Abstract
Older patients are heavy users of emergency department (ED) resources and are at high risk for short-term ED visits, often leading to adverse outcomes. We aim to elucidate the characteristics of older patients who undergo 72-h ED returns, and develop a prediction model for unfavorable outcomes to facilitate clinical practices. This retrospective observational study enrolled older patients who shortly returned to the ED of a tertiary hospital within 72 h between 2019 and 2020. The study population was divided into development and validation datasets. The primary outcome was high-risk ED returns, defined as intensive care unit admission or in-hospital mortality after ED returns. Multivariable logistic regression was performed to identify predictors of high-risk returns, and a prediction model was built accordingly. A total of 1118 encounters were enrolled in our development dataset, with a mean age of 79.4 ± 9.5 years. Through multivariable analysis, independent predictors of high-risk ED returns were identified. A simple prediction model (ReC-FLASH) was developed, demonstrating a C-statistic of 0.862 (95% CI: 0.822-0.903, P < .001), incorporating “
Keywords
Introduction
Older patients are frequent users of emergency departments (EDs), and their utilization rates tend to increase with age.1 -3 In addition to medical needs, several reasons have been proposed to explain this trend, including cognitive impairment, functional decline, such as mobility reduction, and lack of social support. 4 Not only do older patients visit the ED more often than younger ones, but they also have higher unplanned return rates, ranging from 12% to 19% compared to younger adults.2,5 This heightened frequency may arise from the vagueness of their clinical presentations.3,6 ED returns not merely strain already overburdened ED settings and increase healthcare costs, but also serve as a potential indicator of suboptimal emergency care.7,8 From the perspective of care quality, ED returns may indicate inadequate initial assessment, treatment, and patient education. 9 Moreover, patients revisiting the ED may also be correlated with higher hospitalization and mortality rates.
Given the limited literature on older patients revisiting the ED and the adverse relationship between such returns and patient outcomes, this study aimed to elucidate the demographic characteristics of this patient population, identify the factors associated with unfavorable outcomes, and develop a prediction model to classify these patients according to future risk.
Methods
Study Design and Setting
This study was conducted at a tertiary care teaching facility in Taiwan. A retrospective cohort design was used, enrolling patients aged 65 and older who presented to the ED between 2019 and 2020. The study population was divided into 2 parts: the development cohort (January 1, 2019, to June 30, 2020) and the validation cohort (July 1, 2020, to December 31, 2020). The study focused on older patients who had at least 1 ED visit followed by a return within 72 h during the study period. Patients with incomplete records, those with frequent ED visits, or those who left against medical advice were excluded to ensure data quality. Frequent ED users were defined as individuals with more than 5 ED visits within 1 year prior to the index ED visit. 10 As this was a retrospective observational study utilizing all available encounters that met the inclusion criteria during the study period, no a priori sample size calculation or power analysis was performed. The sample size was therefore determined by the total number of eligible 72-h ED return encounters identified from the electronic health record. This approach is commonly applied in clinical prediction model development using real-world datasets.
This study received approval from the Ethics Committee/Institutional Review Board (Protocol Number: 2021-06-027CC), which waived the need for informed patient consent due to the retrospective nature of the analysis. Data were collected by 2 investigators using structured data collection forms. Reliability was assessed at the outset of the analysis and intermittently verified by 2 independent reviewers. Any disagreements or uncertainties were resolved through consultation with external experts (a senior researcher or an emergency physician) and finalized by joint consensus. The methodology of this study is consistent with the STROBE checklist for observational studies.
