Machine Learning to Allocate Palliative Care Consultations During Cancer Treatment.

PURPOSE: For patients with advanced cancer, early consultations with palliative care (PC) specialists reduce costs, improve quality of life, and prolong survival. However, capacity limitations prevent all patients from receiving PC shortly after diagnosis. We evaluated whether a prognostic machine learning system could promote early PC, given existing capacity. METHODS: Using population-level administrative data in Ontario, Canada, we assembled a cohort of patients with incurable cancer who received palliative-intent systemic therapy between July 1, 2014, and December 30, 2019. We developed a machine learning system that predicted death within 1 year of each treatment using demographics, cancer characteristics, treatments, symptoms, laboratory values, and history of acute care admissions. We trained the system in patients who started treatment before July 1, 2017, and evaluated the potential impact of the system on PC in subsequent patients. RESULTS: Among 560,210 treatments received by 54,628 patients, death occurred within 1 year of 45.2% of treatments. The machine learning system recommended the same number of PC consultations observed with usual care at the 60.0% 1-year risk of death, with a first-alarm positive predictive value of 69.7% and an outcome-level sensitivity of 74.9%. Compared with usual care, system-guided care could increase early PC by 8.5% overall (95% CI, 7.5 to 9.5; P < .001) and by 15.3% (95% CI, 13.9 to 16.6; P < .001) among patients who live 6 months beyond their first treatment, without requiring more PC consultations in total or substantially increasing PC among patients with a prognosis exceeding 2 years. CONCLUSION: Prognostic machine learning systems could increase early PC despite existing resource constraints. These results demonstrate an urgent need to deploy and evaluate prognostic systems in real-time clinical practice to increase access to early PC.

Machine Learning-Based Early Warning Systems for Acute Care Utilization During Systemic Therapy for Cancer.

BACKGROUND: Emergency department visits and hospitalizations frequently occur during systemic therapy for cancer. We developed and evaluated a longitudinal warning system for acute care use. METHODS: Using a retrospective population-based cohort of patients who started intravenous systemic therapy for nonhematologic cancers between July 1, 2014, and June 30, 2020, we randomly separated patients into cohorts for model training, hyperparameter tuning and model selection, and system testing. Predictive features included static features, such as demographics, cancer type, and treatment regimens, and dynamic features, such as patient-reported symptoms and laboratory values. The longitudinal warning system predicted the probability of acute care utilization within 30 days after each treatment session. Machine learning systems were developed in the training and tuning cohorts and evaluated in the testing cohort. Sensitivity analyses considered feature importance, other acute care endpoints, and performance within subgroups. RESULTS: The cohort included 105,129 patients who received 1,216,385 treatment sessions. Acute care followed 182,444 (15.0%) treatments within 30 days. The ensemble model achieved an area under the receiver operating characteristic curve of 0.742 (95% CI, 0.739-0.745) and was well calibrated in the test cohort. Important predictive features included prior acute care use, treatment regimen, and laboratory tests. If the system was set to alarm approximately once every 15 treatments, 25.5% of acute care events would be preceded by an alarm, and 47.4% of patients would experience acute care after an alarm. The system underestimated risk for some treatment regimens and potentially underserved populations such as females and non-English speakers. CONCLUSIONS: Machine learning warning systems can detect patients at risk for acute care utilization, which can aid in preventive intervention and facilitate tailored treatment. Future research should address potential biases and prospectively evaluate impact after system deployment.

Incidence and Current Treatment Gaps in Pediatric Stroke and TIA: An Ontario-Wide Population-Based Study.

