Revisits, Readmission, and Mortality From Emergency Department Admissions for Older Adults With Vague Presentations: Longitudinal Observational Study

Introduction

Care for acute illnesses has shifted from outpatient offices to emergency departments (EDs), leading to an increase in ED use that has outpaced population growth [Pitts SR, Carrier ER, Rich EC, Kellermann AL. Where Americans get acute care: increasingly, it’s not at their doctor’s office. Health Aff (Millwood). Sep 2010;29(9):1620-1629. [CrossRef]1]. EDs diagnose and treat acute illnesses [Kellermann AL, Martinez R. The ER, 50 years on. N Engl J Med. Jun 16, 2011;364(24):2278-2279. [CrossRef]2]. Therefore, emergency providers diagnose patients, initiate treatment, and predict the disease trajectory to decide whether to admit the patient to an inpatient unit or discharge the patient home [Tang N, Stein J, Hsia RY, Maselli JH, Gonzales R. Trends and characteristics of US emergency department visits, 1997-2007. JAMA. Aug 11, 2010;304(6):664-670. [CrossRef] [Medline]3,Morganti KG, Bauhoff S, Blanchard JC, et al. The Evolving Role of Emergency Departments in the United States. Vol 3. Rand health quarterly The RAND Corporation; 2013. 4]. Any change to admission decisions can impact outcomes and costs since two-thirds of ED health care costs in the United States come from visits that end in admission [Sabbatini AK, Nallamothu BK, Kocher KE. Reducing variation in hospital admissions from the emergency department for low-mortality conditions may produce savings. Health Aff (Millwood). Sep 2014;33(9):1655-1663. [CrossRef]5,Galarraga JE, Pines JM. Costs of ED episodes of care in the United States. Am J Emerg Med. Mar 2016;34(3):357-365. [CrossRef]6].

Whether to admit or discharge a patient is a weighty decision. This point is best exemplified with older adults (≥65 years of age). Discharging an older adult carries a high risk of adverse health outcomes, especially when compared with younger patients [Aminzadeh F, Dalziel WB. Older adults in the emergency department: a systematic review of patterns of use, adverse outcomes, and effectiveness of interventions. Ann Emerg Med. Mar 2002;39(3):238-247. [CrossRef]7-Rowland K, Maitra AK, Richardson DA, Hudson K, Woodhouse KW. The discharge of elderly patients from an accident and emergency department: functional changes and risk of readmission. Age Ageing. 1990;19(6):415-418. [CrossRef]12]. Hence, increasing interest has been placed on identifying patients at risk for adverse outcomes after ED discharge [Aminzadeh F, Dalziel WB. Older adults in the emergency department: a systematic review of patterns of use, adverse outcomes, and effectiveness of interventions. Ann Emerg Med. Mar 2002;39(3):238-247. [CrossRef]7,Caplan GA, Brown A, Croker WD, Doolan J. Risk of admission within 4 weeks of discharge of elderly patients from the emergency department—the DEED study. Age Ageing. 1998;27(6):697-702. [CrossRef]11,Martin-Gill C, Reiser RC. Risk factors for 72-hour admission to the ED. Am J Emerg Med. Oct 2004;22(6):448-453. [CrossRef]13,Gerson LW, Rousseau EW, Hogan TM, Bernstein E, Kalbfleisch N. Multicenter study of case finding in elderly emergency department patients. Acad Emerg Med. Aug 1995;2(8):729-734. [CrossRef]14] or on developing strategies for following up on discharged patients [Stuck AR, Crowley C, Killeen J, Castillo EM. National survey of emergency physicians concerning home-based care options as alternatives to emergency department–based hospital admissions. J Emerg Med. Nov 2017;53(5):623-628. [CrossRef]15]. Admitting a patient who can be discharged carries its own risks. Older adults are vulnerable to deconditioning and hospital acquired infection, as well as developing delirium and accelerated cognitive decline [Kozak LJ, Hall MJ, Owings MF. Trends in avoidable hospitalizations, 1980–1998. Health Aff (Millwood). Mar 2001;20(2):225-232. [CrossRef]16-Taylor ME, Oppenheim BA. Hospital-acquired infection in elderly patients. J Hosp Infect. Apr 1998;38(4):245-260. [CrossRef]19]. These issues underscore the importance of measuring the trade-off between admitting and discharging a patient.

