A Smartphone-Based Model of Care to Support Patients With Cardiac Disease Transitioning From Hospital to the Community (TeleClinical Care): Pilot Randomized Controlled Trial

Background

Patients hospitalized with acute coronary syndrome (ACS) or heart failure (HF) are frequently readmitted. This is the first randomized controlled trial of a mobile health intervention that combines telemonitoring and education for inpatients with ACS or HF to prevent readmission.

Objective

This study aims to investigate the feasibility, efficacy, and cost-effectiveness of a smartphone app–based model of care (TeleClinical Care [TCC]) in patients discharged after ACS or HF admission.

Methods

In this pilot, 2-center randomized controlled trial, TCC was applied at discharge along with usual care to intervention arm participants. Control arm participants received usual care alone. Inclusion criteria were current admission with ACS or HF, ownership of a compatible smartphone, age ≥18 years, and provision of informed consent. The primary end point was the incidence of unplanned 30-day readmissions. Secondary end points included all-cause readmissions, cardiac readmissions, cardiac rehabilitation completion, medication adherence, cost-effectiveness, and user satisfaction. Intervention arm participants received the app and Bluetooth-enabled devices for measuring weight, blood pressure, and physical activity daily plus usual care. The devices automatically transmitted recordings to the patients’ smartphones and a central server. Thresholds for blood pressure, heart rate, and weight were determined by the treating cardiologists. Readings outside these thresholds were flagged to a monitoring team, who discussed salient abnormalities with the patients’ usual care providers (cardiologists, general practitioners, or HF outreach nurses), who were responsible for further management. The app also provided educational push notifications. Participants were followed up after 6 months.

Results

Overall, 164 inpatients were randomized (TCC: 81/164, 49.4%; control: 83/164, 50.6%; mean age 61.5, SD 12.3 years; 130/164, 79.3% men; 128/164, 78% admitted with ACS). There were 11 unplanned 30-day readmissions in both groups (P=.97). Over a mean follow-up of 193 days, the intervention was associated with a significant reduction in unplanned hospital readmissions (21 in TCC vs 41 in the control arm; P=.02), including cardiac readmissions (11 in TCC vs 25 in the control arm; P=.03), and higher rates of cardiac rehabilitation completion (20/51, 39% vs 9/49, 18%; P=.03) and medication adherence (57/76, 75% vs 37/74, 50%; P=.002). The average usability rating for the app was 4.5/5. The intervention cost Aus $6028 (US $4342.26) per cardiac readmission saved. When modeled in a mainstream clinical setting, enrollment of 237 patients was projected to have the same expenditure compared with usual care, and enrollment of 500 patients was projected to save approximately Aus $100,000 (approximately US $70,000) annually.

Conclusions

TCC was feasible and safe for inpatients with either ACS or HF. The incidence of 30-day readmissions was similar; however, long-term benefits were demonstrated, including fewer readmissions over 6 months, improved medication adherence, and improved cardiac rehabilitation completion.

Trial Registration

Australian New Zealand Clinical Trials Registry ACTRN12618001547235; https://www.anzctr.org.au/Trial/Registration/TrialReview.aspx?id=375945

Process Evaluation of a Randomised Controlled Trial for TeleClinical Care, a Smartphone-App Based Model of Care

Background

A novel smartphone app-based model of care (TeleClinical Care – TCC) for patients with acute coronary syndrome (ACS) and heart failure (HF) was evaluated in a two-site, pilot randomised control trial of 164 participants in Sydney, Australia. The program included a telemonitoring system whereby abnormal blood pressure, weight and heart rate readings were monitored by a central clinical team, who subsequently referred clinically significant alerts to the patients' usual general practitioner (GP, also known as primary care physician in the United States), HF nurse or cardiologist. While the primary endpoint, 30-day readmissions, was neutral, intervention arm participants demonstrated improvements in readmission rates over 6 months, cardiac rehabilitation (CR) completion and medication compliance. A process evaluation was designed to identify contextual factors and mechanisms that influenced the results, as well as strategies of improving site and participant recruitment and the delivery of the intervention, for a planned larger effectiveness trial of over 1,000 patients across the state of New South Wales, Australia (TCC-Cardiac).

Methods

Multiple data sources were used in this mixed-methods process evaluation, including interviews with four TCC team members, three GPs and three cardiologists. CR completion rates, HF outreach service (HFOS) referrals and cardiologist follow-up appointments were audited. A patient questionnaire was also analysed for evidence of improved self-care as a hypothesised mechanism of the TCC app. An implementation research logic model was used to synthesise our findings.

