- Email: [email protected]
Distinct subgroups of emergency department frequent users: A latent class analysis
YAJEM-58192; No of Pages 6 American Journal of Emergency Medicine xxx (xxxx) xxx
Contents lists available at ScienceDirect
American Journal of Emergency Medicine journal homepage: www.elsevier.com/locate/ajem
Distinct subgroups of emergency department frequent users: A latent class analysis☆ Lauren E. Birmingham, PhD, MA a,b,⁎, Vinay K. Cheruvu, PhD, MS, MSc, PGD-MISCA a, Jennifer A. Frey, PhD, CCRP b,c, Kirk A. Stiffler, MD, MPH d, Jonathan VanGeest, PhD a a
Kent State University, College of Public Health, Kent, OH, USA Summa Health System-Akron City Campus, Department of Emergency Medicine, Akron, OH, USA The Ohio State University, Department of Emergency Medicine, Columbus, OH, USA d Northeast Ohio Medical University, Rootstown, OH, USA b c
a r t i c l e
i n f o
Article history: Received 10 March 2019 Received in revised form 10 April 2019 Accepted 12 April 2019 Available online xxxx Keywords: Emergency department ED frequent user Latent class analysis Health care costs Health care utilization
a b s t r a c t Background: Emergency department (ED) frequent users have high resource utilization and associated costs. Many interventions have been designed to reduce utilization, but few have proved effective. This may be because this group is more heterogeneous than initially assumed, limiting the effectiveness of targeted interventions. The purpose of this study was to identify and describe distinct subgroups of ED frequent users and to estimate costs to provide hospital-based care to each group. Methods: Latent class analysis was used to identify homogeneous subgroups of ED frequent users. ED frequent users (n = 5731) from a single urban tertiary hospital-based ED and level 1 trauma center in 2014 were included. Descriptive statistics (counts and percentages) are described to characterize subgroups. A cost analysis was performed to examine differences in direct medical costs between subgroups from the healthcare provider perspective. Results: Four subgroups were identified and characterized: Short-term ED Frequent Users, Heart-related ED Frequent Users, Long-term ED Frequent Users, and Minor Care ED Frequent Users. The Heart-related group had the largest per person costs and the Long-term group had the largest total group costs. Conclusion: Distinct subgroups of ED frequent users were identified and described using a statistically objective method. This taxonomy of ED frequent users allows healthcare organizations to tailor interventions to specific subgroups of ED frequent users who can be targeted with tailored interventions. Cost data suggest intervention for long-term ED frequent users offers the greatest cost-avoidance benefit from a hospital perspective. © 2019 Published by Elsevier Inc.
1. Introduction Emergency department (ED) frequent users, are of great interest to health care organizations, insurers, and researchers alike as they account for a disproportionately large share of overall visits and costs [1]. ED frequent users are traditionally defined as those patients who make four or more ED visits in a 12-month period [2,3]. Overall, it is estimated that ED frequent users account for approximately 4.5–8% of the ED-using population and make 21–28% of all ED visits in the United States [2]. Per person yearly expenses for ED frequent users versus nonfrequent ED users vary substantially with an ED frequent user accruing Abbreviations: ED, emergency department; LCA, latent class analysis; df, degrees of freedom; G2, G-squared; SES, socioeconomic status; Adm, admission; IP, inpatient; Obs, observation. ☆ Presentations: Academy Health Annual Research Meeting, 2017, New Orleans, LA, Ohio Public Health Association, 2017, Columbus, OH. ⁎ Corresponding author at: Kent State University, College of Public Health, 800 Hilltop Drive, Moulton Hall, P.O. Box 5190, Kent, OH 44242, USA. E-mail address: [email protected] (L.E. Birmingham).
$89,033 in costs on average, compared $5700 for a non-frequent ED user [4]. As a result, impacting the health and utilization behaviors of the ED frequent user population poses a significant opportunity to improve health while simultaneously reducing healthcare costs. Growing healthcare costs paired with increasing numbers of ED frequent users highlights the critical need to intervene with this population [5]. However, randomized controlled trials have demonstrated limited success in reducing utilization and healthcare costs for ED frequent users [6,7]. One reason for the limited success found in prior interventions may be the lack of effective tailoring that can be done when a heterogeneous population is targeted with a universal intervention [8]. For example, many patients present with injuries from a fall, but interventions for a 25-year-old with a broken wrist from a fall from alcohol abuse would be different than those for a 75-year-old with a fall due to medication interactions. Program evaluation literature supports the use of targeted interventions over broad-based strategies [9,10]. Furthermore, recent work has demonstrated significant cost-savings associated with tailored interventions based on risk-stratified groups of patients with repeat hospitalizations [11]. We suspect that significant changes
https://doi.org/10.1016/j.ajem.2019.04.029 0735-6757/© 2019 Published by Elsevier Inc.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
2
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
in utilization in the entire population of ED frequent users may require more than one type of intervention to create change, however the subgroups to target are not yet well defined [12-14]. The purpose of this research was to identify and describe distinct subgroups of ED frequent users using an objective statistical methodology. The results of this analysis serve as a logical next step in the development of tailored interventions that could be targeted to the unique needs of clinically relevant subgroups of frequent ED users. Additionally, this analysis presents data to inform decisions about which subgroup would be most optimal to target from a healthcare costs perspective. 2. Materials and methods This was a retrospective cross-sectional analysis of ED visitors presenting to a large adult-only Midwestern hospital-based emergency department with approximately 100,000 ED visits per year and an associated level I trauma center. 