Admission decisions are usually based on well-defined clinical factors and practice guidelines, but often, patients fall into a gray area in which the need for hospital admission is unclear based on objective information or even local standards of care. A large group of such gray-area patients is those presenting with syndromic diagnoses such as falls [Hartholt KA, van der Velde N, Looman CWN, et al. Trends in fall-related hospital admissions in older persons in the Netherlands. Arch Intern Med. May 24, 2010;170(10):905-911. [CrossRef] [Medline]20] or weakness [Eagles D, Stiell IG, Clement CM, et al. International survey of emergency physicians’ priorities for clinical decision rules. Acad Emerg Med. Feb 2008;15(2):177-182. [CrossRef]21], or patients presenting with more definite diagnoses that are associated with wide variability in admission decision, such as syncope [Daccarett M, Jetter TL, Wasmund SL, Brignole M, Hamdan MH. Syncope in the emergency department: comparison of standardized admission criteria with clinical practice. Europace. Nov 1, 2011;13(11):1632-1638. [CrossRef]22], chest pain [Sabbatini AK, Nallamothu BK, Kocher KE. Reducing variation in hospital admissions from the emergency department for low-mortality conditions may produce savings. Health Aff (Millwood). Sep 2014;33(9):1655-1663. [CrossRef]5,Kocher KE, Meurer WJ, Fazel R, Scott PA, Krumholz HM, Nallamothu BK. National trends in use of computed tomography in the emergency department. Ann Emerg Med. Nov 2011;58(5):452-462. [CrossRef]23,Kocher KE. Achieving the holy grail of emergency department evaluation for chest pain. Circ Cardiovasc Qual Outcomes. Oct 2017;10(10). [CrossRef]24], urinary tract infection (UTI) [Sammon JD, Sharma P, Rahbar H, et al. Predictors of admission in patients presenting to the emergency department with urinary tract infection. World J Urol. Jun 2014;32(3):813-819. [CrossRef]25], pneumonia [Venkatesh AK, Chou SC, Li SX, et al. Association between insurance status and access to hospital care in emergency department disposition. JAMA Intern Med. May 1, 2019;179(5):686-693. [CrossRef] [Medline]26], and cellulitis [Volz KA, Canham L, Kaplan E, Sanchez LD, Shapiro NI, Grossman SA. Identifying patients with cellulitis who are likely to require inpatient admission after a stay in an ED observation unit. Am J Emerg Med. Feb 2013;31(2):360-364. [CrossRef]27]. Moreover, other factors influence disposition decisions including triage [Calder LA, Forster AJ, Stiell IG, et al. Mapping out the emergency department disposition decision for high-acuity patients. Ann Emerg Med. Nov 2012;60(5):567-576. [CrossRef]28], crowding [Gorski JK, Batt RJ, Otles E, Shah MN, Hamedani AG, Patterson BW. The impact of emergency department census on the decision to admit. Acad Emerg Med. Jan 2017;24(1):13-21. [CrossRef] [Medline]29], patient home environment [Nugus P, Carroll K, Hewett DG, Short A, Forero R, Braithwaite J. Integrated care in the emergency department: a complex adaptive systems perspective. Soc Sci Med. Dec 2010;71(11):1997-2004. [CrossRef]30], diagnostic testing [Florin TA, French B, Zorc JJ, Alpern ER, Shah SS. Variation in emergency department diagnostic testing and disposition outcomes in pneumonia. Pediatrics. Aug 2013;132(2):237-244. [CrossRef] [Medline]31], patient ethnicity [Studnicki J, Platonova EA, Fisher JW. Hospital-level variation in the percentage of admissions originating in the emergency department. Am J Emerg Med. Oct 2012;30(8):1441-1446. [CrossRef]32], and hospital capacity [Studnicki J, Platonova EA, Fisher JW. Hospital-level variation in the percentage of admissions originating in the emergency department. Am J Emerg Med. Oct 2012;30(8):1441-1446. [CrossRef]32]. Consequently, admission rates vary widely for these patients between providers and between hospitals, leading to potentially harmful practice variation [Pines JM, Mutter RL, Zocchi MS. Variation in emergency department admission rates across the United States. Med Care Res Rev. Apr 2013;70(2):218-231. [CrossRef]33]. Reducing this variation may avoid harmful admissions while safeguarding patient safety, but in order to do so evidence must be used to guide decision-making.