Results

Rates of HFOS referral (83 vs. 72%) and cardiologist follow-up (96 vs. 93%) were similarly high in the intervention and control arms, respectively. Team members were largely positive towards their orientation and training, but highlighted several implementation strategies that could be optimised for TCC-Cardiac: streamlining of the enrolment process, improving the reach of the trial by screening patients in non-cardiac wards, and ensuring team members had adequate time to recruit (>15 h per week). GPs and cardiologists viewed the intervention acceptably regarding potential benefit of closely monitoring, and responding to abnormalities for their patients, though there were concerns of the potential additional workload generated by alerts that did not merit clinical intervention. Clear delineation of which clinician (GP or cardiologist) was primarily responsible for alert management was also recommended, as well as a preference to receive regular summary data. Several patients commented on the mechanisms of improved self-management because of TCC, which could have led to the outcome of improved medication compliance.

Discussion

Use of TCC was associated with several benefits, including higher patient engagement and completion rates with CR. The conduct and delivery of TCC-Cardiac will be improved by the findings of this process evaluation to optimise recruitment, and establishing the roles of GPs and cardiologists as part of the model.

Prediction models for hospital readmissions in patients with heart disease: a systematic review and meta-analysis

OBJECTIVE

To describe the discrimination and calibration of clinical prediction models, identify characteristics that contribute to better predictions and investigate predictors that are associated with unplanned hospital readmissions.

DESIGN

Systematic review and meta-analysis.

DATA SOURCE

Medline, EMBASE, ICTPR (for study protocols) and Web of Science (for conference proceedings) were searched up to 25 August 2020.

ELIGIBILITY CRITERIA FOR SELECTING STUDIES

Studies were eligible if they reported on (1) hospitalised adult patients with acute heart disease; (2) a clinical presentation of prediction models with c-statistic; (3) unplanned hospital readmission within 6 months.

PRIMARY AND SECONDARY OUTCOME MEASURES

Model discrimination for unplanned hospital readmission within 6 months measured using concordance (c) statistics and model calibration. Meta-regression and subgroup analyses were performed to investigate predefined sources of heterogeneity. Outcome measures from models reported in multiple independent cohorts and similarly defined risk predictors were pooled.

RESULTS

Sixty studies describing 81 models were included: 43 models were newly developed, and 38 were externally validated. Included populations were mainly patients with heart failure (HF) (n=29). The average age ranged between 56.5 and 84 years. The incidence of readmission ranged from 3% to 43%. Risk of bias (RoB) was high in almost all studies. The c-statistic was <0.7 in 72 models, between 0.7 and 0.8 in 16 models and >0.8 in 5 models. The study population, data source and number of predictors were significant moderators for the discrimination. Calibration was reported for 27 models. Only the GRACE (Global Registration of Acute Coronary Events) score had adequate discrimination in independent cohorts (0.78, 95% CI 0.63 to 0.86). Eighteen predictors were pooled.

CONCLUSION

Some promising models require updating and validation before use in clinical practice. The lack of independent validation studies, high RoB and low consistency in measured predictors limit their applicability.

PROSPERO REGISTRATION NUMBER

CRD42020159839.

Published models that predict hospital readmission: a critical appraisal

INTRODUCTION

The number of readmission risk prediction models available has increased rapidly, and these models are used extensively for health decision-making. Unfortunately, readmission models can be subject to flaws in their development and validation, as well as limitations in their clinical usefulness.

OBJECTIVE

To critically appraise readmission models in the published literature using Delphi-based recommendations for their development and validation.

METHODS

We used the modified Delphi process to create Critical Appraisal of Models that Predict Readmission (CAMPR), which lists expert recommendations focused on development and validation of readmission models. Guided by CAMPR, two researchers independently appraised published readmission models in two recent systematic reviews and concurrently extracted data to generate reference lists of eligibility criteria and risk factors.

RESULTS

We found that published models (n=81) followed 6.8 recommendations (45%) on average. Many models had weaknesses in their development, including failure to internally validate (12%), failure to account for readmission at other institutions (93%), failure to account for missing data (68%), failure to discuss data preprocessing (67%) and failure to state the model's eligibility criteria (33%).