2.1. Data sources Multiple internal data systems were used to conduct the analysis including the hospital electronic medical record, a quality improvement database, and the hospital accounting database that contained billing information. Data from calendar year 2014 was used in the primary analysis and data from 2015 was used to identify ED frequent users who demonstrated persistent frequent use in 2015. ED frequent users were identified if they made four or more ED visits in a 12-month period. Individuals younger than 18 years of age were excluded from the analysis. Approximately 30 variables were included for assessment in the statistical modeling procedure to support the development of a robust model. Variables included diagnostic categories summarizing clinical reasons for ED use, measures of ED and inpatient utilization (including indicators of very high-frequency use), 30-day readmissions, making an ED visit for a mental health (ICD-9: 295–299, 300.4, 309.1, 311) or substance use disorder (ICD-9: 291.xx, 292.xx, 303.xx, 304.xx, 305.xx), Charlson comorbidity index, demographic indicators (age, sex, race/ ethnicity), as well as measures of socioeconomic status (payer, median family income in zip code). 2.2. Statistical analysis Latent class analysis (LCA) was used to identify subgroups of ED frequent users. LCA is an objective statistical methodology that identifies homogeneous subgroups of individuals termed “classes”. This methodology has been used to create meaningful subgroupings of adults in healthrelated areas of study including alcohol dependency [15], drug adherence [16], and cigarette smoking [17]. SAS version 9.3 (Cary, NC) was used to conduct the data analysis with PROC LCA which was downloaded from The Methodology Center (University of Pennsylvania). The study was approved by the hospital's Institutional Review Board. A cost analysis was performed to determine the total costs of providing care from the hospital perspective. Direct medical costs for providing care were tabulated for each identified subgroup of ED frequent users. Both fixed and variable direct medical costs incurred during the emergency department visit and subsequent hospital encounter (if admitted) were included in the cost analysis. 3. Results A total of 123,594 ED visits were made by 70,959 patients in 2014 to the hospital where the study was conducted. Subsequently, 5814 people were identified as ED frequent users. Eighty-three (83) patients were excluded because they were b18 years of age. The remaining 5731 individuals were included for analysis. Table 1 depicts characteristics of the ED frequent user population. Patient demographics showed
that most patients were white (61.6%), female (60.0%), and publicly insured (73.3%), which is reflective of the total ED-using population at this site, and is also consistent with prior studies describing ED frequent users [2]. Data indicate that few patients exhibited chronic frequent ED use, defined as being an ED frequent user in both 2014 and 2015 (29.8%). 3.1. Latent class analysis models Selecting variables for inclusion in the LCA model was completed prior to fitting the number of classes, as is a best practice in LCA modeling [18]. Several models were tested iteratively using a diverse array of characteristics including diagnostic categories, rates of healthcare system utilization, chronic disease burden, as well as demographic and socioeconomic status measures. Variables were kept in the model if they demonstrated latent class separation as denoted by the item-response probabilities. The final model was determined based on several model fit criteria: G-squared (G2) being greater than the degrees of freedom (df), maximizing entropy, and minimizing the information criterion statistics. Additionally, clinical applicability of a variable, or the ability of a clinician to determine which resources would be helpful to a patient based on the variable itself at the time of presentation, was also considered. The models were reviewed and grounded in this fashion by practicing Emergency Medicine physicians. The final model included six variables: having a majority of 2014 ED visits be for a (1) respiratory, (2) mental health, (3) circulatory, or (4) musculoskeletal condition, as well as (5) lower rate of hospital admission following an ED visit (b50% of ED visits resulted in inpatient (IP) admission), and (6) being a ED frequent user in both 2014 and 2015. Once the final model was selected, the number of subgroups, or classes, was identified. A 4-class model optimized the model fit criteria outlined above. Patient gender was tested as a grouping variable in the final 4-class model to determine if measurement invariance would hold across the latent classes. Gender was evaluated as a grouping variable because it has been shown that men and women use the ED at different rates [19], and there are documented gender inequalities in emergency medicine care [20]. The statistical test for evaluating gender as a grouping variable failed to be rejected, and thus, the assumption of the model's ability to produce similar groupings for both men and women held. Several covariates were evaluated. Age, socioeconomic status (SES, as measured by the median income in zip code being above 200% of the federal poverty level), and very high ED utilization (defined by having 10+ ED visits in 2014) were tested as covariates. Only SES was found to be significantly associated with class membership. This analysis detected that there were lower odds of membership in Class 2 (Heart-related) and Class 3 (Long-term) (43% (95% confidence interval (CI): 0.40–0.83) and 47% (95% CI: 0.41–0.67) respectively) compared to Class 1 (Short-term) for an individual with an income below 200% of the federal poverty line. Class 4 (Minor care) did not differ significantly from Class 1 (Short-term). Given the statistical significance of the SES covariate, SES was included in the final model. The final model is displayed in Table 2. Class 1 (Short-term) members made up 40.31% of the population and were unlikely to be admitted to the hospital for an inpatient stay following an ED visit. Very few were predicted to be chronic ED frequent users (4.26%). Class 2 (Heartrelated) members were primarily characterized by most of their visits being for circulatory disorders. Class 3 (Long-term) members were defined by the high probability of being a chronic ED frequent user (61.75%). Class 4 (Minor care) members had a very high probability of not having ED visits result in inpatient admission (95.62%). One-third of their visits were predicted to be for musculoskeletal conditions (33.05%). Following final model selection and adjustment for SES, individuals were assigned to the class where they had the highest probability of class membership, as is standard practice in latent class analysis [18].
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
3
Table 1 ED frequent user characteristics.