In this study, we focused on older adults in the ED with gray-area diagnoses as follows: diagnoses associated with clinical ambiguity or high rates of potentially preventable hospitalizations and variability in admissions [Kozak LJ, Hall MJ, Owings MF. Trends in avoidable hospitalizations, 1980–1998. Health Aff (Millwood). Mar 2001;20(2):225-232. [CrossRef]16,Nyweide DJ, Anthony DL, Bynum JPW, et al. Continuity of care and the risk of preventable hospitalization in older adults. JAMA Intern Med. Nov 11, 2013;173(20):1879-1885. [CrossRef] [Medline]17]. The goal of this work was to identify how the decision to admit drives subsequent revisit, readmission, and mortality among older adult patients with diagnoses that are syndromic (falls and weakness) or lacking a clear standard of practice (syncope, UTI, pneumonia, or cellulitis). This question was difficult to answer since patients with different disposition decisions differ in their clinical severity, complexity, and needs. Without adjusting for these differences, unfair conclusions can be drawn if patients with different disposition decisions are directly compared. We thus used a causal inference methodology for observational data to measure the relationship between disposition decisions and outcomes for these older adult patients [Cochran AL, Rathouz PJ, Kocher KE, Zayas‐Cabán G. A latent variable approach to potential outcomes for emergency department admission decisions. Stat Med. Sep 10, 2019;38(20):3911-3935. [CrossRef]34,Alvarez Avendaño SA, Cochran AL, Patterson B, Shah M, Smith M, Zayas-Cabán G. The average effect of emergency department admission on readmission and mortality for older adults with chest pain. Med Care. 2020;58(10):881-888. [CrossRef]35].

Methods

Data

ED visits were analyzed using electronic health records (EHRs) from a large Midwestern academic health system. The dataset used was from a common EHR system, and the analyzed population consisted of encounters with a specific large ED between January 1, 2014 and September 27, 2018. ED visits were included for older adult patients (65 years of age or older) with presentations for falls (n=1581), weakness (n=564), syncope (n=468), UTI (n=456), pneumonia (n=299), and cellulitis (n=223). For all 6 presentations, inclusion and exclusion criteria were specified to capture patients who, based on objective criteria present in the EHR occupied a “gray area” with regards to criteria for admission. In general, the strategy was to include patients who met diagnostic criteria for each presentation but did not have further abnormalities indicating a clear indication for admission present within discrete fields of the EHR. General inclusion criteria were an acuity level (emergency severity index; ESI) of 2 or 3, excluding visits assigned the most and least severe ESI levels; and treated and either admitted to the hospital or discharged to their residence, excluding patients who eloped, left against medical advice, or were transferred to another facility. General exclusion criteria were a missing disposition decision, a visit in the last 45 days of the sample period (as these patients had inadequate follow-up for outcomes), indicators of acute coronary syndromes (troponin levels >0.10 ng/mL), severe vital sign abnormalities at any point during the ED visit (systolic blood pressure <80 mmHg, respiration rate >30 BPM, pulse oximetry <88%, or heart rate >120 BPM), or a specific diagnosis that unambiguously required admission (ie, stroke, myocardial infarction, or femur fractures). In addition, patients presenting to the ED with a combination of cellulitis and a higher temperature than 100.3 F were eliminated from the analysis since these encounters had a high admission probability.

Variables

Seven variables represented baseline characteristics as follows: age, sex, insurance, history of diabetes, history of congestive heart failure, history of hypertension, and Centers for Medicare and Medicaid Services Hierarchical Condition Category (HCC) score [Pope GC, Ellis RP, Ash AS, et al. Diagnostic Cost Group Hierarchical Condition Category Models for Medicare Risk Adjustment. Health Economics Research; 2000. 36]. HCC scores predict health care costs and outcomes by assigning risk weights to comorbidities based on ICD codes; higher scores indicate a greater predicted risk and resource use. These baseline characteristics were selected because they are available prior to the ED visit and can capture key clinical variations that could influence a patient’s ED visit and outcomes.

Five variables represented initial observations serving as a proxy for underlying latent health state at the start of the admission process: acuity, temperature, blood pressure, respiratory rate, and heart rate. Acuity was measured using the ESI level, a 5-level triage system where lower ESI levels (eg, 1 or 2) indicate higher acuity, meaning the patient requires more urgent care and significant resources. These proxy variables were included because they are available before the admission process begins and can be considered indirect (or “noisy”) measurements of the patient’s latent health state.

Two variables represented the admission process: treatment time (duration between when a patient is placed in an ED room after triage for treatment and when their admission decision is made) and admission decision. Patients discharged to a Skilled Nursing Facility or Inpatient Rehabilitation Facility were treated as a discharge.

