A practical method for predicting frequent use of emergency department care using routinely available electronic registration data

Exploring Which Patients Use Their Closest Emergency Departments Using Geocoded Data

BACKGROUND

Each year, roughly 20% of U.S. adults visit an emergency department (ED), but little is known about patients' choice of ED.

OBJECTIVES

Examine the discretion patients have to choose among EDs, characteristics associated with ED choice, and relationship between ED choice and self-reported care experiences of ED patients.

METHODS

We surveyed adult patients discharged to the community (DTC) in January-March 2018 from 16 geographically dispersed hospital-based EDs, geocoded patient and hospital-based ED addresses within 100 miles of patient addresses, and calculated travel distances. We examined the likelihood of visiting the closest ED based on patient and ED characteristics. Linear regression models examined the association of choosing the closest ED with seven measures of patient experience of care (scaled 0-100), adjusting for patient characteristics.

RESULTS

43.6% of 4647 responding patients visited the ED nearest their home (on average, 5.7 miles away). Patients who chose a farther ED had more urgent conditions, were more educated, and were less likely to be non-Hispanic White. They were significantly more likely to have visited an ED in a higher-rated, metropolitan, network hospital with major teaching status, a cardiac intensive care unit, and a certified trauma center. Patients who chose a farther ED were more likely to recommend that ED, with "medium-to-large" differences in scores (+4.3% more selected "definitely yes", p < 0.05).

CONCLUSIONS

Fewer than half of patients visited the closest ED. Patients who chose a farther ED tended to seek higher-rated hospitals and report more favorable experiences.

A hybrid machine learning framework to improve prediction of all-cause rehospitalization among elderly patients in Hong Kong

BACKGROUND

Accurately estimating elderly patients' rehospitalisation risk benefits clinical decisions and service planning. However, research in rehospitalisation and repeated hospitalisation yielded only models with modest performance, and the model performance deteriorates rapidly as the prediction timeframe expands beyond 28 days and for older participants.

METHODS

A temporal zero-inflated Poisson (tZIP) regression model was developed and validated retrospectively and prospectively. The data of the electronic health records (EHRs) contain cohorts (aged 60+) in a major public hospital in Hong Kong. Two temporal offset functions accounted for the associations between exposure time and parameters corresponding to the zero-inflated logistic component and the Poisson distribution's expected count. tZIP was externally validated with a retrospective cohort's rehospitalisation events up to 12 months after the discharge date. Subsequently, tZIP was validated prospectively after piloting its implementation at the study hospital. Patients discharged within the pilot period were tagged, and the proposed model's prediction of their rehospitalisation was verified monthly. Using a hybrid machine learning (ML) approach, the tZIP-based risk estimator's marginal effect on 28-day rehospitalisation was further validated, competing with other factors representing different post-acute and clinical statuses.

RESULTS

The tZIP prediction of rehospitalisation from 28 days to 365 days was achieved at above 80% discrimination accuracy retrospectively and prospectively in two out-of-sample cohorts. With a large margin, it outperformed the Cox proportional and linear models built with the same predictors. The hybrid ML revealed that the risk estimator's contribution to 28-day rehospitalisation outweighed other features relevant to service utilisation and clinical status.

CONCLUSIONS

A novel rehospitalisation risk model was introduced, and its risk estimators, whose importance outweighed all other factors of diverse post-acute care and clinical conditions, were derived. The proposed approach relies on four easily accessible variables easily extracted from EHR. Thus, clinicians could visualise patients' rehospitalisation risk from 28 days to 365 days after discharge and screen high-risk older patients for follow-up care at the proper time.

The Data-Adaptive Fellegi-Sunter Model for Probabilistic Record Linkage: Algorithm Development and Validation for Incorporating Missing Data and Field Selection

Background

Quality patient care requires comprehensive health care data from a broad set of sources. However, missing data in medical records and matching field selection are 2 real-world challenges in patient-record linkage.

Objective

In this study, we aimed to evaluate the extent to which incorporating the missing at random (MAR)–assumption in the Fellegi-Sunter model and using data-driven selected fields improve patient-matching accuracy using real-world use cases.