CONCLUSIONS

The high prevalence of weaknesses in model development identified in the published literature is concerning, as these weaknesses are known to compromise predictive validity. CAMPR may support researchers, clinicians and administrators to identify and prevent future weaknesses in model development.

Development of Electronic Health Record-Based Prediction Models for 30-Day Readmission Risk Among Patients Hospitalized for Acute Myocardial Infarction

IMPORTANCE

In the US, more than 600 000 adults will experience an acute myocardial infarction (AMI) each year, and up to 20% of the patients will be rehospitalized within 30 days. This study highlights the need for consideration of calibration in these risk models.

OBJECTIVE

To compare multiple machine learning risk prediction models using an electronic health record (EHR)-derived data set standardized to a common data model.

DESIGN, SETTING, AND PARTICIPANTS

This was a retrospective cohort study that developed risk prediction models for 30-day readmission among all inpatients discharged from Vanderbilt University Medical Center between January 1, 2007, and December 31, 2016, with a primary diagnosis of AMI who were not transferred from another facility. The model was externally validated at Dartmouth-Hitchcock Medical Center from April 2, 2011, to December 31, 2016. Data analysis occurred between January 4, 2019, and November 15, 2020.

EXPOSURES

Acute myocardial infarction that required hospital admission.

MAIN OUTCOMES AND MEASURES

The main outcome was thirty-day hospital readmission. A total of 141 candidate variables were considered from administrative codes, medication orders, and laboratory tests. Multiple risk prediction models were developed using parametric models (elastic net, least absolute shrinkage and selection operator, and ridge regression) and nonparametric models (random forest and gradient boosting). The models were assessed using holdout data with area under the receiver operating characteristic curve (AUROC), percentage of calibration, and calibration curve belts.

RESULTS

The final Vanderbilt University Medical Center cohort included 6163 unique patients, among whom the mean (SD) age was 67 (13) years, 4137 were male (67.1%), 1019 (16.5%) were Black or other race, and 933 (15.1%) were rehospitalized within 30 days. The final Dartmouth-Hitchcock Medical Center cohort included 4024 unique patients, with mean (SD) age of 68 (12) years; 2584 (64.2%) were male, 412 (10.2%) were rehospitalized within 30 days, and most of the cohort were non-Hispanic and White. The final test set AUROC performance was between 0.686 to 0.695 for the parametric models and 0.686 to 0.704 for the nonparametric models. In the validation cohort, AUROC performance was between 0.558 to 0.655 for parametric models and 0.606 to 0.608 for nonparametric models.

CONCLUSIONS AND RELEVANCE

In this study, 5 machine learning models were developed and externally validated to predict 30-day readmission AMI hospitalization. These models can be deployed within an EHR using routinely collected data.

Estimating the health loss due to poor engagement with cardiac rehabilitation in Australia

BACKGROUND

Cardiac rehabilitation (CR) programs are effective in reducing cardiovascular mortality and readmissions. However, most patients are denied the benefits of CR due to low referral rates. Of those patients referred, commencement rates vary from 28.4% to 60%. This paper quantifies the scale of health loss in Australia due to poor engagement with the program, and estimates how much public funding can be justifiably reallocated to address the problem.

METHODS

Economic decision modelling was undertaken to estimate the expected lifetime health loss and costs to Medicare. Key parameters were derived from Australian databases, CR registries and meta-analyses. Population health gains associated with uptake rates of 60%, and 85% were calculated.

RESULTS

CR was associated with a 99.9% probability of being cost-effective, even at a cost-effectiveness threshold lower than conventionally applied. Importantly, an average of 0.52 years of life expectancy are lost due to national uptake being below 60% achieved in some best performing programs in Australia, equivalent to 0.28 quality adjusted life years. The analysis indicates that $12.9 million/year could be justifiably reallocated from public funds to achieve a national uptake rate of 60%, while maintaining cost-effectiveness of CR due to the large health gains that would be expected.

CONCLUSION

CR is a cost-effective service for patients with coronary heart disease. In Australia, less than a third of patients commence CR, potentially resulting in avoidable patient harm. Additional investment in CR is vital and should be a national priority as the health gains for patients far outweigh the costs.