Three primary outcome variables were considered: ED revisits, hospital readmission, and mortality within 30 days of being discharged from either the hospital for admitted patients or the ED for discharged patients. Binary variables for revisits indicated whether an individual returned to the study ED (based on EHR data) within 30 days of discharge. Similarly, binary variables for readmission and mortality indicated whether an individual was readmitted or died within 30 days of discharge, respectively. Alternative windows (3 d and 9 d instead of 30 d) were also considered as secondary outcomes, since a short window is likely to capture only subsequent events related to the original ED visit, whereas a long window is likely to capture all subsequent events related to the original ED visit. To optimize this trade-off, a 9-day window has previously been recommended [Rising KL, Victor TW, Hollander JE, Carr BG. Patient returns to the emergency department: the time-to-return curve. Acad Emerg Med. Aug 2014;21(8):864-871. [CrossRef]37].

Missing continuous and categorical variables were imputed with the median and most frequent category, respectively. Continuous variables were standardized and these transformed variables were then used for estimation.

Analysis

Admission decisions were evaluated using 3 types of analyses: unadjusted, adjusted, and subgroup. Each type of analysis was repeated for each of the 3 primary outcomes, and for the overall sample and each presentation group. Unadjusted analyses involved estimating the observed (ie, unadjusted) difference in risk of an outcome between admitted patients and discharged patients. Wald 95% CIs were recovered.

Adjusted analyses, constituting our primary analyses, involved estimating these same risk differences (RDs) but adjusted for the unmeasured or latent health state in addition to measured variables (ie, baseline characteristics and proxy variables). These estimates were recovered by implementing a latent-variable approach for evaluating admission decisions (see Cochran et al [Cochran AL, Rathouz PJ, Kocher KE, Zayas‐Cabán G. A latent variable approach to potential outcomes for emergency department admission decisions. Stat Med. Sep 10, 2019;38(20):3911-3935. [CrossRef]34] for details and applied in Alvarez Avendaño et al [Alvarez Avendaño SA, Cochran AL, Patterson B, Shah M, Smith M, Zayas-Cabán G. The average effect of emergency department admission on readmission and mortality for older adults with chest pain. Med Care. 2020;58(10):881-888. [CrossRef]35] to chest pain patients). Briefly, this approach involves modeling both the measured variables and a latent “health state” variable, and then fitting this model to data using expectation maximization, from which an estimate of the average treatment effect (ie, average difference in potential outcomes were we to admit vs discharge a patient) and Wald CIs can be recovered. The latent health state is included to account for potential unmeasured confounding between the admission decision and each outcome. More specifically, it accounts for confounding by indication, whereby admission decisions are based on factors not captured in the data. The inclusion of proxy variables, such as acuity and vitals, is crucial as they provide indirect measures of the latent health state, thereby strengthening the model’s ability to mitigate confounding by indication.

The model comprised several regression components: logistic regression to model the latent health state as a function of baseline characteristics; logistic and linear regression to model the proxy variables (eg, acuity and vitals) as a function of latent health and baseline factors; threshold regression to model the admission process (decision and timing) as a function of latent health, baseline characteristics, and proxy variables; and linear regression to model each outcome based on latent health. Once model parameters are estimated, we calculate average treatment effects by comparing outcomes for each visit under hypothetical admission and discharge scenarios while holding all other variables at their original value. We then calculate the difference between these hypothetical outcomes and average this difference across all visits to obtain the overall treatment effect. For subgroup analyses, we recover subgroup-specific average treatment effects by averaging these hypothetical outcomes across only those visits with certain baseline characteristics (eg, female patients).

To check the sensitivity of our conclusions to various factors, several additional analyses were conducted. First, we checked sensitivity to 2 key assumptions for the latent-variable approach, which is that potential outcomes are independent of admission decisions with similar latent health needs and baseline characteristics and that events related to different ED visits are independent. Second, we analyzed the secondary outcomes, which use alternative time windows (3 days and 9 days instead of 30 days). Third, we estimated the same RDs as the main analyses, but only adjusted for the measured variables as opposed to both the measured variables and the latent health state. These estimates were recovered using the causal inference methods known as inverse probability weighting (IPW) and g-estimation.

Due to space considerations, sensitivity analyses are detailed in

Multimedia Appendix 1

Accompanying appendix reporting (1) IPW and g-estimation results, (2) description of the latent variable approach, (3) 3-day outcomes results, (4) sensitivity analyses, (5) subgroup analyses for individual diagnoses, (6) analysis for cellulitis diagnosis, and (7) analysis for syndromic and no clear standard practice diagnoses.

Significance was considered at an α level of 0.05. Hypothesis tests were 2-tailed Wald tests. Multiple comparisons were not adjusted for, and as such, nominal CIs and P values are reported.

Ethical Considerations

This study was reviewed by the UW Minimal Risk Research Institutional Review Board (ID 2024-0106-CP001) and was deemed to meet the federal criteria for exemption.