Methods

We adapted the Fellegi-Sunter model to accommodate missing data using the MAR assumption and compared the adaptation to the common strategy of treating missing values as disagreement with matching fields specified by experts or selected by data-driven methods. We used 4 use cases, each containing a random sample of record pairs with match statuses ascertained by manual reviews. Use cases included health information exchange (HIE) record deduplication, linkage of public health registry records to HIE, linkage of Social Security Death Master File records to HIE, and deduplication of newborn screening records, which represent real-world clinical and public health scenarios. Matching performance was evaluated using the sensitivity, specificity, positive predictive value, negative predictive value, and F1-score.

Results

Incorporating the MAR assumption in the Fellegi-Sunter model maintained or improved F1-scores, regardless of whether matching fields were expert-specified or selected by data-driven methods. Combining the MAR assumption and data-driven fields optimized the F1-scores in the 4 use cases.

Conclusions

MAR is a reasonable assumption in real-world record linkage applications: it maintains or improves F1-scores regardless of whether matching fields are expert-specified or data-driven. Data-driven selection of fields coupled with MAR achieves the best overall performance, which can be especially useful in privacy-preserving record linkage.

Does living near hospital obscure the association between active living environments and hospitalization?

Hospitals tend to be among the destinations that make densely populated, well-connected neighbourhoods more conducive to active living. In this study, we determined whether living near a hospital distorts the association between living in favourable ALEs and hospitalization for physical inactivity-related cardiometabolic diseases. We used a record linkage of 442,345 respondents of the Canadian Community Health Survey and their hospitalization records for cardiometabolic disease. We then assessed respondents' neighbourhoods using the Canadian Active Living Environments measure (Can-ALE), a measure based on ≥3-way intersection density, residential density, and points of interest. We then calculated the distance in kilometers between the centroids of respondents' assigned dissemination areas and the nearest user-contributed location for hospitals from OpenStreetMap. We monitored changes in estimates for the association between ALEs and odds of cardiometabolic disease hospitalization using a series of logistic regressions with indicator variables for distances to hospital of 500 meters to 10 kilometers. We found that living between 500 meters and six kilometers of a hospital and was associated with modestly higher odds of cardiometabolic hospitalization (OR 1.10, 95% CI 1.02 to 1.18 for 500 meters; OR 1.05, 95% CI 1.01 to 1.09 for six kilometers). Living in more favourable ALEs was associated with lower odds of hospitalization (OR 0.79, 95% CI 0.68 to 0.91; comparing the most favourable to least favourable ALEs). Effect estimates between more favourable ALEs and lower odds of hospitalization were marginally strengthened when living within 2-6 kilometers to a hospital was accounted for. This study demonstrates the importance of disentangling interrelated geographic factors and underlines the potential for built environments to elicit reductions in health care.

National travel distances for emergency care

Background

Most emergency department (ED) patients arrive by their own transport and, for various reasons, may not choose the nearest ED. How far patients travel for ED treatment may reflect both patients’ access to care and severity of illness. In this study, we aimed to examine the travel distance and travel time between a patient’s home and ED they visited and investigate how these distances/times vary by patient and hospital characteristics.

Methods

We randomly sampled and collected data from 14,812 patients discharged to the community (DTC) between January and March 2016 from 50 hospital-based EDs nationwide. We geocoded and calculated the distance and travel time between patient and hospital-based ED addresses, examined the travel distances/ times between patients’ home and the ED they visited, and used mixed-effects regression models to investigate how these distances/times vary by patient and hospital characteristics.

Results

Patients travelled an average of 8.0 (SD = 10.9) miles and 17.3 (SD = 18.0) driving minutes to the ED. Patients travelled significantly farther to avoid EDs in lower performing hospitals (p < 0.01) and in the West (p < 0.05) and Midwest (p < 0.05). Patients travelled farther when visiting EDs in rural areas. Younger patients travelled farther than older patients.

Conclusions

Understanding how far patients are willing to travel is indicative of whether patient populations have adequate access to ED services. By showing that patients travel farther to avoid a low-performing hospital, we provide evidence that DTC patients likely do exercise some choice among EDs, indicating some market incentives for higher-quality care, even for some ED admissions. Understanding these issues will help policymakers better define access to ED care and assist in directing quality improvement efforts. To our knowledge, our study is the most comprehensive nationwide characterization of patient travel for ED treatment to date.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-022-07743-7.

Can the adverse childhood experiences (ACEs) checklist be utilized to predict emergency department visits among children and adolescents?