Number of ED visits Number of inpatient & observation Visits Number of 30-day readmissions Median income for zip code Age (years)
Mean
SD
Median
Min
Max
IQR
6.24 1.46 0.51 $40,489 46.89
4.13 2.07 1.04 $15,384 19.69
5 0 0 $36,270 44
4 0 0 $13,015 18
89 18 12 $112,530 101
4–7 0–2 0–1 $30,958–46,909 44–60
Race Black White Other Sex Male Female Payer category Medicare Medicaid Commercial (private) Self-pay Chronic ED frequent user Yes No High inpatient (IP) and observation (Obs) visitors Made 4 + IP/Obs Made b4 IP/Obs High frequency user Made 10 + ED Made b10 ED ED admission (adm) percentage b50% of ED visits results in IP adm 50%+ of ED visits result in IP adm 30-day readmission category 0 readmissions 1 30-day readmission 2–3 30-day readmissions 4+ 30-day readmissions Charlson comorbidity index Score = 0 Score ≥ 1 ED visit for substance abuse disorder Yes No ED visit for mental health disorder Yes No Chronic disease categories 0 chronic diseases 1–2 chronic diseases 3–5 chronic diseases 6+ chronic diseases
Descriptive statistics for the latent classes are presented in Table 3. Each subgroup was assigned a descriptive group name based on the group characteristics. Several Emergency Medicine physicians assisted with the naming process to ensure the names were clinically relevant and descriptive. The first group, named “Short-term Frequent ED Users”, was the youngest group on average (average age 43). This group presented for a variety of conditions without any one diagnostic group being
Number
Percent
2047 3531 153
35.7% 61.6% 2.7%
2297 3434
40.0% 60.0%
1696 2504 769 762
29.6% 43.7% 13.4% 13.3%
1709 4022
29.8% 70.2%
837 4894
14.6% 85.4%
1073 4658
18.7% 81.3%
4370 1361
76.3% 23.8%
1915 479 251 66
70.6% 17.7% 9.3% 2.4%
5556 175
96.9% 3.1%
407 5324
7.1% 92.9%
506 5225
8.8% 91.2%
3069 1873 720 69
53.6% 32.7% 12.6% 1.2%
especially prevalent. Most notably, none (0%) of the group members were chronic frequent ED users. The second group termed “Heart-related Frequent ED Users” was primarily defined by the fact that all group members (100%) made most of their visits to the ED for heart-related conditions. Higher inpatient utilization was observed in this group with a median of 3 inpatient visits during 2014, and 66% of ED visits resulting in an inpatient admission.
Table 2 Final model, adjusted latent class analysis class membership & item-response probabilities. Class # Membership probability (standard error) Respiratory diagnosis Mental health diagnosis Musculoskeletal diagnosis Circulatory diagnosis Low ED admission (admitted after b50% of ED visits) Chronic ED frequent user
Class 1 - Short-term 40.31% (0.1677)
Class 2 - Heart related 4.16% (0.0038)
Class 3 - Long-term 35.46% (0.0581)
Class 4 - Minor care 20.07% (0.1428)
9.61% (0.0299) 4.07% (0.0164) 0.01% (0.0004) 0.67% (0.0051) 93.56% (0.0379) 4.26% (0.1097)
0.01% (0.0006) 0.01% (0.0005) 0.01% (0.0006) 97.83% (0.0443) 20.36% (0.0460) 50.33% (0.0397)
10.57% (0.0105) 10.65% (0.0105) 1.12% (0.0070) 0.00% (0.0002) 52.17% (0.0532) 61.75% (0.0546)
0.01% (0.0004) 0.00% (0.0004) 33.05% (0.2374) 0.00% (0.0003) 95.62% (0.0204) 20.49% (0.0278)
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
4
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
Table 3 Descriptive statistics of latent classes.
Average age % female % White % Medicaid % Medicare % Medicare and less than age 65 Median count of chronic diseases Median number of inpatient visits Average percent of ED visits that result in inpatient admission Chronic ED frequent Majority of ED visits in 2014 were for mental disorders Majority of ED visits in 2014 were for diseases of the circulatory system Majority of ED visits in 2014 were for diseases of the musculoskeletal system and connective tissue
The third group, “Long-term Frequent ED Users” was primarily characterized by the high prevalence of chronic ED frequent users (89% of group members). This group had the highest proportion of females (67%) and group members insured by Medicaid (50%) and Medicare while being b65 years of age (15%). On average, 41% of ED visits resulted in an inpatient admission for this group. This group demonstrated the highest percentage of people making a majority of 2014 ED visits for mental health diagnoses (10%). Lastly, the fourth group was named “Minor Care Frequent ED Users”. This group had the lowest average percentage of ED visits result in an inpatient admission (9%). Across the groups, this group was the least likely to have members insured by Medicaid (39%) and Medicare while being b65 years of age (9%). All members (100%) had visits that resulted in musculoskeletal discharge diagnoses which are inclusive of many diagnostic codes for sprains and strains. 3.2. Cost analysis A cost analysis was conducted to determine which group would generate the greatest cost-saving opportunity if offered an effective intervention to reduce utilization. As shown in Table 4, the Long-term Frequent ED Users had the largest total cost ($4,501,810) but the Heart-related Frequent ED Users had the largest per-person cost ($5609). The Short-term ED Frequent User group was the largest in terms of people (3383 members) but had lower total and per-person costs relative to other groups of ED frequent users. 4. Discussion This analysis used an objective statistical methodology to identify four unique latent classes, or subgroups, of ED frequent users—a Short-term, Long-term, Heart-related, and Minor Care ED Frequent User groups. The Heart-related group was the smallest group, and tended to be older, frequently discharged for heart-related conditions, and had the highest rate of ED visits result in inpatient or observation hospital admissions. Members of this group had the highest average cost per person. This echoes the focus by the Centers for Medicare and Medicaid Services (CMS) on reducing readmissions for patients with heart failure. Given the small size of the Heart-related group (249 patients), effective interventions could lead to relatively fast cost-avoidance benefits. However, given the advanced age and medical
Short-term (n = 3383)
Heart-related (n = 249)
Long-term (n = 1713)
Minor care (n = 386)
43 57% 63% 41% 31% 10% 0 0 14% 0% 4% 0% 0%
70 59% 62% 46% 33% 13% 1 3 66% 47% 0% 100% 0%
52 67% 58% 50% 27% 15% 0 1 41% 89% 10% 0% 1.0%
47 56% 66% 39% 29% 9% 0 0 9% 19% 0% 0% 100%
