R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 15.

In this latest update we discuss real-world evidence (RWE) guidance from the leading oncology professional societies, the American Society of Clinical Oncology and the European Society for Medical Oncology, and the PRINCIPLED practical guide on the design and analysis of causal RWE studies.

R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 14.

In this latest update we highlight: a publication from the US FDA regarding the definitions of real-world data (RWD) and real-world evidence (RWE); a publication from academic researchers on a demonstration project for target trial emulation; a publication from the National Institute of Health and Care Excellence (NICE) on the 1 year anniversary of their RWE framework; and a publication from NICE and Flatiron Health on the utility of US RWD for initial UK health technology assessment decision making.

The impact of the COVID-19 pandemic on cardiovascular disease prevention and corresponding geographical inequalities in England: interrupted time series analysis.

BACKGROUND: There has been disruption to the detection and management of those with hypertension and atrial fibrillation (AF) during the COVID-19 pandemic. This is likely to vary geographically and could have implications for future mortality and morbidity. We aimed to estimate the change in diagnosed prevalence, treatment and prescription indicators for AF and hypertension and assess corresponding geographical inequalities. METHODS: Using the Quality and Outcomes Framework (2016/17 to 2021/22) and the English Prescribing Datasets (2018 to 2022), we described age standardised prevalence, treatment and prescription item rates for hypertension and AF by geography and over time. Using an interrupted time-series (ITS) analysis, we estimated the impact of the pandemic (from April 2020) on missed diagnoses and on the percentage change in medicines prescribed for these conditions. Finally, we described changes in treatment indicators against Public Health England 2029 cardiovascular risk targets. RESULTS: We observed 143,822 fewer (-143,822, 95%CI:-226,144, -61,500, p = 0.001) diagnoses of hypertension, 60,330 fewer (-60,330, 95%CI: -83,216, -37,444, p = 0.001) diagnoses of AF and 1.79% fewer (-1.79%, 95%CI: -2.37%, -1.22%), p < 0.0001) prescriptions for these conditions over the COVID-19 impact period. There was substantial variation across geography in England in terms of the indirect impact of the COVID-19 pandemic on the diagnosis, prescription, and treatment rates of hypertension and AF. 20% of Sub Integrated Care Boards account for approximately 62% of all missed diagnoses of hypertension. The percentage of individuals who had their hypertension controlled fell from 75.8% in 2019/20 to 64.1% in 2021/22 and the percentage of individuals with AF who were risk assessed fell from 97.2% to 90.7%. CONCLUSIONS: Hypertension and AF detection and management were disrupted during the COVID-19 pandemic. The disruption varied considerably across diseases and geography. This highlights the utility of administrative and geographically granular datasets to inform targeted efforts to mitigate the indirect impacts of the pandemic through applied secondary prevention measures.

Developing and externally validating a machine learning risk prediction model for 30-day mortality after stroke using national stroke registers in the UK and Sweden.

OBJECTIVES: We aimed to develop and externally validate a generalisable risk prediction model for 30-day stroke mortality suitable for supporting quality improvement analytics in stroke care using large nationwide stroke registers in the UK and Sweden. DESIGN: Registry-based cohort study. SETTING: Stroke registries including the Sentinel Stroke National Audit Programme (SSNAP) in England, Wales and Northern Ireland (2013-2019) and the national Swedish stroke register (Riksstroke 2015-2020). PARTICIPANTS AND METHODS: Data from SSNAP were used for developing and temporally validating the model, and data from Riksstroke were used for external validation. Models were developed with the variables available in both registries using logistic regression (LR), LR with elastic net and interaction terms and eXtreme Gradient Boosting (XGBoost). Performances were evaluated with discrimination, calibration and decision curves. OUTCOME MEASURES: The primary outcome was all-cause 30-day in-hospital mortality after stroke. RESULTS: In total, 488 497 patients who had a stroke with 12.4% 30-day in-hospital mortality were used for developing and temporally validating the model in the UK. A total of 128 360 patients who had a stroke with 10.8% 30-day in-hospital mortality and 13.1% all mortality were used for external validation in Sweden. In the SSNAP temporal validation set, the final XGBoost model achieved the highest area under the receiver operating characteristic curve (AUC) (0.852 (95% CI 0.848 to 0.855)) and was well calibrated. The performances on the external validation in Riksstroke were as good and achieved AUC at 0.861 (95% CI 0.858 to 0.865) for in-hospital mortality. For Riksstroke, the models slightly overestimated the risk for in-hospital mortality, while they were better calibrated at the risk for all mortality. CONCLUSION: The risk prediction model was accurate and externally validated using high quality registry data. This is potentially suitable to be deployed as part of quality improvement analytics in stroke care to enable the fair comparison of stroke mortality outcomes across hospitals and health systems across countries.

