist to Predict Short-term Death in Older Patients After Emergency Department Admission in Australia and Ireland.

BACKGROUND: Emergency departments (EDs) are pressured environment where patients with supportive and palliative care needs may not be identified. We aimed to test the predictive ability of the CriSTAL (Criteria for Screening and Triaging to Appropriate aLternative care) checklist to flag patients at risk of death within 3 months who may benefit from timely end-of-life discussions. METHODS: Prospective cohorts of >65-year-old patients admitted for at least one night via EDs in five Australian hospitals and one Irish hospital. Purpose-trained nurses and medical students screened for frailty using two instruments concurrently and completed the other risk factors on the CriSTAL tool at admission. Postdischarge telephone follow-up was used to determine survival status. Logistic regression and bootstrapping techniques were used to test the predictive accuracy of CriSTAL for death within 90 days of admission as primary outcome. Predictability of in-hospital death was the secondary outcome. RESULTS: A total of 1,182 patients, with median age 76 to 80 years (IRE-AUS), were included. The deceased had significantly higher mean CriSTAL with Australian mean of 8.1 (95% confidence interval [CI] = 7.7-8.6) versus 5.7 (95% CI = 5.1-6.2) and Irish mean of 7.7 (95% CI = 6.9-8.5) versus 5.7 (95% CI = 5.1-6.2). The model with Fried frailty score was optimal for the derivation (Australian) cohort but prediction with the Clinical Frailty Scale (CFS) was also good (areas under the receiver-operating characteristic [AUROC] = 0.825 and 0.81, respectively). Values for the validation (Irish) cohort were AUROC = 0.70 with Fried and 0.77 using CFS. A minimum of five of 29 variables were sufficient for accurate prediction, and a cut point of 7+ or 6+ depending on the cohort was strongly indicative of risk of death. The most significant independent predictor of short-term death in both cohorts was frailty, carrying a twofold risk of death. CriSTAL's accuracy for in-hospital death prediction was also good (AUROC = 0.795 and 0.81 in Australia and Ireland, respectively), with high specificity and negative predictive values. CONCLUSIONS: The modified CriSTAL tool (with CFS instead of Fried's frailty instrument) had good discriminant power to improve certainty of short-term mortality prediction in both health systems. The predictive ability of models is anticipated to help clinicians gain confidence in initiating earlier end-of-life discussions. The practicalities of embedding screening for risk of death in routine practice warrant further investigation.

Predictive validity of the CriSTAL tool for short-term mortality in older people presenting at Emergency Departments: a prospective study.

ABSTRACT: To determine the validity of the Australian clinical prediction tool Criteria for Screening and Triaging to Appropriate aLternative care (CRISTAL) based on objective clinical criteria to accurately identify risk of death within 3 months of admission among older patients. METHODS: Prospective study of ≥ 65 year-olds presenting at emergency departments in five Australian (Aus) and four Danish (DK) hospitals. Logistic regression analysis was used to model factors for death prediction; Sensitivity, specificity, area under the ROC curve and calibration with bootstrapping techniques were used to describe predictive accuracy. RESULTS: 2493 patients, with median age 78-80 years (DK-Aus). The deceased had significantly higher mean CriSTAL with Australian mean of 8.1 (95% CI 7.7-8.6 vs. 5.8 95% CI 5.6-5.9) and Danish mean 7.1 (95% CI 6.6-7.5 vs. 5.5 95% CI 5.4-5.6). The model with Fried Frailty score was optimal for the Australian cohort but prediction with the Clinical Frailty Scale (CFS) was also good (AUROC 0.825 and 0.81, respectively). Values for the Danish cohort were AUROC 0.764 with Fried and 0.794 using CFS. The most significant independent predictors of short-term death in both cohorts were advanced malignancy, frailty, male gender and advanced age. CriSTAL's accuracy was only modest for in-hospital death prediction in either setting. CONCLUSIONS: The modified CriSTAL tool (with CFS instead of Fried's frailty instrument) has good discriminant power to improve prognostic certainty of short-term mortality for ED physicians in both health systems. This shows promise in enhancing clinician's confidence in initiating earlier end-of-life discussions.

Efficacy of a tool to predict short-term mortality in older people presenting at emergency departments: Protocol for a multi-centre cohort study.