Data Collection and Outcomes
Routine patient demographics, Charlson Comorbidity Index scores, underlying medical history, chief complaints, triage levels of index visits and returns, length of ED stay of index ED and return ED, interval between 2 ED visits, marital status, polypharmacy, functional status of daily living, education levels, living arrangement, and subsequent outcomes after returns, including return with out-of-hospital cardiac arrest (OHCA), hospital admission, intensive care unit (ICU) admission, and in-hospital mortality, were retrospectively obtained from the electronic health records (EHR). Triage levels were determined by the Taiwan Triage and Acuity Scale (TTAS), which categorized patients into 5 levels based on their severity. 11 Levels with lower numbers are considered more life-threatening, thus requiring more timely intervention. Changes in triage scores were computed as the difference in score between the return and index visits, with a negative value suggesting a deterioration in health status. Mild liver disease was defined as chronic hepatitis or cirrhosis without portal hypertension, while moderate to severe liver disease was defined as cirrhosis and portal hypertension with or without a variceal bleeding history. Polypharmacy was defined as the regular use of 5 or more different medications. The primary outcome was high-risk ED return, which was defined as intensive care unit admission or in-hospital mortality after ED returns.
Statistical Analysis
Normally distributed continuous variables were described as mean and standard deviation (SD), while non-normally distributed continuous variables were expressed as median and interquartile range (IQR). Continuous variables were analyzed using the 2 independent-samples t-test, and categorical variables were examined using the chi-square test. A predictive model was constructed using the development dataset through multivariable logistic regression analysis. Variables with a significance level of P < .10 in univariate analysis were entered into the multivariable model, followed by backward stepwise elimination. Adjusted odds ratios (ORs) and their corresponding 95% confidence intervals (CIs) were reported. Each predictor’s OR from the final model was used to assign weighted points, with every 0.5 increase in OR corresponding to 0.5 points, thereby forming a clinical risk scoring system based on variables independently associated with high-risk returns.
The model’s external validity was subsequently evaluated and discriminative performance of the scoring system was assessed by constructing receiver operating characteristic (ROC) curves for both the development and validation cohorts, with the area under the curve (AUCs, or Harrell’s C statistics) presented alongside 95% CIs. Calibration of the model was examined using the Hosmer–Lemeshow goodness-of-fit test, applied to deciles of predicted high-risk return probabilities in both cohorts as well as in 2000 bootstrap resamples. Differences were considered significant with a two-tailed P < .05. All the analyses were performed using IBM SPSS Statistics software (version 20.0; IBM Corp., Armonk, NY, USA).
Results
During the study period, 1118 and 342 encounters were identified as eligible patients for the development and validation datasets who experienced 72 h of ED returns (Supplemental Figure 1), with a mean age of 79.4 ± 9.5 years, of which 57.3% were male in the development dataset; 78.0 ± 8.9 years, 52.6% male in the validation dataset (Table 1 and Supplemental Table 1). In the development dataset, the mean Charlson Comorbidity Index score was 5.7 ± 2.5, with hypertension (64.8%), cancer (29.4%), and diabetes mellitus (29.0%) being the most prevalent underlying diseases. Regarding the outcomes of the study subjects upon ED returns, 84 (7.5%) high-risk returns were identified, including 58 (5.2%) requiring ICU admission and 40 (3.6%) experiencing in-hospital mortality. Only 1 patient presented with OHCA when returning to the ED.
Baseline Characteristics of Older People with 72-H ED Returns (N = 1118).
Note. ED = emergency department; S.D. = standard deviation; HD = hemodialysis; ICU = intensive care unit; LOS = length of hospital stay; OHCA = out-of-hospital cardiac arrest.
Compared with the non-high-risk return group, the proportion of older patients in the high-risk return group who were bedridden (44.0% vs 15.9%, P < .001), had low education levels (51.2% vs 42.4%, P = .017), and resided in nursing homes (6.0% vs 2.7%, P = .028) were significantly higher, while the proportion of fully independent functional status was significantly lower (10.7% vs 35.7%, P < .001; Table 2). There was no significant difference between the high-risk return and non-high-risk return groups regarding marital status and polypharmacy in the development dataset. Chief complaints of shortness of breath (25.0% vs 4.9%, P < .001), focal weakness/suspect stroke (4.8% vs 1.6%, P = .043), and cancer-related issues (3.6% vs 0.8%, P = .012) were significantly higher in the high-risk return group (Supplemental Table 2). Altered mental status (8.3% vs 3.9%, P = .050) also showed marginal significance. Regarding triage information, high-risk return individuals were more likely to have index visit triage levels ≤ 2 (16.7% vs 9.5%, P = .035), return triage levels ≤ 2 (60.7% vs 12.0%, P < .001), and escalation of triage levels upon returns (56.0% vs 13.2%, P < .001). There was no significant difference in the interval between 2 ED visits and length of stay of index ED and return ED.