R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 13.

In this latest update we highlight a report from the European Medicines Agency on their use of real-world evidence (RWE) in decision making, RWE reporting guidance from the Canadian Agency for Drugs and Technologies in Health and highlight some new data demonstrating the value medicines for spinal muscular atrophy have brought patients.

Do Stroke Care Processes Modify Clinical Outcomes in Patients with Stroke-Associated Pneumonia? A Registry Cohort Study in England and Wales.

INTRODUCTION: It is not known whether modern stroke unit care reduces the impact of stroke complications, such as stroke-associated pneumonia (SAP), on clinical outcomes. We investigated the relationship between SAP and clinical outcomes, adjusting for the confounding effects of stroke care processes and their timing. METHODS: The Sentinel Stroke National Audit Programme provided patient data for all confirmed strokes between April 2013 and December 2018. SAP was defined as new antibiotic initiation for suspected pneumonia within the first 7 days from stroke admission. We compared outcomes after SAP versus non-SAP in appropriate multilevel mixed models. Each model was adjusted for patient and clinical characteristics, as well as markers of stroke care and their timing within the first 72 h. The appropriate effect estimates and corresponding 95% confidence intervals (CIs) were reported. RESULTS: Of 201,778 patients, SAP was present in 14.2%. After adjustment for timing of acute stroke care processes and clinical characteristics, adverse outcomes remained for SAP versus non-SAP patients. In these adjusted analyses, patients with SAP maintained an increased risk of longer length of in-hospital stay (IRR of 1.27; 95% CI: 1.25, 1.30), increased odds of worse functional outcome at discharge (OR of 2.9; 95% CI: 2.9, 3.0), and increased risk of in-hospital mortality (HR of 1.78; 95% CI: 1.74, 1.82). CONCLUSION: We show for the first time that SAP remains associated with worse clinical outcomes, even after adjusting for processes of acute stroke care and their timing. These findings highlight the importance of continued research efforts aimed at preventing SAP.

Risk prediction of 30-day mortality after stroke using machine learning: a nationwide registry-based cohort study.

BACKGROUNDS: We aimed to develop and validate machine learning (ML) models for 30-day stroke mortality for mortality risk stratification and as benchmarking models for quality improvement in stroke care. METHODS: Data from the UK Sentinel Stroke National Audit Program between 2013 to 2019 were used. Models were developed using XGBoost, Logistic Regression (LR), LR with elastic net with/without interaction terms using 80% randomly selected admissions from 2013 to 2018, validated on the 20% remaining admissions, and temporally validated on 2019 admissions. The models were developed with 30 variables. A reference model was developed using LR and 4 variables. Performances of all models was evaluated in terms of discrimination, calibration, reclassification, Brier scores and Decision-curves. RESULTS: In total, 488,497 stroke patients with a 12.3% 30-day mortality rate were included in the analysis. In 2019 temporal validation set, XGBoost model obtained the lowest Brier score (0.069 (95% CI: 0.068-0.071)) and the highest area under the ROC curve (AUC) (0.895 (95% CI: 0.891-0.900)) which outperformed LR reference model by 0.04 AUC (p < 0.001) and LR with elastic net and interaction term model by 0.003 AUC (p < 0.001). All models were perfectly calibrated for low (< 5%) and moderate risk groups (5-15%) and ≈1% underestimation for high-risk groups (> 15%). The XGBoost model reclassified 1648 (8.1%) low-risk cases by the LR reference model as being moderate or high-risk and gained the most net benefit in decision curve analysis. CONCLUSIONS: All models with 30 variables are potentially useful as benchmarking models in stroke-care quality improvement with ML slightly outperforming others.