BACKGROUND: Prognostic uncertainty inhibits clinicians from initiating timely end-of-life discussions and advance care planning. This study evaluates the efficacy of the CriSTAL (Criteria for Screening and Triaging to Appropriate aLternative care) checklist in emergency departments. METHODS: Prospective cohort study of patients aged ≥65 years with any diagnosis admitted via emergency departments in ten hospitals in Australia, Denmark and Ireland. Electronic and paper clinical records will be used to extract risk factors such as nursing home residency, physiological deterioration warranting a rapid response call, personal history of active chronic disease, history of hospitalisations or intensive care unit admission in the past year, evidence of proteinuria or ECG abnormalities, and evidence of frailty to be concurrently measured with Fried Score and Clinical Frailty Scale. Patients or their informal caregivers will be contacted by telephone around three months after initial assessment to ascertain survival, self-reported health, post-discharge frailty and health service utilisation since discharge. Logistic regression and bootstrapping techniques and AUROC curves will be used to test the predictive accuracy of CriSTAL for death within 90 days of admission and in-hospital death. DISCUSSION: The CriSTAL checklist is an objective and practical tool for use in emergency departments among older patients to determine individual probability of death in the short-term. Its validation in this cohort is expected to reduce clinicians' prognostic uncertainty on the time to patients' death and encourage timely end-of-life conversations to support clinical decisions with older frail patients and their families about their imminent or future care choices.

What is inappropriate hospital use for elderly people near the end of life? A systematic review.

BACKGROUND: Older people with advance chronic illness use hospital services repeatedly near the end of life. Some of these hospitalizations are considered inappropriate. AIM: To investigate extent and causes of inappropriate hospital admission among older patients near the end of life. METHODS: English language publications in Medline, EMBASE, PubMed, Cochrane library, and the grey literature (January 1995-December 2016) covering community and nursing home residents aged ≥60years admitted to hospital. OUTCOMES: measurements of inappropriateness. A 17-item quality score was estimated independently by two authors. RESULTS: The definition of 'Inappropriate admissions' near the end of life incorporated system factors, social and family factors. The prevalence of inappropriate admissions ranged widely depending largely on non-clinical reasons: poor availability of alternative sites of care or failure of preventive actions by other healthcare providers (1.7-67.0%); family requests (up to 10.5%); or too late an admission to be of benefit (1.7-35.0%). The widespread use of subjective parameters not routinely collected in practice, and the inclusion of non-clinical factors precluded the true estimation of clinical inappropriateness. CONCLUSIONS: Clinical inappropriateness and system factors that preclude alternative community care must be measured separately. They are two very different justifications for hospital admissions, requiring different solutions. Society has a duty to ensure availability of community alternatives for the management of ambulatory-sensitive conditions and facilitate skilling of staff to manage the terminally ill in non-acute settings. Only then would the evaluation of local variations in clinically inappropriate admissions and inappropriate length of stay be possible to undertake.

Pre-existing risk factors for in-hospital death among older patients could be used to initiate end-of-life discussions rather than Rapid Response System calls: A case-control study.

AIM: To investigate associations between clinical parameters - beyond the evident physiological deterioration and limitations of medical treatment - with in-hospital death for patients receiving Rapid Response System (RRS) attendances. METHODS: Retrospective case-control analysis of clinical parameters for 328 patients aged 60 years and above at their last RRS call during admission to a single teaching hospital in the 2012-2013 calendar years. Generalised estimating equation modelling was used to compare the deceased with a randomly selected sample of those who had RRS calls and survived admission (controls), matched by age group, sex, and hospital ward. RESULTS: In addition to a pre-existing order for limitation of treatment or cardiac arrest (OR 6.92; 95%CI 4.61-10.27), nursing home residence, proteinuria, advanced malignancy, acute myocardial infarction, chronic kidney disease, cognitive impairment and frailty were associated with high risk of death. After adjusting for all the clinical indicators investigated, the strongest risk factors for in-hospital death for patients with a RRS call were advanced malignancy (OR 3.95; 95%CI 2.16-7.21) and new myocardial infarction (OR 2.79; 95%CI 1.86-4.20). Patients with cognitive impairment, frailty indicator or chronic kidney disease were twice as likely to die as patients without those risk factors. CONCLUSION: In a sample of older deteriorated patients requiring a RRS attendance, multiple indicators of chronic illness, cognitive impairment and frailty were significantly associated with high risk of death. These clinical features beyond the evident orders for limitation of medical treatment should signal the need for clinicians to initiate end-of-life discussions that may prevent futile interventions.