complexity of this group, some level of ED and inpatient hospital utilization will likely be unavoidable. The Minor Care ED Frequent User group presented most often for musculoskeletal complaints such as sprains and strains and was rarely admitted for inpatient or observation admission. This group has the lowest costs (total and per person) and had a low percentage of chronic ED frequent users (19%) suggesting that most of these patients resolve on their own within 12-months, as is generally true of most ED frequent users [12,21]. As such, these patients may not be an optimal group to target with intervention given their relatively low costs and high prevalence of self-resolution in a short period of time. This contrasts with recent policies by insurance companies focused on reducing low acuity ED utilization to reduce healthcare expenditures. These data clearly demonstrate that intervening with this group offers the smallest costavoidance benefit. Similarly, the Short-term group exhibited low per person costs, similar to the Minor Care group. However, as the largest group (3383 patients) it had the highest total costs. These individuals were primarily characterized by a complete lack of chronic ED frequent users in the group (0%), suggesting that intervention with this group may not be a good use of resources as they appear to naturally resolve on their own within one year. In sum, these data suggest that intervening with groups other than those presenting for minor illness or short-term bouts of frequent ED use will offer greater cost-avoidance benefits. Conversely the Long-term ED Frequent User group was primarily characterized by a high frequency of persistent ED use, and encompassed the highest percentage of ED frequent users who made the majority of their ED visits for mental health disorders. Furthermore, the Long-term group was the highest total cost group, and exhibited the second highest per-person costs. This combination of chronic use and high costs suggests that it would be the optimal subgroup to target with intervention(s). This is consistent with conclusions drawn by Smulowitz et al. in their analysis of ED cost-reduction strategies [22]. The present study builds on the body of literature examining characteristics of ED frequent users by determining how characteristics cooccur, and how this information can be used to create tailored and targeted interventions. This study takes the next logical step in creating effective interventions for this population. The present findings can inform future interventions in EDs or elsewhere, by providing information about the unique needs of distinct subgroups of ED frequent users to
Table 4 Cost analysis of ED frequent user subgroups.
Count of patients Sum of full costs Average full costs
Short-term frequent ED users
Heart-related frequent ED users
Long-term frequent ED users
Minor care frequent ED users
3383 $3,879,519 $1196
249 $1,273,288 $5609
1713 $4,501,810 $2807
386 $344,903 $922
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
assist in the development of tailored and targeted interventions. Recent data shows this strategy is effective. One study demonstrated approximately $26 million in charge avoidance by targeting interventions tailored to distinct needs of super-utilizers [11]. Additionally, this analysis suggests a strategy for determining which group should receive intervention to promote cost-effective decision-making. The present analysis is the first-step in creating targeted and tailored interventions for ED frequent users. An additional advantage of the development of a standard typology of ED frequent users is the potential for utilization of predictive models. Predictive models could prospectively sort patients into preventative interventions to decrease the likelihood of high ED utilization at the first sign of increased use and could potentially be incorporated into existing electronic medical record technology. Based on the current understanding of factors that drive frequent ED use, interventions will likely need to address not only healthcare needs, but also basic daily living, behavioral, and psychosocial needs [23]. Such preventative interventions could be aimed at addressing the root cause of frequent ED use (such as underlying health issues, social and environmental problems), rather than the symptom (ED visits). The present research was conducted at a single adult-only emergency department which could potentially limit the generalizability of the findings. ED utilization outside of the health system where this research was conducted could not be accessed and included in the analysis which may result in underestimation of ED and IP utilization rates. Future research should examine the potential for the existence of a standard typology of ED frequent users in a large national emergency department dataset and examine potential differences between typologies for pediatric and adult populations. Additionally, conducting this style of analysis with more recent data may influence the resulting latent classes since many emergency departments have recently seen large increases in rates of ED visits for opioid-related overdoses [24]. Cost data from a more inclusive perspective, such as the societal or payer perspective may provide more insight into the cost-savings that could be achieved through effective intervention. The costs included in this analysis were limited to only hospital-based costs, thus the few physicians who were consultants billing separately from the hospital were not included due to this data being unavailable. This may explain why costs per person reported in the present analysis are slightly lower than some other estimates [4]. While mortality was not included in the present analysis, future research can examine differing mortality rates across latent classes, and may inform opportunities for early palliative care involvement. Inclusion of additional information on unmet psychosocial needs, social determinants of health, or other confounders related to population health could increase the usefulness of this type of analysis, as these factors are largely believed to influence emergency department utilization [25-29]. The strength of this study is that few analyses have endeavored to create objectively defined distinct subgroups of ED frequent users, and none to our knowledge have utilized LCA. A few recent studies have analyzed groups of super-utilizers (primarily focused on frequent inpatient hospital utilization) using structural equation modeling techniques [30-32]. These studies of super-utilizers provide some validation of the subgroups identified in the present analysis since they also identified distinct groups with high levels of cardiac conditions [30,32], and another group with high levels of ED utilization and psychiatric and substance-related disorders [31]. A sizable portion of ED revisits may be driven by factors not readily amenable to modification by EDs as they stand today [33]; instead pointing to well-recognized fragmentations of the medical system through which patients fail to receive continuous or coordinated medical services, and as a result often end up in the ED. Others have pointed to the need for the ED to have greater involvement in population health interventions like care management, or involvement with community health agencies that provide care at home, pushing the ED to function beyond its traditional scope [22,34]. In this fashion, the ED has become