Implementation and Continuous Monitoring of an Electronic Health Record Embedded Readmissions Clinical Decision Support Tool

Unplanned hospital readmissions represent a significant health care value problem with high costs and poor quality of care. A significant percentage of readmissions could be prevented if clinical inpatient teams were better able to predict which patients were at higher risk for readmission. Many of the current clinical decision support models that predict readmissions are not configured to integrate closely with the electronic health record or alert providers in real-time prior to discharge about a patient's risk for readmission. We report on the implementation and monitoring of the Epic electronic health record-"Unplanned readmission model version 1"-over 2 years from 1/1/2018-12/31/2019. For patients discharged during this time, the predictive capability to discern high risk discharges was reflected in an AUC/C-statistic at our three hospitals of 0.716-0.760 for all patients and 0.676-0.695 for general medicine patients. The model had a positive predictive value ranging from 0.217-0.248 for all patients. We also present our methods in monitoring the model over time for trend changes, as well as common readmissions reduction strategies triggered by the score.

Acute Myocardial Infarction Readmission Risk Prediction Models: A Systematic Review of Model Performance

BACKGROUND

Hospitals are subject to federal financial penalties for excessive 30-day hospital readmissions for acute myocardial infarction (AMI). Prospectively identifying patients hospitalized with AMI at high risk for readmission could help prevent 30-day readmissions by enabling targeted interventions. However, the performance of AMI-specific readmission risk prediction models is unknown.

METHODS AND RESULTS

We systematically searched the published literature through March 2017 for studies of risk prediction models for 30-day hospital readmission among adults with AMI. We identified 11 studies of 18 unique risk prediction models across diverse settings primarily in the United States, of which 16 models were specific to AMI. The median overall observed all-cause 30-day readmission rate across studies was 16.3% (range, 10.6%-21.0%). Six models were based on administrative data; 4 on electronic health record data; 3 on clinical hospital data; and 5 on cardiac registry data. Models included 7 to 37 predictors, of which demographics, comorbidities, and utilization metrics were the most frequently included domains. Most models, including the Centers for Medicare and Medicaid Services AMI administrative model, had modest discrimination (median C statistic, 0.65; range, 0.53-0.79). Of the 16 reported AMI-specific models, only 8 models were assessed in a validation cohort, limiting generalizability. Observed risk-stratified readmission rates ranged from 3.0% among the lowest-risk individuals to 43.0% among the highest-risk individuals, suggesting good risk stratification across all models.

CONCLUSIONS

Current AMI-specific readmission risk prediction models have modest predictive ability and uncertain generalizability given methodological limitations. No existing models provide actionable information in real time to enable early identification and risk-stratification of patients with AMI before hospital discharge, a functionality needed to optimize the potential effectiveness of readmission reduction interventions.

POLAR Diversion: Using General Practice Data to Calculate Risk of Emergency Department Presentation at the Time of Consultation

OBJECTIVE

This project examined and produced a general practice (GP) based decision support tool (DST), namely POLAR Diversion, to predict a patient's risk of emergency department (ED) presentation. The tool was built using both GP/family practice and ED data, but is designed to operate on GP data alone.

METHODS

GP data from 50 practices during a defined time frame were linked with three local EDs. Linked data and data mapping were used to develop a machine learning DST to determine a range of variables that, in combination, led to predictive patient ED presentation risk scores. Thirteen percent of the GP data was kept as a control group and used to validate the tool.

RESULTS

The algorithm performed best in predicting the risk of attending ED within the 30-day time category, and also in the no ED attendance tests, suggesting few false positives. At 0 to 30 days the positive predictive value (PPV) was 74%, with a sensitivity/recall of 68%. Non-ED attendance had a PPV of 82% and sensitivity/recall of 96%.

CONCLUSION

Findings indicate that the POLAR Diversion algorithm performed better than previously developed tools, particularly in the 0 to 30 day time category. Its utility increases because of it being based on the data within the GP system alone, with the ability to create real-time "in consultation" warnings. The tool will be deployed across GPs in Australia, allowing us to assess the clinical utility, and data quality needs in further iterations.