Background

Extensive literature has shown an association of Adverse Childhood Experiences (ACEs) with adverse health outcomes; however, its ability to predict events or stratify risks is less known. Individuals with mental illness and ACE exposure have been shown to visit emergency departments (ED) more often than those in the general population. This study thus examined the ability of the ACEs checklist to predict ED visits within the subsequent year among children and adolescents presenting to mental health clinics with pre-existing mental health issues.

Methods

The study analyzed linked data (n = 6100) from two databases provided by Alberta Health Services (AHS). The Regional Access and Intake System (RAIS 2016–2018) database provided data on the predictors (ACE items, age, sex, residence, mental health program type, and primary diagnosis) regarding children and adolescents (aged 0–17 years) accessing addiction and mental health services within Calgary Zone, and the National Ambulatory Care Reporting System (NACRS 2016–2019) database provided data on ED visits. A 25% random sample of the data was reserved for validation purposes. Two Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression models, each employing a different method to tune the shrinkage parameter lambda (namely cross-validated and adaptive) and performing 10-fold cross-validation for a set of 100 lambdas in each model were examined.

Results

The adaptive LASSO model had a slightly better fit in the validation dataset than the cross-validated model; however, it still demonstrated poor discrimination (AUC 0.60, sensitivity 37.8%, PPV 49.6%) and poor calibration (over-triaged in low-risk and under-triaged in high-risk subgroups). The model’s poor performance was evident from an out-of-sample deviance ratio of − 0.044.

Conclusion

The ACEs checklist did not perform well in predicting ED visits among children and adolescents with existing mental health concerns. The diverse causes of ED visits may have hindered accurate predictions, requiring more advanced statistical procedures. Future studies exploring other machine learning approaches and including a more extensive set of childhood adversities and other important predictors may produce better predictions. Furthermore, despite highly significant associations being observed, ACEs may not be deterministic in predicting health-related events at the individual level, such as general ED use.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-021-01392-w.

Adherence to the Use of Home Telehealth Technologies and Emergency Room Visits in Veterans with Heart Failure

Background: Prior studies have posited poor patient adherence to remote patient monitoring as the reason for observed lack of benefits. Introduction: The purpose of this study was to examine the relationship between average adherence to the daily use of home telehealth (HT) and emergency room (ER) visits in Veterans with heart failure. Materials and Methods: This was a retrospective study using administrative data of Veterans with heart failure enrolled in Veterans Affairs (VA) HT Program in the first half of 2014. Zero-inflated negative binomial regression was used to determine which predictors affect the probability of having an ER visit and the number of ER visits. Results: The final sample size was 3,449 with most being white and male. There were fewer ER visits after HT enrollment (mean ± standard deviation of 1.85 ± 2.8) compared with the year before (2.2 ± 3.4). Patient adherence was not significantly associated with ER visits. Age and being from a racial minority group (not white or black) and belonging to a large HT program were associated with having an ER visit. Being in poorer health was associated with higher expected count of ER visits. Discussion: Subgroups of patients (e.g., with depression, sicker, or from a racial minority group) may benefit from added interventions to decrease ER use. Conclusions: This study found that adherence was not associated with ER visits. Reasons other than adherence should be considered when looking at ER use in patients with heart failure enrolled in remote patient monitoring programs.

Evaluating the effect of data standardization and validation on patient matching accuracy

OBJECTIVE

This study evaluated the degree to which recommendations for demographic data standardization improve patient matching accuracy using real-world datasets.

MATERIALS AND METHODS

We used 4 manually reviewed datasets, containing a random selection of matches and nonmatches. Matching datasets included health information exchange (HIE) records, public health registry records, Social Security Death Master File records, and newborn screening records. Standardized fields including last name, telephone number, social security number, date of birth, and address. Matching performance was evaluated using 4 metrics: sensitivity, specificity, positive predictive value, and accuracy.

RESULTS

Standardizing address was independently associated with improved matching sensitivities for both the public health and HIE datasets of approximately 0.6% and 4.5%. Overall accuracy was unchanged for both datasets due to reduced match specificity. We observed no similar impact for address standardization in the death master file dataset. Standardizing last name yielded improved matching sensitivity of 0.6% for the HIE dataset, while overall accuracy remained the same due to a decrease in match specificity. We noted no similar impact for other datasets. Standardizing other individual fields (telephone, date of birth, or social security number) showed no matching improvements. As standardizing address and last name improved matching sensitivity, we examined the combined effect of address and last name standardization, which showed that standardization improved sensitivity from 81.3% to 91.6% for the HIE dataset.