5
a center of care for entire communities, connecting patients at high-risk to community health resources and other agencies. 5. Conclusions A greater understanding of subgroups of ED frequent users will be indispensable in helping population health practitioners tailor interventions to the distinct needs of subgroups. Tailored interventions have demonstrated greater effectiveness than universal, non-targeted interventions [9-11,35]. This analysis identified four homogeneous groups of ED frequent users that have different needs and user patterns and will likely require different interventions or resources. Interventions that are tailored and targeted to distinct homogeneous groups of ED frequent users, such as those identified here, may help drive improved results with respect to costs, utilization, and health status in the frequent ED user population. The present analysis also provides a framework for those providing financial resources for interventions to consider how they will achieve the greatest cost-savings with frequent ED user interventions. Declarations of interest None. References [1] Cook LJ, Knight S, Junkins Jr EP, Mann NC, Dean JM, Olson LM. Repeat patients to the emergency department in a statewide database. Acad Emerg Med 2004;11(3): 256–63. [2] LaCalle E, Rabin E. Frequent users of emergency departments: the myths, the data, and the policy implications. Ann Emerg Med 2010;56(1):42–8. [3] Hunt KA, Weber EJ, Showstack JA, Colby DC, Callaham ML. Characteristics of frequent users of emergency departments. Ann Emerg Med 2006;48(1):1–8. [4] South Carolina Public Health Institute. A report on frequent users of hospital emergency departments in South Carolina; 2011. [5] Martin GB, Stokes-Buzzelli SA, Peltzer-Jones JM, Schultz LR. Ten years of frequent users in an urban emergency department. West J Emerg Med 2013;14(3):243–6. [6] Althaus F, Paroz S, Hugli O, Ghali WA, Daeppen JB, Peytremann-Bridevaux I, et al. Effectiveness of interventions targeting frequent users of emergency departments: a systematic review. Ann Emerg Med 2011;58(1):41–52 [e42]. [7] Soril LJ, Leggett LE, Lorenzetti DL, Noseworthy TW, Clement FM. Reducing frequent visits to the emergency department: a systematic review of interventions. PLoS One 2015;10(4):e0123660. [8] Brown RS, Peikes D, Peterson G, Schore J, Razafindrakoto CM. Six features of Medicare coordinated care demonstration programs that cut hospital admissions of high-risk patients. Health Aff (Millwood) 2012;31(6):1156–66. [9] Issel LM. Health program planning and evaluation: a practical, systematic approach for community health. Boston, MA: Jones and Barlett Publishers; 2004. [10] Cutrona SL, Choudhry NK, Fischer MA, Servi AD, Stedman M, Liberman JN, et al. Targeting cardiovascular medication adherence interventions. J Am Pharm Assoc (2003) 2012;52(3):381–97. [11] Durfee J, Johnson T, Batal H, Long J, Rinehart D, Everhart R, et al. The impact of tailored intervention services on charges and mortality for adult super-utilizers. Healthc (Amst) 2018;6(4):253–8. [12] Johnson TL, Rinehart DJ, Durfee J, Brewer D, Batal H, Blum J, et al. For many patients who use large amounts of health care services, the need is intense yet temporary. Health Aff 2015;34(8):1312–9. [13] Krieg C, Hudon C, Chouinard MC, Dufour I. Individual predictors of frequent emergency department use: a scoping review. BMC Health Serv Res 2016;16(1):594. [14] Kanzaria HK, Niedzwiecki MJ, Montoy JC, Raven MC, Hsia RY. Persistent frequent emergency department use: core group exhibits extreme levels of use for more than a decade. Health Aff 2017;36(10):1720–8. [15] Sacco P, Bucholz KK, Spitznagel EL. Alcohol use among older adults in the National Epidemiologic Survey on Alcohol and Related Conditions: a latent class analysis. J Stud Alcohol Drugs 2009;70(6):829–38. [16] Ahn J, McCombs JS, Jung C, Croudace TJ, McDonnell D, Ascher-Svanum H, et al. Classifying patients by antipsychotic adherence patterns using latent class analysis: characteristics of nonadherent groups in the California Medicaid (Medi-Cal) program. Value Health 2008;11(1):48–56. [17] Niu L, Luo D, Silenzio VM, Xiao S, Tian Y. Are informing knowledge and supportive attitude enough for tobacco control? A latent class analysis of cigarette smoking patterns among medical teachers in China. Int J Environ Res Public Health 2015;12(10): 12030–42. [18] Lanza ST, Collins LM, Lemmon DR, Schafer JL. PROC LCA: a SAS procedure for latent class analysis. Struct Equ Model 2007;14(4):671–94. [19] Anson O, Carmel S, Levin M. Gender differences in the utilization of emergency department services. Women Health 1991;17(2):91–104. [20] Fassler J. How doctor's take women's pain less seriously. The Atlantic 2015 [Oct 15]. [21] Billings J, Raven MC. Dispelling an urban legend: frequent emergency department users have substantial burden of disease. Health Aff 2013;32(12):2099–108.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
6
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
[22] Smulowitz PB, Honigman L, Landon BE. A novel approach to identifying targets for cost reduction in the emergency department. Ann Emerg Med 2013;61(3):293–300. [23] Birmingham LE, Cochran T, Frey JA, Stiffler KA, Wilber ST. Emergency department use and barriers to wellness: a survey of emergency department frequent users. BMC Emerg Med 2017;17(1):16. [24] Birmingham LE, Nielson JA. An increase in per-patient naloxone requirements in an opioid epidemic. Am J Emerg Med 2017;35(12):1958–9. [25] McCusker J, Roberge D, Ciampi A, Levesque JF, Pineault R, Belzile E, et al. Primary care organization and outcomes of an emergency visit among seniors. Healthc Policy 2009;5(1):e115–31. [26] Haggerty JL, Roberge D, Pineault R, Larouche D, Touati N. Features of primary healthcare clinics associated with patients' utilization of emergency rooms: urbanrural differences. Healthc Policy 2007;3(2):72–85. [27] Kushel MB, Gupta R, Gee L, Haas JS. Housing instability and food insecurity as barriers to health care among low-income Americans. J Gen Intern Med 2006;21 (1):71–7. [28] Bodenmann P, Velonaki VS, Ruggeri O, Hugli O, Burnand B, Wasserfallen JB, et al. Case management for frequent users of the emergency department: study protocol of a randomised controlled trial. BMC Health Serv Res 2014;14:264.