Readmissions following hospitalisations for cardiovascular disease: a scoping review of the Australian literature

Objective International studies suggest high rates of readmissions after cardiovascular hospitalisations, but the burden in Australia is uncertain. We summarised the characteristics, frequency, risk factors of readmissions and interventions to reduce readmissions following cardiovascular hospitalisation in Australia. Methods A scoping review of the published literature from 2000-2016 was performed using Medline, EMBASE and Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases and relevant grey literature. Results We identified 35 studies (25 observational, 10 reporting outcomes of interventions). Observational studies were typically single-centre (11/25) and reported readmissions following hospitalisations for heart failure (HF; 10/25), acute coronary syndrome (7/25) and stroke (6/25), with other conditions infrequently reported. The definition of a readmission was heterogeneous and was assessed using diverse methods. Readmission rate, most commonly reported at 1 month (14/25), varied from 6.3% to 27%, with readmission rates of 10.1-27% for HF, 6.5-11% for stroke and 12.7-17% for acute myocardial infarction, with a high degree of heterogeneity among studies. Of the 10 studies of interventions to reduce readmissions, most (n=8) evaluated HF management programs and three reported a significant reduction in readmissions. We identified a lack of national studies of readmissions and those assessing the cost and resource impact of readmissions on the healthcare system as well as a paucity of successful interventions to lower readmissions. Conclusions High rates of readmissions are reported for cardiovascular conditions, although substantial methodological heterogeneity exists among studies. Nationally standardised definitions are required to accurately measure readmissions and further studies are needed to address knowledge gaps and test interventions to lower readmissions in Australia. What is known about the topic? International studies suggest readmissions are common following cardiovascular hospitalisations and are costly to the health system, yet little is known about the burden of readmission in the Australian setting or the effectiveness of intervention to reduce readmissions. What does this paper add? We found relatively high rates of readmissions following common cardiovascular conditions although studies differed in their methodology making it difficult to accurately gauge the readmission rate. We also found several knowledge gaps including lack of national studies, studies assessing the impact on the health system and few interventions proven to reduce readmissions in the Australian setting. What are the implications for practitioners? Practitioners should be cautious when interpreting studies of readmissions due the heterogeneity in definitions and methods used in Australian literature. Further studies are needed to test interventions to reduce readmissions in the Australians setting.

Development of Quality Indicators for Cardiac Rehabilitation in Australia: A Modified Delphi Method and Pilot Test

BACKGROUND

International guidelines recommend cardiac rehabilitation (CR) for secondary prevention of cardiovascular disease, however, it is underutilised and the quality of content and delivery varies widely. Quality indicators (QIs) for CR are used internationally to measure clinical practice performance, but are lacking in the Australian context. This study reports the development of QIs for minimum dataset (MDS) for CR and the results of a pilot test for feasibility and applicability in clinical practice in Australia.

METHODS

A modified Delphi method was used to develop initial QIs which involved a consensus approach through a series of face-to-face and teleconference meetings of an expert multidisciplinary panel (n=8), supplemented by an environmental scan of the literature and a multi-site pilot test.

RESULTS

Eight (8) QIs were proposed and sent to CR clinicians (n=250) electronically to rate importance, current data collection status, and feasibility of future collection. The top six of these QIs were selected with an additional two key performance indicators from the New South Wales (NSW) Ministry of Health and two QIs from international registers for a draft MDS. The pilot test in 16 sites (938 patient cases) demonstrated median performance of 93% (IQR 47.1-100%). All 10 QIs were retained and one further QI related to diabetes was added for a final draft MDS.

CONCLUSIONS

The MDS of 11 QIs for CR provides an important foundation for collection of data to promote the quality of CR nationally and the opportunity to participate in international benchmarking.

Clinical Profile of Acute Myocardial Infarction Patients Included in the Hospital Readmissions Reduction Program

Background Medicare's Hospital Readmissions Reduction Program assesses financial penalties to hospitals based on risk-standardized readmission rates after specific episodes of care, including acute myocardial infarction. Detailed information about the type of patients included in the penalty is unknown. Methods and Results Starting with administrative data from Medicare, we conducted physician-adjudicated chart reviews of all patients considered 30-day readmissions after acute myocardial infarction from July 2012 to June 2015. Of 197 readmissions, 68 (34.5%) received percutaneous coronary intervention and 18 (9.1%) underwent coronary artery bypass grafting on index hospitalization. The remaining 111 patients did not receive any intervention. Of the 197 patients, 56 patients (28.4%) were considered too high risk for invasive management, 23 (11.7%) had nonobstructive coronary artery disease on diagnostic catheterization and therefore no indication for revascularization, 19 patients had a type II myocardial infarction (9.6%) for which noninvasive, outpatient workup was recommended, and 13 (6.6%) declined further care. The most common readmission diagnoses were cardiac causes and noncardiac chest discomfort, infection, and gastrointestinal bleeding. Conclusions Our results demonstrate that more than a quarter of the patients included in the penalty do not receive revascularization either because of provider assessment of risk or patient preference, and nearly one tenth have type II myocardial infarction. As such, administrative codes for prohibitive procedural risk, patient-initiated "do not resuscitate" status, or type II myocardial infarction may improve the risk-adjustment of the metric. Furthermore, provider organizations seeking to reduce readmission rates should focus resources on the needs of these patients, such as care coordination, hospice services when requested by patients, and treatment of noncardiac conditions.