Geriatric Parameter of Older People with 72-H ED Returns (N = 1118).
Note. ED = emergency department.
P < .05.
Multivariable analysis revealed that return triage levels ≤ 2 (adjusted OR 5.175, P = .001), hypertension history (adjusted OR 1.960, P = .040), stroke history (adjusted OR 2.156, P = .018), moderate to severe liver disease (adjusted OR 5.085, P < .001), cancer (adjusted OR 2.053, P = .010), bed-ridden status (adjusted OR 2.900, P < .001) and complaint of shortness of breath (adjusted OR 2.846, P = .004) were identified as independent predictors of high-risk ED returns (Supplemental Table 3). The sum of these 7 factors constituted a new scoring model with the mnemonic ReC-FLASH, consisting of “
Point Allocation for Predictors of Older People with High-Risk 72-H ED Returns as ReC-FLASH Score.
Note. Re = Return triage; C = Cancer; F = Functional bed-ridden status; L = Liver disease, moderate to severe; A = Air hunger (shortness of breath); S = Stroke; H = Hypertension.
The area under the receiver operating characteristic curve (AUROC) of the ReC-FLASH model.
Discussion
In this retrospective study, we investigated the characteristics of older patients with 72-h ED returns and highlighted the significant predictors of adverse outcomes for these patients. Key findings indicated a higher likelihood of high-risk returns in patients with several conditions, such as hypertension, cancer, moderate to severe liver disease, stroke, fully dependent functional status, shortness of breath, and return triage levels. A simple and practical prediction model, the ReC-FLASH score, was developed to assist clinicians in the early identification and management of older patients at increased risk of deterioration upon ED return. Among these predictors, “Air Hunger” was used in the mnemonic to maintain the acronym structure. This component specifically represents shortness of breath, a well-established clinical predictor of adverse outcomes among older ED patients. The ReC-FLASH score demonstrated a strong ability to identify high-risk ED returns and to stratify older patients into low-, moderate-, and high-risk groups across 3 adverse outcomes: hospital admission, ICU admission, and in-hospital mortality.
To our best knowledge, this is the first study to propose such a prediction model specifically for older patients with short-term ED returns. It offers clinicians a quick tool to foresee the likelihood of unfavorable outcomes, allowing them to provide anticipatory guidance and take precautions against adverse events. McCusker et al 12 proposed the Identification of Seniors At Risk (ISAR) screening tool to identify older patients at risk of adverse outcomes in 1999. Apart from disparate definitions of adverse outcomes, the ISAR was essentially a self-report screening tool, subject to potential reference bias and lack of objectivity. 13 It was designed for older people during their index ED visits or hospital inpatient discharge rather than ED returns. However, older patients are vulnerable and tend to utilize ED care more frequently.1-3,5 In particular, those in the last months of life have even higher ED visit rates. 14 A recent retrospective cohort study from Singapore 15 similarly examined older adults who returned to the ED and reported that high-acuity presentations, dyspnea, and multiple comorbidities were prominent characteristics among patients experiencing adverse outcomes. Their findings align closely with several components of the ReC-FLASH score—particularly the predictive value of shortness of breath, functional dependence, and complex chronic disease burden. However, their study described revisit characteristics without developing a risk-prediction tool, whereas the current work advances the field by constructing a clinically actionable scoring system to stratify risk at the time of return presentation. Our study was conducted at a single center with a modest number of adverse events, which constrained the development of an index-visit–only prediction model. Although we performed exploratory analyses using only index-visit variables, model performance did not reach clinically actionable thresholds. As such, the present score is intended specifically for use at the return visit, and further multicenter prospective research is necessary to create a reliable index-visit tool.