Results

Baseline Characteristics

Sample characteristics are summarized in Table 1. Patients were predominantly female (2102/3591, 58.5%) with Medicare insurance (1460/1589, 91.9%) and had an average age of 79 years. Patients were diagnosed with falls (1581/3591, 44%), weakness (564/3591, 16%), syncope (468/3591, 13%), UTI (456/3591, 13%), pneumonia (299/3591, 8%), and cellulitis (223/3591, 6%).

	All (n=3591)	Falls (n=1581)	Weakness (n=564)	Syncope (n=468)	UTIa (n=456)	Pneumonia (n=299)	Cellulitis (n=223)
Age (years), mean (SD)	79.23 (8.93)	80.56 (8.94)	78.66 (8.4)	76.8 (8.73)	79.22 (8.91)	78.9 (9.27)	76.758 (8.36)
Comorbidity (HCCb), mean (SD)	1.65 (1.41)	1.49 (1.22)	1.82 (1.46)	1.32 (1.36)	1.879 (1.6)	2.107 (1.57)	1.965 (1.72)
Heart rate (BPM), mean (SD)	78.07 (14.66)	76.41 (13.54)	78.39 (15.06)	73.11 (13.52)	80.79 (14.82)	87.32 (15.74)	81.41 (14.59)
Temperature (°F), mean (SD)	97.61 (1.4)	97.44 (1.69)	97.56 (0.85)	97.34 (0.72)	97.88 (1.18)	98.55 (1.56)	97.73 (0.8)
Blood pressure (mmHg), mean (SD)	74.2 (13.7)	76.23 (13.92)	74.36 (12.9)	71.91 (12.19)	73.49 (14.72)	69.47 (13.57)	71.99 (12.41)
Respiration rate (BPM), mean (SD)	18.07 (3.18)	17.82 (2.92)	17.86 (3.23)	17.72 (3.57)	18.28 (3.07)	19.8 (3.72)	18.41 (2.58)
Treatment time (hours), mean (SD)	1.82 (0.12)	0.18 (0.1)	0.2 (0.12)	0.22 (0.2)	0.16 (0.07)	0.13 (0.06)	0.15 (0.08)
Female, n (%)	2102 (58.5)	994 (62.9)	299 (53)	255 (54.5)	290 (63.6)	149 (49.8)	115 (51.6)
Insurance, n (%)
Medicaid/Badger Care	15 (0.4)	3 (0.2)	3 (0.5)	5 (1.1)	2 (0.4)	1 (0.3)	1 (0.4)
Medicare	3278 (91.9)	1460 (92.9)	526 (93.6)	415 (90.8)	409 (90.1)	273 (91.3)	195 (87.4)
Commercial/Worker’s Compensation	271 (7.6)	109 (6.9)	33 (5.9)	35 (7.7)	42 (9.3)	25 (8.4)	27 (12.1)
Self-pay	3 (0.1)	0 (0)	0 (0)	2 (0.4)	1 (0.2)	0 (0)	0 (0)
Diabetes, n (%)	744 (20.7)	293 (18.5)	138 (24.5)	81 (17.3)	99 (21.7)	72 (24.1)	61 (27.4)
Congestive Heart failure, n (%)	413 (11.5)	173 (10.9)	62 (11)	33 (7.1)	56 (12.3)	53 (17.1)	36 (16.1)
Hypertension, n (%)	2081 (58.0)	923 (58.4)	350 (62.1)	244 (52.1)	255 (55.9)	170 (56.9)	139 (62.3)
Acuity (=2), n (%)	993 (27.7)	425 (26.9)	139 (24.6)	195 (41.7)	103 (22.6)	106 (35.5)	25 (11.2)
Admitted, n (%)	1401 (39.0)	333 (21.1)	252 (44.7)	179 (38.2)	240 (52.6)	255 (85.3)	142 (63.7)
30-day revisits, n (%)	644 (17.9)	293 (18.5)	108 (19.1)	53 (11.3)	105 (23)	40 (13.4)	45 (20.2)
30-day readmission, n (%)	207 (5.8)	50 (3.2)	36 (6.4)	23 (4.9)	43 (9.4)	33 (11)	22 (9.9)
30-day mortality, n (%)	127 (3.5)	50 (3.2)	30 (5.3)	6 (1.3)	11 (2.4)	27 (9)	3 (1.3)

^aUTI: urinary tract infection.

^bHCC: hierarchical condition category.