Prediction of Incident Hypertension Within the Next Year: Prospective Study Using Statewide Electronic Health Records and Machine Learning

BACKGROUND

As a high-prevalence health condition, hypertension is clinically costly, difficult to manage, and often leads to severe and life-threatening diseases such as cardiovascular disease (CVD) and stroke.

OBJECTIVE

The aim of this study was to develop and validate prospectively a risk prediction model of incident essential hypertension within the following year.

METHODS

Data from individual patient electronic health records (EHRs) were extracted from the Maine Health Information Exchange network. Retrospective (N=823,627, calendar year 2013) and prospective (N=680,810, calendar year 2014) cohorts were formed. A machine learning algorithm, XGBoost, was adopted in the process of feature selection and model building. It generated an ensemble of classification trees and assigned a final predictive risk score to each individual.

RESULTS

The 1-year incident hypertension risk model attained areas under the curve (AUCs) of 0.917 and 0.870 in the retrospective and prospective cohorts, respectively. Risk scores were calculated and stratified into five risk categories, with 4526 out of 381,544 patients (1.19%) in the lowest risk category (score 0-0.05) and 21,050 out of 41,329 patients (50.93%) in the highest risk category (score 0.4-1) receiving a diagnosis of incident hypertension in the following 1 year. Type 2 diabetes, lipid disorders, CVDs, mental illness, clinical utilization indicators, and socioeconomic determinants were recognized as driving or associated features of incident essential hypertension. The very high risk population mainly comprised elderly (age>50 years) individuals with multiple chronic conditions, especially those receiving medications for mental disorders. Disparities were also found in social determinants, including some community-level factors associated with higher risk and others that were protective against hypertension.

CONCLUSIONS

With statewide EHR datasets, our study prospectively validated an accurate 1-year risk prediction model for incident essential hypertension. Our real-time predictive analytic model has been deployed in the state of Maine, providing implications in interventions for hypertension and related diseases and hopefully enhancing hypertension care.

Early Readmissions After Acute Myocardial Infarction

This study aims to evaluate the rate, predictors, and causes of 30-day readmissions in a single tertiary hospital in the United Kingdom. We conducted a retrospective study of all patients admitted between 2012 and 2014 with a diagnosis of acute myocardial infarction, who were in the Myocardial Infarction National Audit Project register. Data on patient demographics, comorbidities, care received, and in-hospital mortality were collected. Rates of 30-day readmission and causes of readmission were evaluated. Univariate and multiple logistic regressions were used to identify predictors of all-cause, cardiac, and noncardiac readmission. A total of 1,869 patients were included in the analysis and 171 had an unplanned readmission with 30 days (9%). Noncardiac problems represented half of all readmissions with the dominant cause noncardiac chest pain (50%). A variety of other noncardiac causes for readmission were identified and the most common were lower respiratory tract infection (4.3%), gastrointestinal problems (4.9%), bleeding (3.7%), dizziness, syncope, or fall (3.0%), and pulmonary embolus (2.4%). For cardiac causes of readmissions, common causes included acute coronary syndrome (17.1%), stable angina (11.6%), and heart failure (9.8%). Readmitted patients were more likely to be older, anemic, and less likely to receive coronary angiogram and percutaneous coronary intervention. After adjustment, the only predictor of all-cause readmission was older age. For noncardiac readmission, previous myocardial infarction was associated with significantly fewer readmissions. Our results suggest that early readmission after discharge with diagnosis of acute myocardial infarction is common. Chest pain is the most frequent cause of readmission, and interventions to reduce noncardiac chest pain admissions are needed.

Using the Electronic Medical Record to Identify Patients at High Risk for Frequent Emergency Department Visits and High System Costs

BACKGROUND

A small proportion of patients account for a high proportion of healthcare use. Accurate preemptive identification may facilitate tailored intervention. We sought to determine whether machine learning techniques using text from a family practice electronic medical record can be used to predict future high emergency department use and total costs by patients who are not yet high emergency department users or high cost to the healthcare system.