Our study is one of the few that identify the risk factors of unfavorable outcomes specifically in older ED patients with short-term returns. The analysis incorporated symptom-oriented factors (air hunger), underlying comorbidities (cancer, moderate to severe liver disease, stroke, and hypertension), and objective characteristics (return triage and bed-ridden status). Although high-acuity triage (TTAS ≤ 2) appropriately identifies patients requiring immediate attention, our findings indicate that triage alone does not capture all older adults at risk of deterioration after a 72-h ED return. In our cohort, 39.3% of patients who subsequently required ICU admission or died during hospitalization returned with TTAS levels ≥ 3, suggesting that a substantial proportion of high-risk patients would not have been flagged as urgent based on triage alone. This subgroup represents the key clinical niche for the ReC-FLASH score. Importantly, the ReC-FLASH score is not intended to replace triage, but rather to complement conventional triage by identifying additional high-risk older patients among those classified as lower acuity (TTAS ≥ 3). By integrating geriatric-specific vulnerabilities and comorbidities, the score provides incremental risk stratification in older patients who may appear clinically stable at presentation when revisiting yet remain at elevated risk for adverse outcomes. The ReC-FLASH scoring system, derived from concise but critical clinical factors, enables clinicians to identify those with high-risk returns in advance. Since older patients are frequent ED users, our findings provide clinicians with valuable insights and guidance for both early detection and prevention in this population.
Both old age and frequent ED visits are reported to be associated with poor survival.16 -18 Several reasons account for such a situation, including atypical clinical presentations, physiological changes of aging, adverse effects of polypharmacy, and multiple comorbidities.14,19 -21 However, the relationship between ED returns and poor outcomes remains inconsistent in the current literature.22 -24 In our study, a TTAS triage category of less than 2 upon revisiting indicated higher patient acuity and was strongly associated with ICU admission and in-hospital mortality. This aligns with the study conducted by Cheng et al, 23 which showed higher morbidity and mortality rates in return patients. By contrast, Sabbatini et al 24 found lower in-hospital mortality and ICU admission rates among adult inpatients admitted through ED returns compared with those admitted through initial ED visits.
TTAS is a 5-level system adapted from the Canadian Triage and Acuity Scale (CTAS), categorizing patients based on chief complaints and vital signs. 25 The comparison systems exhibited substantial structural variation. The CTAS is a standardized 5-level system using computer support software. 26 The Emergency Severity Index (ESI) in the United States is likewise a 5-level system focusing on physical signs and resource utilization, implemented with a 2-page checklist. 27 The Manchester Triage Scale (MTS) employs 5 levels with 52 flowcharts. 28 TTAS demonstrated comparable performance to the CTAS when both were implemented in Taiwanese EDs. 26 The subsequent validation of 5-level TTAS showed a weighted kappa of .87 for inter-rater reliability, demonstrating high reproducibility. 11 Against the ESI, the TTS showed meaningful distributional differences. 27 For routine ED operations across mixed patient populations, 5-level systems (TTAS, CTAS, ESI, and MTS) demonstrate superior discrimination compared to simpler schemes, 26 with computerized decision support enhancing reliability. 11 Nonetheless, because triage criteria and real-world application may vary across regions, external validation using CTAS, ESI, or MTS-based datasets remains necessary to confirm transportability.
Though ED returns have been considered surrogates for suboptimal quality of care and adopted for ED performance measurement, several studies found otherwise.9,24,29 -32 For instance, no mortality difference was observed even in studies showing higher ICU admission rates among those making returns. 24 This could be explained by the natural disease history, namely expected disease progression, rather than diagnostic or treatment errors. 32 In addition, a root-cause analysis revealed that most ED returns among older patients were unpreventable, with only a few attributed to medical mistakes, considering that old age is a crucial predictor of mortality in patients with ED returns. 32 Furthermore, some returns are scheduled rather than unexpected. It is also not uncommon for patients to return merely to seek reassurance. Given the complex needs and atypical presentations of older patients, a simple and practical risk assessment tool is crucial for clarifying which older people will experience a poor prognosis when revisiting.