Revisits

Table 2 summarizes the unadjusted RD between admission and discharge for 30-day revisits across all presentations and for individual presentations. Across all presentations, admission carried a significantly lower unadjusted risk than a discharge of 30-day revisits (RD=−5.2%, 95% CI −6.3 to −4.1). Individual presentations yielded unadjusted estimates that generally agreed with that of the entire sample, with 1 exception. For syncope patients, admission carried a greater unadjusted risk than a discharge of 30-day revisits (RD=2.5%, 95% CI 0.0 to 5.0).

	Unadjusted RD (95% CI)	Adjusted RD (95% CI)
All	−5.2 (−6.3 to −4.1)	−6.4 (−7.8 to −5.1)
Falls	−4.5 (−6.5 to −2.5)	−33.2 (−35.5 to −30.8)
Weakness	−8.8 (−11.6 to −6.0)	4.4 (−2.8 to 11.6)
Syncope	2.5 (0 to 5)	−38.6 (−43.2 to −34.1)
UTIa	−10.8 (−14.1 to −7.5)	−56.7 (−61.3 to −52.2)
Pneumonia	−3.0 (−7.6 to 1.7)	23.4 (18.7 to 28.1)

^aUTI: urinary tract infection.

Table 2 also summarizes the estimated RDs, adjusted for latent health state and measured patient variables, between admission and discharge for 30-day revisits. Across all presentations, admission carried a significantly lower adjusted risk than a discharge of 30-day revisits (RD=−6.4%, 95% CI −7.8 to −5.0). Individual presentations yielded adjusted estimates that generally agreed with that of the entire sample, with the following exceptions. For patients with weakness, admission carried a numerically greater adjusted risk than a discharge of 30-day revisits (RD=4.4%, 95% CI −2.8 to 11.6). For patients with pneumonia, admission carried a significantly greater adjusted risk than a discharge of 30-day revisits (RD=23.4%, 95% CI 18.7 to 28.1).

Readmission

Table 3 summarizes the unadjusted RD between admission and discharge for 30-day readmissions across all diagnoses and for individual diagnoses. Across all diagnoses, admission carried a significantly greater unadjusted risk than a discharge of 30-day readmission (RD =14.8%, 95% CI 14.1 to 15.4). Individual diagnoses yielded unadjusted estimates that agreed in terms of direction, significance, and magnitude (within 3%) with that of the entire sample.

	Unadjusted RD (95% CI)	Adjusted RD (95% CI)
All	14.8 (14.1 to 15.4)	5.8 (5.0 to 6.5)
Falls	15.0 (14.1 to 15.9)	3.2 (2.3 to 4.0)
Weakness	14.3 (12.6 to 16.0)	61.6 (57.7 to 65.5)
Syncope	12.8 (11.1 to 14.6)	4.9 (3.0 to 6.9)
UTIa	17.9 (15.6 to 20.2)	9.4 (6.8 to 12.1)
Pneumonia	12.9 (8.7 to 17.2)	25.8 (20.9 to 30.6)

^aUTI: urinary tract infection.

Table 3 also summarizes estimated RDs, adjusted for latent health state and measured patient variables, between admission and discharge for 30-day readmissions. Across all diagnoses, admission carried a significantly greater adjusted risk than a discharge of 30-day readmission (RD=5.8%, 95% CI 5.0 to 6.5). Individual diagnoses yielded adjusted estimates in the same direction as that of the entire sample, but the magnitude was notably larger in a few cases. For patients with weakness, admission carried a greater adjusted risk than a discharge 30-day readmission (RD=61.6%, 95% CI 57.7 to 65.5). For patients with pneumonia, admission carried a significantly greater adjusted risk than a discharge of 30-day readmission (RD=25.8%, 95% CI [20.9, 30.6]).

Mortality

Table 4 summarizes the unadjusted RD between admission and discharge for 30-day mortality across all diagnoses and for individual diagnoses. Across all diagnoses, admission carried a significantly greater unadjusted risk than a discharge of 30-day mortality (RD=3.8%, 95% CI 3.3 to 4.3). Individual diagnoses yielded unadjusted estimates that agreed in terms of direction, significance, and magnitude (within 3%) with those of the entire sample.

	Unadjusted RD (95% CI)	Adjusted RD (95% CI)
All	3.8 (3.3 to 4.3)	1.0 (0.4 to 1.6)
Falls	2.8 (1.9 to 3.7)	−0.9 (−1.8 to 0.0)
Weakness	6.9 (5.3 to 8.5)	58.1 (54.1 to 62.0)
Syncope	2.4 (1.6 to 3.3)	−21.9 (−53.6 to 9.7)
UTIa	1.9 (0.7 to 3.1)	1.1 (−0.3 to 2.5)
Pneumonia	5.3 (1.4 to 9.2)	7.7 (4.4 to 11.0)

^aUTI: urinary tract infection.