[29] Kangovi S, Barg FK, Carter T, Long JA, Shannon R, Grande D. Understanding why patients of low socioeconomic status prefer hospitals over ambulatory care. Health Aff 2013;32(7):1196–203. [30] Lee NS, Whitman N, Vakharia N, Ph DG, Rothberg MB. High-cost patients: hot-spotters don't explain the half of it. J Gen Intern Med 2017;32(1):28–34. [31] Rinehart DJ, Oronce C, Durfee MJ, Ranby KW, Batal HA, Hanratty R, et al. Identifying subgroups of adult superutilizers in an urban safety-net system using latent class analysis: implications for clinical practice. Med Care 2018;56(1):e1–9. [32] Davis AC, Shen E, Shah NR, Glenn BA, Ponce N, Telesca D, et al. Segmentation of highcost adults in an integrated healthcare system based on empirical clustering of acute and chronic conditions. J Gen Intern Med 2018;33(12):2171–9. [33] Pham JC, Bayram JD, Moss D. Characteristics of frequent users of three hospital emergency departments. Agency for Healthcare Research & Quality: Department of Health & Human Services; 2017. [34] Sarfaty M. Health affairs blog. [Internet]. Available from: Health Affairs. https:// www.healthaffairs.org/do/10.1377/hblog20140103.036069/full/; 2014. [35] Meropol SB, Schiltz NK, Sattar A, Stange KC, Nevar AH, Davey C, et al. Practice-tailored facilitation to improve pediatric preventive care delivery: a randomized trial. Pediatrics 2014;133(6):e1664–75.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
Contents lists available at ScienceDirect
American Journal of Emergency Medicine journal homepage: www.elsevier.com/locate/ajem
Distinct subgroups of emergency department frequent users: A latent class analysis☆ Lauren E. Birmingham, PhD, MA a,b,⁎, Vinay K. Cheruvu, PhD, MS, MSc, PGD-MISCA a, Jennifer A. Frey, PhD, CCRP b,c, Kirk A. Stiffler, MD, MPH d, Jonathan VanGeest, PhD a a
Kent State University, College of Public Health, Kent, OH, USA Summa Health System-Akron City Campus, Department of Emergency Medicine, Akron, OH, USA The Ohio State University, Department of Emergency Medicine, Columbus, OH, USA d Northeast Ohio Medical University, Rootstown, OH, USA b c
a r t i c l e
i n f o
Article history: Received 10 March 2019 Received in revised form 10 April 2019 Accepted 12 April 2019 Available online xxxx Keywords: Emergency department ED frequent user Latent class analysis Health care costs Health care utilization
a b s t r a c t Background: Emergency department (ED) frequent users have high resource utilization and associated costs. Many interventions have been designed to reduce utilization, but few have proved effective. This may be because this group is more heterogeneous than initially assumed, limiting the effectiveness of targeted interventions. The purpose of this study was to identify and describe distinct subgroups of ED frequent users and to estimate costs to provide hospital-based care to each group. Methods: Latent class analysis was used to identify homogeneous subgroups of ED frequent users. ED frequent users (n = 5731) from a single urban tertiary hospital-based ED and level 1 trauma center in 2014 were included. Descriptive statistics (counts and percentages) are described to characterize subgroups. A cost analysis was performed to examine differences in direct medical costs between subgroups from the healthcare provider perspective. Results: Four subgroups were identified and characterized: Short-term ED Frequent Users, Heart-related ED Frequent Users, Long-term ED Frequent Users, and Minor Care ED Frequent Users. The Heart-related group had the largest per person costs and the Long-term group had the largest total group costs. Conclusion: Distinct subgroups of ED frequent users were identified and described using a statistically objective method. This taxonomy of ED frequent users allows healthcare organizations to tailor interventions to specific subgroups of ED frequent users who can be targeted with tailored interventions. Cost data suggest intervention for long-term ED frequent users offers the greatest cost-avoidance benefit from a hospital perspective. © 2019 Published by Elsevier Inc.
1. Introduction Emergency department (ED) frequent users, are of great interest to health care organizations, insurers, and researchers alike as they account for a disproportionately large share of overall visits and costs [1]. ED frequent users are traditionally defined as those patients who make four or more ED visits in a 12-month period [2,3]. Overall, it is estimated that ED frequent users account for approximately 4.5–8% of the ED-using population and make 21–28% of all ED visits in the United States [2]. Per person yearly expenses for ED frequent users versus nonfrequent ED users vary substantially with an ED frequent user accruing Abbreviations: ED, emergency department; LCA, latent class analysis; df, degrees of freedom; G2, G-squared; SES, socioeconomic status; Adm, admission; IP, inpatient; Obs, observation. ☆ Presentations: Academy Health Annual Research Meeting, 2017, New Orleans, LA, Ohio Public Health Association, 2017, Columbus, OH. ⁎ Corresponding author at: Kent State University, College of Public Health, 800 Hilltop Drive, Moulton Hall, P.O. Box 5190, Kent, OH 44242, USA. E-mail address: [email protected] (L.E. Birmingham).