METHODS

Text from fields of the cumulative patient profile within an electronic medical record of 43,111 patients was indexed. Separate training and validation cohorts were created. After processing, 11,905 words were used to fit a logistic regression model. The primary outcomes of interest in the 12 months after prediction were 3 or more emergency department visits and being in the top 5% in healthcare expenditures. Outcomes were assessed through linkage to administrative databases housed at the Institute for Clinical Evaluative Sciences.

RESULTS

In the model to predict frequent emergency department visits, after excluding patients who were high emergency department users in the previous year, the area under the receiver operating characteristic curve was 0.71. By using the same methodology, the model to predict the top 5% in total system costs had an area under the receiver operating characteristic curve of 0.76.

CONCLUSIONS

Machine learning techniques can be applied to analyze free text contained in electronic medical records. This dataset is more predictive of patients who will generate future high costs than future emergency department visits. It remains to be seen whether these predictions can be used to reduce costs by early interventions in this cohort of patients.

A Framework for Mixed-Type Multioutcome Prediction With Applications in Healthcare

Health analysis often involves prediction of multiple outcomes of mixed type. The existing work is restrictive to either a limited number or specific outcome types. We propose a framework for mixed-type multioutcome prediction. Our proposed framework proposes a cumulative loss function composed of a specific loss function for each outcome type-as an example, least square (continuous outcome), hinge (binary outcome), Poisson (count outcome), and exponential (nonnegative outcome). To model these outcomes jointly, we impose a commonality across the prediction parameters through a common matrix normal prior. The framework is formulated as iterative optimization problems and solved using an efficient block-coordinate descent method. We empirically demonstrate both scalability and convergence. We apply the proposed model to a synthetic dataset and then on two real-world cohorts: a cancer cohort and an acute myocardial infarction cohort collected over a two-year period. We predict multiple emergency-related outcomes-as example, future emergency presentations (binary), emergency admissions (count), emergency length of stay days (nonnegative), and emergency time to next admission day (nonnegative). We show that the predictive performance of the proposed model is better than several state-of-the-art baselines.

Predicting mortality over different time horizons: which data elements are needed?

OBJECTIVE

Electronic health records (EHRs) are a resource for "big data" analytics, containing a variety of data elements. We investigate how different categories of information contribute to prediction of mortality over different time horizons among patients undergoing hemodialysis treatment.

MATERIAL AND METHODS

We derived prediction models for mortality over 7 time horizons using EHR data on older patients from a national chain of dialysis clinics linked with administrative data using LASSO (least absolute shrinkage and selection operator) regression. We assessed how different categories of information relate to risk assessment and compared discrete models to time-to-event models.

RESULTS

The best predictors used all the available data (c-statistic ranged from 0.72-0.76), with stronger models in the near term. While different variable groups showed different utility, exclusion of any particular group did not lead to a meaningfully different risk assessment. Discrete time models performed better than time-to-event models.

CONCLUSIONS

Different variable groups were predictive over different time horizons, with vital signs most predictive for near-term mortality and demographic and comorbidities more important in long-term mortality.

Utility of models to predict 28-day or 30-day unplanned hospital readmissions: an updated systematic review

OBJECTIVE

To update previous systematic review of predictive models for 28-day or 30-day unplanned hospital readmissions.

DESIGN

Systematic review.

SETTING/DATA SOURCE

CINAHL, Embase, MEDLINE from 2011 to 2015.

PARTICIPANTS

All studies of 28-day and 30-day readmission predictive model.

OUTCOME MEASURES

Characteristics of the included studies, performance of the identified predictive models and key predictive variables included in the models.

RESULTS

Of 7310 records, a total of 60 studies with 73 unique predictive models met the inclusion criteria. The utilisation outcome of the models included all-cause readmissions, cardiovascular disease including pneumonia, medical conditions, surgical conditions and mental health condition-related readmissions. Overall, a wide-range C-statistic was reported in 56/60 studies (0.21-0.88). 11 of 13 predictive models for medical condition-related readmissions were found to have consistent moderate discrimination ability (C-statistic ≥0.7). Only two models were designed for the potentially preventable/avoidable readmissions and had C-statistic >0.8. The variables 'comorbidities', 'length of stay' and 'previous admissions' were frequently cited across 73 models. The variables 'laboratory tests' and 'medication' had more weight in the models for cardiovascular disease and medical condition-related readmissions.