Power of big data to improve patient care in gastroenterology.

Big data is defined as being large, varied or frequently updated, and usually generated from real-world interaction. With the unprecedented availability of big data, comes an obligation to maximise its potential for healthcare improvements in treatment effectiveness, disease prevention and healthcare delivery. We review the opportunities and challenges that big data brings to gastroenterology. We review its sources for healthcare improvement in gastroenterology, including electronic medical records, patient registries and patient-generated data. Big data can complement traditional research methods in hypothesis generation, supporting studies and disseminating findings; and in some cases holds distinct advantages where traditional trials are unfeasible. There is great potential power in patient-level linkage of datasets to help quantify inequalities, identify best practice and improve patient outcomes. We exemplify this with the UK colorectal cancer repository and the potential of linkage using the National Endoscopy Database, the inflammatory bowel disease registry and the National Health Service bowel cancer screening programme. Artificial intelligence and machine learning are increasingly being used to improve diagnostics in gastroenterology, with image analysis entering clinical practice, and the potential of machine learning to improve outcome prediction and diagnostics in other clinical areas. Big data brings issues with large sample sizes, real-world biases, data curation, keeping clinical context at analysis and General Data Protection Regulation compliance. There is a tension between our obligation to use data for the common good and protecting individual patient's data. We emphasise the importance of engaging with our patients to enable them to understand their data usage as fully as they wish.

The Timing of Stroke Care Processes and Development of Stroke Associated Pneumonia: A National Registry Cohort Study.

Introduction: Timely stroke care can result in significant improvements in stroke recovery. However, little is known about how stroke care processes relate to complications such as the development of stroke associated pneumonia (SAP). Here we investigated associations between stroke care processes, their timing and development of SAP. Methods: We obtained patient-level data from the Sentinel Stroke National Audit Programme for all confirmed strokes between 1st April 2013 and 31st December 2018. SAP was identified if new antibiotic initiation for pneumonia occurred within the first 7 days of admission. Time to key stroke care processes in the pre-hospital, hyperacute and acute phase were investigated. A mixed effects logistic regression model estimated the association between SAP [Odds ratios (OR) with 95% CI] and each process of care after controlling for pre-determined confounders such as age, stroke severity and comorbidities. Results: SAP was identified in 8.5% of 413,133 patients in 169 stroke units. A long time to arrival at a stroke unit after symptom onset or time last seen well [OR (95% CI) = 1.29 (1.23-1.35)], from admission to assessment by a stroke specialist [1.10 (1.06-1.14)] and from admission to assessment by a physiotherapist [1.16 (1.12-1.21)] were all independently associated with SAP. Short door to needle times were associated with lower odds of SAP [0.90 (0.83-0.97)]. Conclusion: Times from stroke onset and admission to certain key stroke care processes were associated with SAP. Understanding how timing of these care processes relate to SAP may enable development of preventive interventions to reduce antibiotic use and improve clinical outcomes.

Investigating the association between inpatient stroke therapy and disability, destination on discharge, length of stay and mortality: a prospective cohort study using the Sentinel Stroke National Audit Programme.