Table 4 also summarizes estimated RDs, adjusted for latent health state and measured patient variables, between admission and discharge for 30-day mortality. Across all diagnoses, admission carried a significantly greater adjusted risk than a discharge of 30-day mortality (RD=1.0%, 95% CI 0.4 to 1.6). Individual diagnoses yielded adjusted estimates differing from that of the entire sample in several meaningful ways.

Reporting from high to low, admission for patients with weakness carried a significantly greater adjusted risk than a discharge of 30-day mortality (RD=58.1%, 95% CI 54.1 to 62.0). For patients with pneumonia, admission also carried a significantly adjusted risk than a discharge of 30-day mortality (RD=7.7%, 95% CI 4.4 to 11.0). For patients with UTI, admission carried a numerically greater adjusted risk than a discharge of 30-day mortality (RD=1.1%, 95% CI −0.3 to 2.5). For patients with syncope, admission carried a numerically lower adjusted risk of 30-day mortality (RD=−21.9%, 95% CI −53.6 to 9.7). For patients with falls, admission carried a lower adjusted risk than a discharge of 30-day mortality (RD=−0.9%, 95% CI −1.8 to 0.0).

Subgroup Analyses

Table 5 summarizes subgroup-specific estimates of RDs, adjusted for latent health state and measured patient variables, between admission and discharge for 30-day outcomes. All subgroup estimates had the same direction as that of the entire sample: negative for 30-day revisits and positive for 30-day readmission and mortality. For all 30-day outcomes, 5 subgroups had numerically greater adjusted estimates than that of the entire sample. These subgroups (in decreasing order of prominence) were individuals with congenital heart disease, individuals with diabetes, individuals with HCC scores greater than the mean score of 1.65, individuals greater than the mean age of 79.2 years, and individuals with hypertension. Of note, all these subgroups tend to have worse baseline health than their counterpart. The subgroup with the numerically lowest adjusted estimate for all 30-day outcomes was individuals with insurance other than Medicare.

Discussion

Principal Findings

Our goal was to investigate the impact of admitting older adult patients with 6 common but variably managed ED presentations: falls, weakness, syncope, UTI, pneumonia, and cellulitis. In particular, these conditions show considerable variability in admission practices, and understanding their specific outcomes can help clinicians better determine when admission is beneficial or potentially harmful. Using causal inference methods, we analyzed EHR data to compare outcomes such as ED revisits, hospital readmissions, and mortality, measuring these outcomes at 3, 9, and 30 days after discharge. By focusing on this “gray area” where no standard practice exists, we wanted to provide insights that support more consistent and evidence-based admission decisions for these presentations.

Our primary finding is that admission was associated with a significant decrease in the risk of an ED revisit, but with a significant increase in the risk of hospital readmission and mortality. The association with an increased readmission risk was consistent for all individual diagnoses, for all time frames, and for all estimation approaches (unadjusted, IPW, g-estimation, and latent variable as the primary analysis). The associations with a decreased revisit risk and with an increased mortality risk were generally consistent for individual diagnoses, all time frames, and all approaches, with a few exceptions discussed below. These results would suggest that while admission may help prevent immediate returns to the ED, it could lead to worse long-term outcomes for certain patients. This balance of risks highlights the importance of carefully weighing the short-term benefits of admission against the potential for readmission and mortality, especially in older adults with these conditions.

In our primary analysis, admission was associated with an increased revisit risk for patients with weakness and pneumonia, along with a notably higher readmission and mortality risk for these conditions compared with others. This suggests that patients with weakness or pneumonia may be especially vulnerable to poor outcomes from admission. In contrast, admission was associated with a numerically lower 30-day mortality risk for patients with falls or syncope, indicating that these individuals may benefit more from being admitted. These findings can help guide clinicians in prioritizing which patients may require closer monitoring, alternative care options, or potential admission.

Our subgroup analysis identified patient factors that may influence outcomes from hospital admission. Overall, admission appears less favorable for older individuals (>79 years) and those with comorbidities (ie, congenital heart disease, diabetes, high HCC scores, or hypertension). Important factors for admission, however, may vary by specific presentation, as shown in our subgroup analysis in

Multimedia Appendix 1

Accompanying appendix reporting (1) IPW and g-estimation results, (2) description of the latent variable approach, (3) 3-day outcomes results, (4) sensitivity analyses, (5) subgroup analyses for individual diagnoses, (6) analysis for cellulitis diagnosis, and (7) analysis for syndromic and no clear standard practice diagnoses.