CONCLUSIONS

Data standardization can improve match rates, thus ensuring that patients and clinicians have better data on which to make decisions to enhance care quality and safety.

Risk of Frequent Emergency Department Use Among an Ambulatory Care Sensitive Condition Population: A Population-based Cohort Study

BACKGROUND

A small fraction of patients use a disproportionately large amount of emergency department (ED) resources. Identifying these patients, especially those with ambulatory care sensitive conditions (ACSC), would allow health care professionals to enhance their outpatient care.

OBJECTIVE

The objectives of the study were to determine predictive factors associated with frequent ED use in a Quebec adult population with ACSCs and to compare several models predicting the risk of becoming an ED frequent user following an ED visit.

RESEARCH DESIGN

This was an observational population-based cohort study extracted from Quebec's administrative data.

SUBJECTS

The cohort included 451,775 adult patients, living in nonremote areas, with an ED visit between January 2012 and December 2013 (index visit), and previously diagnosed with an ACSC but not dementia.

MEASURES

The outcome was frequent ED use (≥4 visits) during the year following the index visit. Predictors included sociodemographics, physical and mental comorbidities, and prior use of health services. We developed several logistic models (with different sets of predictors) on a derivation cohort (2012 cohort) and tested them on a validation cohort (2013 cohort).

RESULTS

Frequent ED users represented 5% of the cohort and accounted for 36% of all ED visits. A simple 2-variable prediction model incorporating history of hospitalization and number of previous ED use accurately predicted future frequent ED use. The full model with all sets of predictors performed only slightly better than the simple model (area under the receiver-operating characteristic curve: 0.786 vs. 0.759, respectively; similar positive predictive value and number needed to evaluate curves).

CONCLUSIONS

The ability to identify frequent ED users based only on previous ED and hospitalization use provides an opportunity to rapidly target this population for appropriate interventions.

Patient-Reported Symptoms Improve Performance of Risk Prediction Models for Emergency Department Visits Among Patients With Cancer: A Population-Wide Study in Ontario Using Administrative Data

CONTEXT

Prior work shows measurements of symptom severity using the Edmonton Symptom Assessment System (ESAS) which are associated with emergency department (ED) visits in patients with cancer; however, it is not known if symptom severity improves the ability to predict ED visits.

OBJECTIVES

To determine whether information on symptom severity improves the ability to predict ED visits among patients with cancer.

METHODS

This was a population-based study of patients who were diagnosed with cancer and had at least one ESAS assessment completed between 2007 and 2015 in Ontario, Canada. After splitting the cohort into training and test sets, two ED visit risk prediction models using logistic regression were developed on the training cohort, one without ESAS and one with ESAS. The predictive performance of each risk model was assessed on the test cohort and compared with respect to area under the curve and calibration.

RESULTS

The full cohort consisted of 212,615 unique patients with a total of 1,267,294 ESAS assessments. The risk prediction model including ESAS was superior in sensitivity, specificity, accuracy, and discrimination. The area under the curve was 73.7% under the model with ESAS, whereas it was 70.1% under the model without ESAS. The model with ESAS was also better calibrated. This improvement in calibration was particularly noticeable among patients in the higher deciles of predicted risk.

CONCLUSION

This study demonstrates the importance of incorporating symptom measurements when developing an ED visit risk calculator for patients with cancer. Improved predictive models for ED visits using measurements of symptom severity may serve as an important clinical tool to prompt timely interventions by the cancer care team before an ED visit is necessary.

Statistical tools used for analyses of frequent users of emergency department: a scoping review

OBJECTIVE

Frequent users represent a small proportion of emergency department users, but they account for a disproportionately large number of visits. Their use of emergency departments is often considered suboptimal. It would be more efficient to identify and treat those patients earlier in their health problem trajectory. It is therefore essential to describe their characteristics and to predict their emergency department use. In order to do so, adequate statistical tools are needed. The objective of this study was to determine the statistical tools used in identifying variables associated with frequent use or predicting the risk of becoming a frequent user.