CONCLUSIONS

The predictive models which focused on general medical condition-related unplanned hospital readmissions reported moderate discriminative ability. Two models for potentially preventable/avoidable readmissions showed high discriminative ability. This updated systematic review, however, found inconsistent performance across the included unique 73 risk predictive models. It is critical to define clearly the utilisation outcomes and the type of accessible data source before the selection of the predictive model. Rigorous validation of the predictive models with moderate-to-high discriminative ability is essential, especially for the two models for the potentially preventable/avoidable readmissions. Given the limited available evidence, the development of a predictive model specifically for paediatric 28-day all-cause, unplanned hospital readmissions is a high priority.

Opportunities and challenges in developing risk prediction models with electronic health records data: a systematic review

OBJECTIVE

Electronic health records (EHRs) are an increasingly common data source for clinical risk prediction, presenting both unique analytic opportunities and challenges. We sought to evaluate the current state of EHR based risk prediction modeling through a systematic review of clinical prediction studies using EHR data.

METHODS

We searched PubMed for articles that reported on the use of an EHR to develop a risk prediction model from 2009 to 2014. Articles were extracted by two reviewers, and we abstracted information on study design, use of EHR data, model building, and performance from each publication and supplementary documentation.

RESULTS

We identified 107 articles from 15 different countries. Studies were generally very large (median sample size = 26 100) and utilized a diverse array of predictors. Most used validation techniques (n = 94 of 107) and reported model coefficients for reproducibility (n = 83). However, studies did not fully leverage the breadth of EHR data, as they uncommonly used longitudinal information (n = 37) and employed relatively few predictor variables (median = 27 variables). Less than half of the studies were multicenter (n = 50) and only 26 performed validation across sites. Many studies did not fully address biases of EHR data such as missing data or loss to follow-up. Average c-statistics for different outcomes were: mortality (0.84), clinical prediction (0.83), hospitalization (0.71), and service utilization (0.71).

CONCLUSIONS

EHR data present both opportunities and challenges for clinical risk prediction. There is room for improvement in designing such studies.

Stabilizing l1-norm prediction models by supervised feature grouping

Emerging Electronic Medical Records (EMRs) have reformed the modern healthcare. These records have great potential to be used for building clinical prediction models. However, a problem in using them is their high dimensionality. Since a lot of information may not be relevant for prediction, the underlying complexity of the prediction models may not be high. A popular way to deal with this problem is to employ feature selection. Lasso and l1-norm based feature selection methods have shown promising results. But, in presence of correlated features, these methods select features that change considerably with small changes in data. This prevents clinicians to obtain a stable feature set, which is crucial for clinical decision making. Grouping correlated variables together can improve the stability of feature selection, however, such grouping is usually not known and needs to be estimated for optimal performance. Addressing this problem, we propose a new model that can simultaneously learn the grouping of correlated features and perform stable feature selection. We formulate the model as a constrained optimization problem and provide an efficient solution with guaranteed convergence. Our experiments with both synthetic and real-world datasets show that the proposed model is significantly more stable than Lasso and many existing state-of-the-art shrinkage and classification methods. We further show that in terms of prediction performance, the proposed method consistently outperforms Lasso and other baselines. Our model can be used for selecting stable risk factors for a variety of healthcare problems, so it can assist clinicians toward accurate decision making.

Is demography destiny? Application of machine learning techniques to accurately predict population health outcomes from a minimal demographic dataset

For years, we have relied on population surveys to keep track of regional public health statistics, including the prevalence of non-communicable diseases. Because of the cost and limitations of such surveys, we often do not have the up-to-date data on health outcomes of a region. In this paper, we examined the feasibility of inferring regional health outcomes from socio-demographic data that are widely available and timely updated through national censuses and community surveys. Using data for 50 American states (excluding Washington DC) from 2007 to 2012, we constructed a machine-learning model to predict the prevalence of six non-communicable disease (NCD) outcomes (four NCDs and two major clinical risk factors), based on population socio-demographic characteristics from the American Community Survey. We found that regional prevalence estimates for non-communicable diseases can be reasonably predicted. The predictions were highly correlated with the observed data, in both the states included in the derivation model (median correlation 0.88) and those excluded from the development for use as a completely separated validation sample (median correlation 0.85), demonstrating that the model had sufficient external validity to make good predictions, based on demographics alone, for areas not included in the model development. This highlights both the utility of this sophisticated approach to model development, and the vital importance of simple socio-demographic characteristics as both indicators and determinants of chronic disease.