Identifying Frequent Health Care Users and Care Consumption Patterns: Process Mining of Emergency Medical Services Data

Exploring Acute Pancreatitis Clinical Pathways Using a Novel Process Mining Method

Mining process models of medical behavior from electronic medical records is an effective way to optimize clinical pathways. However, clinical medical behavior is an extremely complex field with high nonlinearity and variability, and thus we need to adopt a more effective method. In this study, we developed a fuzzy process mining method for complex clinical pathways. Firstly, we designed a multi-level expert classification system with fuzzy values to preserve finer details. Secondly, we categorized medical events into long-term and temporary events for more specific data processing. Subsequently, we utilized electronic medical record (EMR) data of acute pancreatitis spanning 9 years, collected from a large general hospital in China, to evaluate the effectiveness of our method. The results demonstrated that our modeling process was simple and understandable, allowing for a more comprehensive representation of medical intricacies. Moreover, our method exhibited high patient coverage (>0.94) and discrimination (>0.838). These findings were corroborated by clinicians, affirming the accuracy and effectiveness of our approach.

Appropriateness of frequent use of emergency departments: A retrospective analysis in Rome, Italy

Background

Frequent users (FUs) are patients who repeatedly and inappropriately visit the emergency department (ED) for low-grade symptoms that could be treated outside the hospital setting. This study aimed to investigate the phenomenon of the FU in Rome by profiling such users and analyzing ED attendance by FUs.

Methods

The analysis was carried out for attendance in 2021 at 15 EDs in the Local Health Authority Roma 1 geographical area. A digital app collected data, including information on the following variables: number of attendance, demographic characteristics, emergency medical service (EMS) usage, triage code, and appropriateness of attendance. COVID-19 diagnosis was also studied to analyze any possible influence on ED attendance. Differences between FUs and non-FUs were investigated statistically by t-test and chi-square test. Univariate analysis and multivariable logistic regression were performed to analyze the associated factors.

Results

A total of 122,762 ED attendance and 89,036 users were registered. The FU category represented 2.9% of all users, comprising 11.9% of total ED attendance. There was a three times higher frequency of non-urgent codes in attendance of FU patients (FU: 9.7%; non-FU: 3.2%). FUs were slightly more likely to have used the EMS (13.6% vs. 11.4%) and had a lower frequency of appropriate ED attendance (23.8% vs. 27.0%). Multivariate logistic analysis confirmed a significant effect of triage code, gender, age, EMS usage, and COVID-19 diagnosis for the appropriateness of attendance. The results were statistically significant (p < 0.001).

Conclusion

The FU profile describes mostly non-urgent and inappropriate attendance at the ED, including during the COVID-19 pandemic. This study represents an important tool for strengthening preventive policies outside the hospital setting. The Italian National Recovery and Resilience Plan represents an excellent opportunity for the development of new strategies to mitigate the phenomenon of FUs.

Defining a threshold above which an adult can be considered to frequently use ambulance services: a retrospective cross-sectional study of emergency calls to an ambulance service in England

Objective

There is no empirical definition of adult frequent use of ambulance services. This study aimed to define a threshold, and utilise this to explore characteristics of people frequently using services.

Methods

This was a retrospective cross-sectional study in a single ambulance service in England. Routinely collected, pseudo-anonymised call- and patient-level data were collected for two months (January and June 2019). Incidents, defined as independent episodes of care, were analysed using a zero-truncated Poisson regression model to determine a suitable frequent-use threshold, with comparisons subsequently made between frequent and non-frequent users.

Results

A total of 101,356 incidents involving 83,994 patients were included in the analysis. Two potentially appropriate thresholds were identified: five incidents per month (A); and six incidents per month (B). Threshold A produced 3137 incidents from 205 patients, with five patients likely false-positive identifications. Threshold B produced 2217 incidents from 95 patients, with no false-positive identifications but 100 false-negatives compared to threshold A. Regardless of threshold, frequent users compared to non-frequent users had relatively reduced service use between 08:00 and 15:00, were younger and were more likely to receive lower-priority responses (all p < 0.001). We identified several chief complaints indicative of increased frequent use, including chest pain, psychiatric/suicide attempt and abdominal pains/problems.

Conclusions

We suggest a threshold of five incidents per month, with recognition that a small number of patients may be incorrectly identified as using ambulance services frequently. The rationale for this choice is discussed. This threshold may be applicable in wider UK settings and could be used for the routine automated identification of people using ambulance services frequently. The identified characteristics can help inform interventions. Future research should examine applicability of this threshold in other UK ambulance services and countries where patterns and determinants of frequent ambulance use may differ.