METHODS

We performed a scoping review following an established 5-stage methodological framework. We searched PubMed, Scopus and CINAHL databases in February 2019 using search strategies defined with the help of an information specialist. Out of 4534 potential abstracts, we selected 114 articles based on defined criteria and presented in a content analysis.

RESULTS

We identified four classes of statistical tools. Regression models were found to be the most common practice, followed by hypothesis testing. The logistic regression was found to be the most used statistical tool, followed by χ2 test and t-test of associations between variables. Other tools were marginally used.

CONCLUSIONS

This scoping review lists common statistical tools used for analysing frequent users in emergency departments. It highlights the fact that some are well established while others are much less so. More research is needed to apply appropriate techniques to health data or to diversify statistical point of views.

Towards a Learning Health System to Reduce Emergency Department Visits at a Population Level

High utilizers of the Emergency Department (ED) often have complex needs that require coordination of care between multiple organizations. We describe a Learning Health Systems (LHS) approach to reducing ED visits, in which an intervention is delivered to a cohort of high utilizers identified using population-level data and predictive modeling. We focus on the development and validation of a random forest model that utilizes electronic health record data from three health systems across two counties in Michigan to predict the number of ED visits each resident will incur in the next six months. Using 5-fold cross-validation, the model achieves a root-mean-squared-error of 0.51 visits and a mean absolute error of 0.24 visits. Using time-based validation, the model achieves a root-mean-squared error of 0.74 visits and a mean absolute error of 0.29 visits. Patients projected to have high ED utilization are being enrolled in a community-wide care coordination intervention using twelve sites across two counties. We believe that the repeated cycles of modeling and intervention demonstrate an LHS in action.

Factors Associated With Emergency Department Use by Patients With and Without Mental Health Diagnoses

IMPORTANCE

An association between frequent use of the emergency department (ED) and mental health diagnoses is frequently documented in the literature, but little has been done to more thoroughly understand why mental illness is associated with increased ED use.

OBJECTIVE

To determine which factors were associated with higher ED use in the near future among patients with and without mental health diagnoses.

DESIGN, SETTING, AND PARTICIPANTS

A retrospective case-control study of all patients presenting to the ED in California in 2013 using past ED data to predict future ED use. Data from January 1, 2012, through December 31, 2014, from California's Office of Statewide Health Planning and Development were analyzed.

MAIN OUTCOMES AND MEASURES

Factors associated with higher ED use in the year following an index visit for patients with vs without a mental health diagnosis.

RESULTS

Among the 3 446 338 individuals in the study (accounting for 7 678 706 ED visits), 44.6% (1 537 067) were male; 31.6% (1 089 043) were between the ages of 18 and 30 years, 40.3% (1 338 874) were between the ages of 31 and 50 years, and 28.1% (968 421) were between the ages of 51 and 64 years. The mean (SD) number of ED visits per patient per year was 1.69 (2.56), and 29.1% of patients (1 002 884) had at least 1 mental health diagnosis. Previous hospitalization and high rates of lagged ED visits were associated with higher future ED use. The severity of the mental health diagnosis (mild, moderate, or severe) was associated with increased ED visits (incidence rate ratio [IRR], 1.029; 95% CI, 1.02-1.04 for mild; IRR, 1.121; 95% CI, 1.11-1.13 for moderate; and IRR, 1.226; 95% CI, 1.22-1.24 for severe). Little evidence was found for interaction effects between mental health diagnoses and other diagnoses in predicting increased future ED use.

CONCLUSIONS AND RELEVANCE

Certain classes of mental health diagnoses were associated with higher ED use. The presence of a mental illness diagnosis did not appear to interact with other patient-level factors in a way that meaningfully altered associations with future ED use.

Predicting Emergency Department Visits

High utilizers of emergency departments account for a disproportionate number of visits, often for nonemergency conditions. This study aims to identify these high users prospectively. Routinely recorded registration data from the Indiana Public Health Emergency Surveillance System was used to predict whether patients would revisit the Emergency Department within one month, three months, and six months of an index visit. Separate models were trained for each outcome period, and several predictive models were tested. Random Forest models had good performance and calibration for all outcome periods, with area under the receiver operating characteristic curve of at least 0.96. This high performance was found to be due to non-linear interactions among variables in the data. The ability to predict repeat emergency visits may provide an opportunity to establish, prioritize, and target interventions to ensure that patients have access to the care they require outside an emergency department setting.