OBJECTIVE: 'More is better' is a recognised mantra within stroke therapy, however, this has been developed in patients receiving long term rehabilitation. We investigated the relationship between amount of therapy received (from therapists and psychologists) and key patient outcomes during inpatient care. DESIGN: A secondary analysis of data from a prospective cohort study was performed. Multilevel mixed models adjusting for measured confounders (eg, severity), explored the relationship between therapy dose (average minutes per day of stay) and outcomes (disability, length of stay, home at discharge and mortality). Therapy was explored using simple linear terms and flexible natural cubic splines to allow for more complex relationships. SETTING: Data from the Sentinel Stroke National Audit Programme, covering England, Wales and Northern Ireland between July 2013 and July 2015 contained 94 905 adults with a stroke and still an inpatient after 72 hours. These patients received 92% (physiotherapy), 88% (occupational therapy), 57% (speech and language therapy) and 5% (clinical psychology), respectively. RESULTS: The average amount of therapy, for individual and 'any' therapy combined per day of stay was low. Overall, 41% were discharged with an 'independent' modified Rankin Scale (≤2), 14% died, 44% were discharged home, and the median length of stay was 16 days. We observed complex relationships between amount of therapy received and outcomes. An additional minute of 'any' therapy, occupational therapy, speech and language therapy and clinical psychology was associated with improved outcomes. Conversely, more physiotherapy was also associated with lower mortality and shorter length of stay, but also lower independence and discharge home. CONCLUSIONS: Our findings suggest for stroke inpatients requiring therapy, 'More is better' may be overly simplistic. Strong limitations associated with analysis of routine data restrict further robust investigation of the therapy-response relationship. Robust prospective work is urgently needed to further investigate the relationships observed here.

Variation of stroke-associated pneumonia in stroke units across England and Wales: A registry-based cohort study.

BACKGROUND: Pneumonia is common in stroke patients and is associated with worse clinical outcomes. Prevalence of stroke-associated pneumonia varies between studies, and reasons for this variation remain unclear. We aimed to describe the variation of observed stroke-associated pneumonia in England and Wales and explore the influence of patient baseline characteristics on this variation. METHODS: Patient data were obtained from the Sentinel Stroke National Audit Programme for all confirmed strokes between 1 April 2013 and 31 December 2018. Stroke-associated pneumonia was defined by new antibiotic initiation for pneumonia within the first seven days of admission. The probability of stroke-associated pneumonia occurrence within stroke units was estimated and compared using a multilevel mixed model with and without adjustment for patient-level characteristics at admission. RESULTS: Of the 413,133 patients included, median National Institutes of Health Stroke Scale was 4 (IQR: 2-10) and 42.3% were aged over 80 years. Stroke-associated pneumonia was identified in 8.5% of patients. The median within stroke unit stroke-associated pneumonia prevalence was 8.5% (IQR: 6.1-11.5%) with a maximum of 21.4%. The mean and variance of the predicted stroke-associated pneumonia probability across stroke units decreased from 0.08 (0.68) to 0.05 (0.63) when adjusting for patient admission characteristics. This difference in the variance suggests that clinical characteristics account for 5% of the observed variation in stroke-associated pneumonia between units. CONCLUSIONS: Patient-level clinical characteristics contributed minimally to the observed variation of stroke-associated pneumonia between stroke units. Additional explanations for the observed variation in stroke-associated pneumonia need to be explored which could reduce variation in antibiotic use for stroke patients.

Are Differences in Dysphagia Assessment, Oral Care Provision, or Nasogastric Tube Insertion Associated with Stroke-Associated Pneumonia? A Nationwide Survey Linked to National Stroke Registry Data.

INTRODUCTION: Stroke-associated pneumonia (SAP) is a common complication associated with poor outcomes. Early dysphagia screening and specialist assessment is associated with a reduced risk of SAP. Evidence about oral care and nasogastric tube (NGT) placement is equivocal. This study aimed to expose variations in dysphagia management practices and explore their associations with SAP. PARTICIPANTS AND METHODS: Speech pathologists from 166 stroke units in England and Wales were surveyed about dysphagia assessment and management, oral care, and NGT placement. Survey data were then linked to the Sentinel Stroke National Audit Programme (SSNAP), the national register of stroke. Univariable and multivariable linear regression models were fitted to estimate the association between dysphagia management practices and SAP incidence. RESULTS: 113 hospitals completed the survey (68%). Variation was evident in dysphagia screening protocols (DSPs), oral care, and NGT practice while specialist swallow assessment data patterns were more consistent. Multivariable analysis showed no evidence of an association in incidence of SAP when using a water-only hospital DSP compared to a multiconsistency DSP (B -0.688, 95% CI: -2.912 to 1.536), when using written swallow assessment guidelines compared to not using written guidelines (B 0.671, 95% CI: -1.567 to 2.908), when teams inserted NGTs overnight compared to teams which did not (B -0.505, 95% CI: -2.759 to 1.749), and when teams had a written oral care protocol compared to those which did not (B -1.339, 95% CI: -3.551 to 0.873). DISCUSSION AND CONCLUSION: Variation exists in dysphagia screening and management, but there was no evidence of an association between clinical practice patterns and incidence of SAP. Further research with larger sample sizes is needed to examine association with SAP.