$89,033 in costs on average, compared $5700 for a non-frequent ED user [4]. As a result, impacting the health and utilization behaviors of the ED frequent user population poses a significant opportunity to improve health while simultaneously reducing healthcare costs. Growing healthcare costs paired with increasing numbers of ED frequent users highlights the critical need to intervene with this population [5]. However, randomized controlled trials have demonstrated limited success in reducing utilization and healthcare costs for ED frequent users [6,7]. One reason for the limited success found in prior interventions may be the lack of effective tailoring that can be done when a heterogeneous population is targeted with a universal intervention [8]. For example, many patients present with injuries from a fall, but interventions for a 25-year-old with a broken wrist from a fall from alcohol abuse would be different than those for a 75-year-old with a fall due to medication interactions. Program evaluation literature supports the use of targeted interventions over broad-based strategies [9,10]. Furthermore, recent work has demonstrated significant cost-savings associated with tailored interventions based on risk-stratified groups of patients with repeat hospitalizations [11]. We suspect that significant changes
https://doi.org/10.1016/j.ajem.2019.04.029 0735-6757/© 2019 Published by Elsevier Inc.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
2
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
in utilization in the entire population of ED frequent users may require more than one type of intervention to create change, however the subgroups to target are not yet well defined [12-14]. The purpose of this research was to identify and describe distinct subgroups of ED frequent users using an objective statistical methodology. The results of this analysis serve as a logical next step in the development of tailored interventions that could be targeted to the unique needs of clinically relevant subgroups of frequent ED users. Additionally, this analysis presents data to inform decisions about which subgroup would be most optimal to target from a healthcare costs perspective. 2. Materials and methods This was a retrospective cross-sectional analysis of ED visitors presenting to a large adult-only Midwestern hospital-based emergency department with approximately 100,000 ED visits per year and an associated level I trauma center. 2.1. Data sources Multiple internal data systems were used to conduct the analysis including the hospital electronic medical record, a quality improvement database, and the hospital accounting database that contained billing information. Data from calendar year 2014 was used in the primary analysis and data from 2015 was used to identify ED frequent users who demonstrated persistent frequent use in 2015. ED frequent users were identified if they made four or more ED visits in a 12-month period. Individuals younger than 18 years of age were excluded from the analysis. Approximately 30 variables were included for assessment in the statistical modeling procedure to support the development of a robust model. Variables included diagnostic categories summarizing clinical reasons for ED use, measures of ED and inpatient utilization (including indicators of very high-frequency use), 30-day readmissions, making an ED visit for a mental health (ICD-9: 295–299, 300.4, 309.1, 311) or substance use disorder (ICD-9: 291.xx, 292.xx, 303.xx, 304.xx, 305.xx), Charlson comorbidity index, demographic indicators (age, sex, race/ ethnicity), as well as measures of socioeconomic status (payer, median family income in zip code). 2.2. Statistical analysis Latent class analysis (LCA) was used to identify subgroups of ED frequent users. LCA is an objective statistical methodology that identifies homogeneous subgroups of individuals termed “classes”. This methodology has been used to create meaningful subgroupings of adults in healthrelated areas of study including alcohol dependency [15], drug adherence [16], and cigarette smoking [17]. SAS version 9.3 (Cary, NC) was used to conduct the data analysis with PROC LCA which was downloaded from The Methodology Center (University of Pennsylvania). The study was approved by the hospital's Institutional Review Board. A cost analysis was performed to determine the total costs of providing care from the hospital perspective. Direct medical costs for providing care were tabulated for each identified subgroup of ED frequent users. Both fixed and variable direct medical costs incurred during the emergency department visit and subsequent hospital encounter (if admitted) were included in the cost analysis. 3. Results A total of 123,594 ED visits were made by 70,959 patients in 2014 to the hospital where the study was conducted. Subsequently, 5814 people were identified as ED frequent users. Eighty-three (83) patients were excluded because they were b18 years of age. The remaining 5731 individuals were included for analysis. Table 1 depicts characteristics of the ED frequent user population. Patient demographics showed
that most patients were white (61.6%), female (60.0%), and publicly insured (73.3%), which is reflective of the total ED-using population at this site, and is also consistent with prior studies describing ED frequent users [2]. Data indicate that few patients exhibited chronic frequent ED use, defined as being an ED frequent user in both 2014 and 2015 (29.8%). 