In our study, history of cancer, stroke, moderate to severe liver disease and hypertension were main comorbidities correlated to adverse outcomes after ED returns. Cancer patients are high-frequency ED users, and malignancy has been known for its association with morbidity and mortality in patients returning to the ED.33 -35 Hematologic malignancy, lung cancer, liver cancer, and metastatic cancer are the most common contributing cancers.34,35 This tendency exists across all age groups, and the majority of their returns are related to infection. 35 On the other hand, stroke may also lead to poor outcomes due to the disease itself or complications following acute debility.36,37 Besides, dyspnea has been reported as a predictor of mortality and ICU admission for ED return patients, 30 which is coherent with our findings as well.
This study underscored the need to prioritize high-risk older patients within ED return care pathways. Implementing policies that integrate risk stratification tools, such as the ReC-FLASH score, may optimize resource allocation and reduce preventable adverse outcomes in vulnerable older adults. Tailored interventions, including comprehensive geriatric assessments and follow-up plans post-discharge, could be mandated to enhance care quality and reduce the burden of repeated ED visits.
In clinical practice, this ReC-FLASH score presents a practical, evidence-based tool that empowers frontline clinicians to swiftly identify older patients at heightened risk for ICU admission or in-hospital mortality upon ED returns. Integrating this tool into (EHR can streamline real-time risk assessment, enabling prompt and personalized management strategies. High-risk patients identified through the score may benefit from closer monitoring, early specialist referrals, and proactive measures to address modifiable factors contributing to adverse outcomes, such as functional decline and untreated comorbidities. It is important to note that the ReC-FLASH score is not designed to guide discharge safety at the index visit. Instead, it targets a different clinical challenge: identifying older adults who are already returning within 72 h and are at high risk of ICU admission or in-hospital mortality. For this reason, return-visit characteristics, such as return triage level—one of the strongest predictors of adverse outcomes—were appropriately included. This aligns the model with its intended clinical context: risk stratification upon return, enabling rapid escalation of care for high-risk older adults.
Additionally, our study highlighted the need for external validation of the ReC-FLASH score across diverse healthcare settings and populations. Future investigations should explore the integration of dynamic clinical data and longitudinal functional status changes to refine predictive accuracy further. Moreover, qualitative studies examining barriers to implementing risk stratification tools in busy ED environments can inform implementation strategies. There is also scope to evaluate the impact of risk-based interventions on reducing returns and improving patient-centered outcomes in older adults. Collectively, these implications emphasize a comprehensive approach, integrating robust prediction tools with proactive clinical and policy measures, to improve the quality of ED care for older adults and reduce avoidable harm during 72-h ED returns.
There are several limitations in this study. First, it was conducted at a single medical center, primarily serving veterans. While this offers a platform for geriatric research because of the larger proportion of older patients, the health conditions of this study population might differ from those in other medical facilities. In view of the unique characteristics of a veteran population, which may influence both the prevalence of comorbidities and patterns of emergency care use, findings should be interpreted cautiously when applying the score to general ED settings. Second, although TTAS is a validated triage system, it has not been implemented outside Taiwan. Hence, the criteria for each triage category may be inconsistent with other systems. Still, most systems divide ED patients into 5 levels, with the first 2 levels requiring immediate attention, which is compatible with our study’s consideration of triage levels upon returns less than 2 to be a risk factor. Fourth, because this study retrospectively analyzed all eligible cases within a fixed 2-year period, no formal sample size or power calculation was conducted. Although the resulting sample size was adequate for model development and internal validation, the absence of an a priori power analysis may limit the assessment of whether the sample was optimal for predicting rare outcomes. Future studies with prospective designs should incorporate sample size estimation to further validate and refine the ReC-FLASH score. Lastly, despite the high C-statistic of the ReC-FLASH score, the accuracy of the prediction rule is not validated in other populations, therefore warranting further investigation to generalize the result.