Stroke Care in the United Kingdom During the COVID-19 Pandemic.

BACKGROUND AND PURPOSE: The coronavirus disease 2019 (COVID-19) pandemic has potentially caused indirect harm to patients with other conditions via reduced access to health care services. We aimed to describe the impact of the initial wave of the pandemic on admissions, care quality, and outcomes in patients with acute stroke in the United Kingdom. METHODS: Registry-based cohort study of patients with acute stroke admitted to hospital in England, Wales, and Northern Ireland between October 1, 2019, and April 30, 2020, and equivalent periods in the 3 prior years. RESULTS: One hundred fourteen hospitals provided data for a study cohort of 184 017 patients. During the lockdown period (March 23 to April 30), there was a 12% reduction (6923 versus 7902) in the number of admissions compared with the same period in the 3 previous years. Admissions fell more for ischemic than hemorrhagic stroke, for older patients, and for patients with less severe strokes. Quality of care was preserved for all measures and in some domains improved during lockdown (direct access to stroke unit care, 1-hour brain imaging, and swallow screening). Although there was no change in the proportion of patients discharged with good outcome (modified Rankin Scale score, ≤2; 48% versus 48%), 7-day inpatient case fatality increased from 6.9% to 9.4% (P<0.001) and was 22.0% in patients with confirmed or suspected COVID-19 (adjusted rate ratio, 1.41 [1.11-1.80]). CONCLUSIONS: Assuming that the true incidence of acute stroke did not change markedly during the pandemic, hospital avoidance may have created a cohort of untreated stroke patients at risk of poorer outcomes or recurrent events. Unanticipated improvements in stroke care quality should be used as an opportunity for quality improvement and to learn about how to develop resilient health care systems.

Stroke impairment categories: A new way to classify the effects of stroke based on stroke-related impairments.

OBJECTIVE: To create a classification system based on stroke-related impairments. DATA SOURCE: All adults with stroke admitted for at least 72 hours in England, Wales and Northern Ireland from July 2013 to July 2015 extracted from the Sentinel Stroke National Audit Programme. ANALYSIS: Impairments were defined using the National Institute of Health Stroke Scale scores at admission. Common combinations of impairments were identified based on geometric coding and expert knowledge. Validity of the classification was assessed using standard descriptive statistics to report and compare patients' characteristics, therapy received and outcomes in each group. RESULTS: Data from 94,905 patients were extracted. The items of the National Institute of Health Stroke Scale (on admission) were initially grouped into four body systems: Cognitive, Motor, Sensory and Consciousness. Seven common combinations of these impairments were identified (in order of stroke severity); Patients with Loss of Consciousness (n = 6034, 6.4%); those with Motor + Cognitive + Sensory impairments (n = 28,226, 29.7%); Motor + Cognitive impairments (n = 16,967, 17.9%); Motor + Sensory impairments (n = 9882, 10.4%); Motor Only impairments (n = 20,471, 21.6%); Any Non-Motor impairments (n = 7498, 7.9%); and No Impairments (n = 5827, 6.1%). There was a gradation of age, premorbid disability, mortality and disability on discharge. People with the most and least severe categories were least likely to receive therapy, and received least therapy (-20 minutes/day of stay) compared to -35 minutes/day of stay for the moderately severe categories. CONCLUSIONS: A classification system of seven Stroke Impairment Categories has been presented.