BACKGROUND: Reported incidence rates of pediatric stroke and transient ischemic attack (TIA) range widely. Treatment gaps are poorly characterized. We sought to evaluate in -Ontario, the incidence and characteristics of pediatric stroke and TIA including care gaps and the predictive value of International Classification of Diseases (ICD) codes. METHODS: A retrospective chart review was conducted at 147 Ontario pediatric and adult acute care hospitals. Pediatric stroke and TIA cases (age < 18 years) were identified using ICD-10 code searches in the 2010/11 Canadian Institute for Health Information's Discharge Abstract Database (CIHI-DAD) and National Ambulatory Care Reporting System (NACRS) databases in the Ontario Stroke Audit. RESULTS: Among 478 potential pediatric stroke and TIA cases identified in the CIHI-DAD and NACRS databases, 163 were confirmed as cases of stroke and TIA during the 1-year study period. The Ontario stroke and TIA incidence rate was 5.9 per 100,000 children (3.3 ischemic, 1.8 hemorrhagic and 0.8 TIA). Mean age was 6.4 years (16% neonate). Nearly half were not imaged within 24 h of arrival in emergency and only 56% were given antithrombotic treatment. At discharge, 83 out of 121 (69%) required health care services post-discharge. Overall positive predictive value (PPV) of ICD-10 stroke and TIA codes was 31% (range 5-74%) and yield ranged from 2.4 to 29% for acute stroke or TIA event; code I63 achieved maximal PPV and yield. CONCLUSION: Our population-based study yielded a higher incidence rate than prior North-American studies. Important care gaps exist including delayed diagnosis, lack of expert care, and departure from published treatment guidelines. Variability in ICD PPV and yield underlines the need for prospective data collection and for improving the pediatric stroke and TIA coding processes.

Outcomes following telestroke-assisted thrombolysis for stroke in Ontario, Canada.

Introduction Since 2002, the Ontario Telestroke Program has provided hospitals in under-served regions of the province the opportunity to offer intravenous thrombolysis with tissue plasminogen activator (IV tPA) to eligible patients. The purpose of this study was to determine whether telestroke-assisted IV tPA patients had similar risks of 7- and 90-day mortality, symptomatic intracerebral haemorrhage (sICH), and poor functional outcome compared to patients who received IV tPA with on-site expertise. Methods Data from two audits of patients with acute ischaemic stroke hospitalized in Ontario, Canada in 2010 and 2012 were analysed. We modelled the risk of all-cause death within 7 and 90 days of receiving IV tPA using proportional hazards adjusting for hospital type, patient characteristics, and whether IV tPA was administered as part of a telestroke consultation. Outcomes of sICH and modified Rankin Scale ≥ 3 at discharge were modelled using generalized estimating equations adjusting for the same variables used in the mortality model. Results There was no difference in 7- or 90-day mortality among those who received IV tPA with telestroke ( n = 214) compared to those without ( n = 1885) (7-day adjusted hazard ratio (aHR) 1.29 (95% confidence interval (CI) 0.68, 2.44); 90-day aHR 1.01 (95% CI 0.67, 1.50)). Complications were similar between groups, with an adjusted odds ratio (aOR) for sICH of 0.71 (95% CI 0.29, 1.71) and an aOR of 0.75 (95% CI 0.46, 1.23) for poor functional ability at discharge. Discussion Patients receiving IV tPA supported by telestroke had similar outcomes to those managed with on-site expertise.

Benchmarks for acute stroke care delivery.

OBJECTIVE: Despite widespread interest in many jurisdictions in monitoring and improving the quality of stroke care delivery, benchmarks for most stroke performance indicators have not been established. The objective of this study was to develop data-derived benchmarks for acute stroke quality indicators. DESIGN: Nine key acute stroke quality indicators were selected from the Canadian Stroke Best Practice Performance Measures Manual. PARTICIPANTS: A population-based retrospective sample of patients discharged from 142 hospitals in Ontario, Canada, between 1 April 2008 and 31 March 2009 (N = 3191) was used to calculate hospital rates of performance and benchmarks. INTERVENTION: The Achievable Benchmark of Care (ABC™) methodology was used to create benchmarks based on the performance of the upper 15% of patients in the top-performing hospitals. MAIN OUTCOME MEASURES: Benchmarks were calculated for rates of neuroimaging, carotid imaging, stroke unit admission, dysphasia screening and administration of stroke-related medications. RESULTS: The following benchmarks were derived: neuroimaging within 24 h, 98%; admission to a stroke unit, 77%; thrombolysis among patients arriving within 2.5 h, 59%; carotid imaging, 93%; dysphagia screening, 88%; antithrombotic therapy, 98%; anticoagulation for atrial fibrillation, 94%; antihypertensive therapy, 92% and lipid-lowering therapy, 77%. ABC™ acute stroke care benchmarks achieve or exceed the consensus-based targets required by Accreditation Canada, with the exception of dysphagia screening. CONCLUSIONS: Benchmarks for nine hospital-based acute stroke care quality indicators have been established. These can be used in the development of standards for quality improvement initiatives.