By applying the ReC-FLASH score in clinical practice, healthcare providers can prioritize high-risk individuals and optimize tailored care plans that address the specific needs of older patients with significant comorbidities and functional dependencies in the ED setting, eventually rescuing more lives. Continuous evaluation and refinement of the prediction model are essential to maintain its accuracy and effectiveness. Incorporating new data and adjusting for changes in patient demographics or clinical practices will help sustain its predictive power.
Conclusion
In conclusion, by identifying the predictors of high-risk returns, ICU admission or in-hospital mortality among older patients revisiting the ED shortly, the ReC-FLASH score, derived from integrating those predictors, exhibits ideal performance in categorizing these patients into different risk groups. Understanding which patients are at high risk can help allocate healthcare resources more effectively and efficiently, ensuring that high-risk older patients receive the necessary attention and care to mitigate their risk. Targeted interventions, such as thorough geriatric assessments, closer monitoring, and proactive management of comorbid conditions, can reduce adverse outcomes after ED returns. Furthermore, the ReC-FLASH score can be integrated into the EHR system to provide real-time risk assessment for older adults presenting to the ED, particularly during return visits. This can help clinicians promptly identify high-risk older patients and tailor their management accordingly.
Supplemental Material
sj-doc-1-inq-10.1177_00469580261433441 – Supplemental material for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score
Supplemental material, sj-doc-1-inq-10.1177_00469580261433441 for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score by Chung-Ting Chen, Yu-Hsiang Meng, Hsin-Hua Yu, Chorng-Kuang How and Yu-Chi Tung in INQUIRY: The Journal of Health Care Organization, Provision, and Financing
Supplemental Material
sj-docx-2-inq-10.1177_00469580261433441 – Supplemental material for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score
Supplemental material, sj-docx-2-inq-10.1177_00469580261433441 for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score by Chung-Ting Chen, Yu-Hsiang Meng, Hsin-Hua Yu, Chorng-Kuang How and Yu-Chi Tung in INQUIRY: The Journal of Health Care Organization, Provision, and Financing
Supplemental Material
sj-pdf-3-inq-10.1177_00469580261433441 – Supplemental material for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score
Supplemental material, sj-pdf-3-inq-10.1177_00469580261433441 for Predicting Adverse Outcomes in Older Adults with 72-Hour Emergency Department Returns: A Retrospective Cohort Study Developing the Rec-FLASH Score by Chung-Ting Chen, Yu-Hsiang Meng, Hsin-Hua Yu, Chorng-Kuang How and Yu-Chi Tung in INQUIRY: The Journal of Health Care Organization, Provision, and Financing
Footnotes
Ethics Considerations
This study was approved by the Ethics Committee/Institutional Review Board of the Taipei Veterans General Hospital (Protocol Number: 2021-06-027CC), which waived the requirement for informed patient consent because of the retrospective nature of the analysis.
Consent to Participate
Not applicable.
Consent for Publication
Not applicable.
Author Contributions
Study concept and design: CTC, YCT; acquisition of the data: CTC, YHM, HHY; analysis and interpretation of the data: CTC, YHM, YCT; drafting of the manuscript: CTC, YHM, HHY, CKH, YCT; critical revision of the manuscript: CTC, CKH, YCT; statistical expertise: CTC, YHM, YCT.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by the National Science and Technology Council (NSTC, previously called as Ministry of Science and Technology [MOST]) (grant number NSTC 113–2410-H-002–052-MY3 and MOST 110–2410-H-002– 116-MY3) and the Population Health and Welfare Research Center from the Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (grant number NTU-115L900401) in Taiwan.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The datasets generated and analyzed during the current study are not publicly available but are available from the corresponding author upon reasonable request.
Supplemental Material
Supplemental material for this article is available online.