Monitoring indirect impact of COVID-19 pandemic on services for cardiovascular diseases in the UK.

OBJECTIVE: To monitor hospital activity for presentation, diagnosis and treatment of cardiovascular diseases during the COVID-19) pandemic to inform on indirect effects. METHODS: Retrospective serial cross-sectional study in nine UK hospitals using hospital activity data from 28 October 2019 (pre-COVID-19) to 10 May 2020 (pre-easing of lockdown) and for the same weeks during 2018-2019. We analysed aggregate data for selected cardiovascular diseases before and during the epidemic. We produced an online visualisation tool to enable near real-time monitoring of trends. RESULTS: Across nine hospitals, total admissions and emergency department (ED) attendances decreased after lockdown (23 March 2020) by 57.9% (57.1%-58.6%) and 52.9% (52.2%-53.5%), respectively, compared with the previous year. Activity for cardiac, cerebrovascular and other vascular conditions started to decline 1-2 weeks before lockdown and fell by 31%-88% after lockdown, with the greatest reductions observed for coronary artery bypass grafts, carotid endarterectomy, aortic aneurysm repair and peripheral arterial disease procedures. Compared with before the first UK COVID-19 (31 January 2020), activity declined across diseases and specialties between the first case and lockdown (total ED attendances relative reduction (RR) 0.94, 0.93-0.95; total hospital admissions RR 0.96, 0.95-0.97) and after lockdown (attendances RR 0.63, 0.62-0.64; admissions RR 0.59, 0.57-0.60). There was limited recovery towards usual levels of some activities from mid-April 2020. CONCLUSIONS: Substantial reductions in total and cardiovascular activities are likely to contribute to a major burden of indirect effects of the pandemic, suggesting they should be monitored and mitigated urgently.

Factors influencing the amount of therapy received during inpatient stroke care: an analysis of data from the UK Sentinel Stroke National Audit Programme.

OBJECTIVES: To understand why most stroke patients receive little therapy. We investigated the factors associated with the amount of stroke therapy delivered. METHODS: Data regarding adults admitted to hospital with stroke for at least 72 hours (July 2013-July 2015) were extracted from the UK's Sentinel Stroke National Audit Programme. Descriptive statistics and multilevel mixed effects regression models explored the factors that influenced the amount of therapy received while adjusting for confounding. RESULTS: Of the 94,905 patients in the study cohort (mean age: 76 (SD: 13.2) years, 78% had a mild or moderate severity stroke. In all, 92% required physiotherapy, 87% required occupational therapy, 57% required speech therapy but only 5% were considered to need psychology. The average amount of therapy ranged from 2 minutes (psychology) to 14 minutes (physiotherapy) per day of inpatient stay. Unmodifiable characteristics (such as stroke severity) dominated the variation in the amount of therapy. However important, modifiable organizational factors were the day and time of admission, type of stroke team, timely therapy assessments, therapy and nursing staffing levels (qualified and support staff), and presence of weekend or early supported discharge services. CONCLUSION: The amount of stroke therapy is associated with unmodifiable patient-related characteristics and modifiable organizational factors in that more therapy was associated with higher therapy and nurse staffing levels, specialist stroke rehabilitation services, timely therapy assessments, and the presence of weekend and early discharge services.

A systematic review of machine learning models for predicting outcomes of stroke with structured data.