For comparison, another study examined the impact of admitting older adult patients presenting to the ED with chest pain, another presentation with variable admission practices, and found that admission had a lower readmission risk but an increased mortality risk [Alvarez Avendaño SA, Cochran AL, Patterson B, Shah M, Smith M, Zayas-Cabán G. The average effect of emergency department admission on readmission and mortality for older adults with chest pain. Med Care. 2020;58(10):881-888. [CrossRef]35]. This study similarly estimated an increased mortality risk, yet it also found an increase in readmission risk for almost all presentations. The association of admission with both increased and decreased risks for revisits, readmission, and mortality, underscores why standardized care pathways are lacking for these presentations [Eagles D, Stiell IG, Clement CM, et al. International survey of emergency physicians’ priorities for clinical decision rules. Acad Emerg Med. Feb 2008;15(2):177-182. [CrossRef]21]. While our results suggest that, in general, discharge may be associated with lower readmission and mortality risks, discharging an older adult can still carry significant risks. These findings support the need for safe alternatives to hospitalization, such as outpatient care coordination or at-home ambulatory providers [Kilaru AS, Resnick D, Flynn D, et al. Practical Alternative to Hospitalization for Emergency Department Patients (PATH): A Feasibility Study. Healthcare Elsevier; 2021:100545. [CrossRef]38]. Such alternatives have been shown to effectively reduce hospital readmission rates on older adult patients discharged from the ED [Guttman A, Afilalo M, Guttman R, et al. An emergency department–based nurse discharge coordinator for elder patients: does it make a difference? Acad Emerg Med. Dec 2004;11(12):1318-1327. [CrossRef]39,Caplan GA, Williams AJ, Daly B, Abraham K. A randomized, controlled trial of comprehensive geriatric assessment and multidisciplinary intervention after discharge of elderly from the emergency department—the DEED II study. J Am Geriatr Soc. Sep 2004;52(9):1417-1423. [CrossRef]40].

Limitations

A primary limitation of this study is the potential for unmeasured confounding, especially confounding by indication, where treatment decisions are influenced by clinical indicators not fully captured in the EHR. This limitation is significant because it directly impacts our ability to interpret the findings causally. Although we used multiple methods, including a latent variable approach to account for unobserved health status [Cochran AL, Rathouz PJ, Kocher KE, Zayas‐Cabán G. A latent variable approach to potential outcomes for emergency department admission decisions. Stat Med. Sep 10, 2019;38(20):3911-3935. [CrossRef]34], no method can fully guarantee that all confounding is addressed. Consequently, our findings remain subject to potential bias if residual confounding exists. This is particularly relevant for interpreting the effect of admission on outcomes, as patients who are “sicker” based on unmeasured factors are more likely to be admitted, and these same factors could drive poor outcomes. While the latent variable approach yielded estimates generally more favorable to admission—suggesting some adjustment for underlying severity—and results were consistent across different time frames and presentations, caution is warranted in interpreting these effects as causal. This limitation highlights the need for further research to strengthen the causal inference of admission effects on outcomes.

In addition, this was a retrospective, observational study conducted using EHR data from an academic ED in the US Midwest, thereby limiting the generalizability of our findings. Moreover, the resulting patient population for specific presentations after applying our filters was relatively small, leading to nonsignificant estimates. Replicating this analysis in different EDs with larger samples is needed.

Conclusions

This study is one of the few to quantify the risks and benefits associated with admission decisions for ED presentations associated with unclear need for hospitalization. Existing studies have mainly focused on identifying factors that influence hospital admission of noncritically ill patients, often through physician surveys [Lewis Hunter AE, Spatz ES, Bernstein SL, Rosenthal MS. Factors influencing hospital admission of non-critically Ill patients presenting to the emergency department: a cross-sectional study. J Gen Intern Med. Jan 2016;31(1):37-44. [CrossRef] [Medline]41] or identifying risk patterns [Pines JM, Mutter RL, Zocchi MS. Variation in emergency department admission rates across the United States. Med Care Res Rev. Apr 2013;70(2):218-231. [CrossRef]33,Oster A, Bindman AB. Emergency department visits for ambulatory care sensitive conditions: insights into preventable hospitalizations. Med Care. 2003;41(1):198-207. [CrossRef]42,Pappas G, Hadden WC, Kozak LJ, Fisher GF. Potentially avoidable hospitalizations: inequalities in rates between US socioeconomic groups. Am J Public Health. May 1997;87(5):811-816. [CrossRef]43]. This study is intended to quantify the trade-offs of admission at a population level rather than directly influencing bedside decisions. Our findings indicate that admission is not always the “safe” choice, underscoring the need for further research to establish specific criteria that can better guide admission decisions for older adults in the ED.