BACKGROUND AND PURPOSE: Machine learning (ML) has attracted much attention with the hope that it could make use of large, routinely collected datasets and deliver accurate personalised prognosis. The aim of this systematic review is to identify and critically appraise the reporting and developing of ML models for predicting outcomes after stroke. METHODS: We searched PubMed and Web of Science from 1990 to March 2019, using previously published search filters for stroke, ML, and prediction models. We focused on structured clinical data, excluding image and text analysis. This review was registered with PROSPERO (CRD42019127154). RESULTS: Eighteen studies were eligible for inclusion. Most studies reported less than half of the terms in the reporting quality checklist. The most frequently predicted stroke outcomes were mortality (7 studies) and functional outcome (5 studies). The most commonly used ML methods were random forests (9 studies), support vector machines (8 studies), decision trees (6 studies), and neural networks (6 studies). The median sample size was 475 (range 70-3184), with a median of 22 predictors (range 4-152) considered. All studies evaluated discrimination with thirteen using area under the ROC curve whilst calibration was assessed in three. Two studies performed external validation. None described the final model sufficiently well to reproduce it. CONCLUSIONS: The use of ML for predicting stroke outcomes is increasing. However, few met basic reporting standards for clinical prediction tools and none made their models available in a way which could be used or evaluated. Major improvements in ML study conduct and reporting are needed before it can meaningfully be considered for practice.

How do patients pass through stroke services? Identifying stroke care pathways using national audit data.

OBJECTIVE: To map and describe how patients pass through stroke services. METHODS: Data from 94,905 stroke patients (July 2013-July 2015) who were still inpatients 72 hours after hospital admission were extracted from a national stroke register and were used to identify the routes patients took through hospital and community stroke services. We sought to categorize these routes through iterative consultations with clinical experts and to describe patient characteristics, therapy provision, outcomes and costs within each category. RESULTS: We identified 874 routes defined by the type of admitting stroke team and subsequent transfer history. We consolidated these into nine distinct routes and further summarized these into three overlapping 'pathways' that accounted for 99% of the patients. These were direct discharge (44%), community rehabilitation (47%) and inpatient transfer (19%) with 12% of the patients receiving both inpatient transfer and community rehabilitation. Patients with the mildest and most severe strokes were more likely to follow the direct discharge pathway. Those perceived to need most therapy were more likely to follow the inpatient transfer pathway. Costs were lowest and mortality was highest for patients on the direct discharge pathway. Outcomes were best for patients on the community rehabilitation pathway and costs were highest where patients underwent inpatient transfers. CONCLUSION: Three overarching stroke care pathways were identified which differ according to patient characteristics, therapy needs and outcomes. This pathway mapping provides a benchmark to develop and plan clinical services, and for future research.

Can clinical audits be enhanced by pathway simulation and machine learning? An example from the acute stroke pathway.

OBJECTIVE: To evaluate the application of clinical pathway simulation in machine learning, using clinical audit data, in order to identify key drivers for improving use and speed of thrombolysis at individual hospitals. DESIGN: Computer simulation modelling and machine learning. SETTING: Seven acute stroke units. PARTICIPANTS: Anonymised clinical audit data for 7864 patients. RESULTS: Three factors were pivotal in governing thrombolysis use: (1) the proportion of patients with a known stroke onset time (range 44%-73%), (2) pathway speed (for patients arriving within 4 hours of onset: per-hospital median arrival-to-scan ranged from 11 to 56 min; median scan-to-thrombolysis ranged from 21 to 44 min) and (3) predisposition to use thrombolysis (thrombolysis use ranged from 31% to 52% for patients with stroke scanned with 30 min left to administer thrombolysis). A pathway simulation model could predict the potential benefit of improving individual stages of the clinical pathway speed, whereas a machine learning model could predict the benefit of 'exporting' clinical decision making from one hospital to another, while allowing for differences in patient population between hospitals. By applying pathway simulation and machine learning together, we found a realistic ceiling of 15%-25% use of thrombolysis across different hospitals and, in the seven hospitals studied, a realistic opportunity to double the number of patients with no significant disability that may be attributed to thrombolysis. CONCLUSIONS: National clinical audit may be enhanced by a combination of pathway simulation and machine learning, which best allows for an understanding of key levers for improvement in hyperacute stroke pathways, allowing for differences between local patient populations. These models, based on standard clinical audit data, may be applied at scale while providing results at individual hospital level. The models facilitate understanding of variation and levers for improvement in stroke pathways, and help set realistic targets tailored to local populations.