3.1. Latent class analysis models Selecting variables for inclusion in the LCA model was completed prior to fitting the number of classes, as is a best practice in LCA modeling [18]. Several models were tested iteratively using a diverse array of characteristics including diagnostic categories, rates of healthcare system utilization, chronic disease burden, as well as demographic and socioeconomic status measures. Variables were kept in the model if they demonstrated latent class separation as denoted by the item-response probabilities. The final model was determined based on several model fit criteria: G-squared (G2) being greater than the degrees of freedom (df), maximizing entropy, and minimizing the information criterion statistics. Additionally, clinical applicability of a variable, or the ability of a clinician to determine which resources would be helpful to a patient based on the variable itself at the time of presentation, was also considered. The models were reviewed and grounded in this fashion by practicing Emergency Medicine physicians. The final model included six variables: having a majority of 2014 ED visits be for a (1) respiratory, (2) mental health, (3) circulatory, or (4) musculoskeletal condition, as well as (5) lower rate of hospital admission following an ED visit (b50% of ED visits resulted in inpatient (IP) admission), and (6) being a ED frequent user in both 2014 and 2015. Once the final model was selected, the number of subgroups, or classes, was identified. A 4-class model optimized the model fit criteria outlined above. Patient gender was tested as a grouping variable in the final 4-class model to determine if measurement invariance would hold across the latent classes. Gender was evaluated as a grouping variable because it has been shown that men and women use the ED at different rates [19], and there are documented gender inequalities in emergency medicine care [20]. The statistical test for evaluating gender as a grouping variable failed to be rejected, and thus, the assumption of the model's ability to produce similar groupings for both men and women held. Several covariates were evaluated. Age, socioeconomic status (SES, as measured by the median income in zip code being above 200% of the federal poverty level), and very high ED utilization (defined by having 10+ ED visits in 2014) were tested as covariates. Only SES was found to be significantly associated with class membership. This analysis detected that there were lower odds of membership in Class 2 (Heart-related) and Class 3 (Long-term) (43% (95% confidence interval (CI): 0.40–0.83) and 47% (95% CI: 0.41–0.67) respectively) compared to Class 1 (Short-term) for an individual with an income below 200% of the federal poverty line. Class 4 (Minor care) did not differ significantly from Class 1 (Short-term). Given the statistical significance of the SES covariate, SES was included in the final model. The final model is displayed in Table 2. Class 1 (Short-term) members made up 40.31% of the population and were unlikely to be admitted to the hospital for an inpatient stay following an ED visit. Very few were predicted to be chronic ED frequent users (4.26%). Class 2 (Heartrelated) members were primarily characterized by most of their visits being for circulatory disorders. Class 3 (Long-term) members were defined by the high probability of being a chronic ED frequent user (61.75%). Class 4 (Minor care) members had a very high probability of not having ED visits result in inpatient admission (95.62%). One-third of their visits were predicted to be for musculoskeletal conditions (33.05%). Following final model selection and adjustment for SES, individuals were assigned to the class where they had the highest probability of class membership, as is standard practice in latent class analysis [18].
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
3
Table 1 ED frequent user characteristics.
Number of ED visits Number of inpatient & observation Visits Number of 30-day readmissions Median income for zip code Age (years)
Mean
SD
Median
Min
Max
IQR
6.24 1.46 0.51 $40,489 46.89
4.13 2.07 1.04 $15,384 19.69
5 0 0 $36,270 44
4 0 0 $13,015 18
89 18 12 $112,530 101
4–7 0–2 0–1 $30,958–46,909 44–60
Race Black White Other Sex Male Female Payer category Medicare Medicaid Commercial (private) Self-pay Chronic ED frequent user Yes No High inpatient (IP) and observation (Obs) visitors Made 4 + IP/Obs Made b4 IP/Obs High frequency user Made 10 + ED Made b10 ED ED admission (adm) percentage b50% of ED visits results in IP adm 50%+ of ED visits result in IP adm 30-day readmission category 0 readmissions 1 30-day readmission 2–3 30-day readmissions 4+ 30-day readmissions Charlson comorbidity index Score = 0 Score ≥ 1 ED visit for substance abuse disorder Yes No ED visit for mental health disorder Yes No Chronic disease categories 0 chronic diseases 1–2 chronic diseases 3–5 chronic diseases 6+ chronic diseases
Descriptive statistics for the latent classes are presented in Table 3. Each subgroup was assigned a descriptive group name based on the group characteristics. Several Emergency Medicine physicians assisted with the naming process to ensure the names were clinically relevant and descriptive. The first group, named “Short-term Frequent ED Users”, was the youngest group on average (average age 43). This group presented for a variety of conditions without any one diagnostic group being
Number
Percent
2047 3531 153
35.7% 61.6% 2.7%
2297 3434
40.0% 60.0%
1696 2504 769 762
29.6% 43.7% 13.4% 13.3%
1709 4022
29.8% 70.2%
837 4894
14.6% 85.4%
1073 4658
18.7% 81.3%
4370 1361
76.3% 23.8%
1915 479 251 66
70.6% 17.7% 9.3% 2.4%
5556 175
96.9% 3.1%
407 5324
7.1% 92.9%
506 5225
8.8% 91.2%
3069 1873 720 69
53.6% 32.7% 12.6% 1.2%
especially prevalent. Most notably, none (0%) of the group members were chronic frequent ED users. The second group termed “Heart-related Frequent ED Users” was primarily defined by the fact that all group members (100%) made most of their visits to the ED for heart-related conditions. Higher inpatient utilization was observed in this group with a median of 3 inpatient visits during 2014, and 66% of ED visits resulting in an inpatient admission.
Table 2 Final model, adjusted latent class analysis class membership & item-response probabilities. Class # Membership probability (standard error) Respiratory diagnosis Mental health diagnosis Musculoskeletal diagnosis Circulatory diagnosis Low ED admission (admitted after b50% of ED visits) Chronic ED frequent user
Class 1 - Short-term 40.31% (0.1677)
Class 2 - Heart related 4.16% (0.0038)
Class 3 - Long-term 35.46% (0.0581)
Class 4 - Minor care 20.07% (0.1428)
9.61% (0.0299) 4.07% (0.0164) 0.01% (0.0004) 0.67% (0.0051) 93.56% (0.0379) 4.26% (0.1097)
0.01% (0.0006) 0.01% (0.0005) 0.01% (0.0006) 97.83% (0.0443) 20.36% (0.0460) 50.33% (0.0397)
10.57% (0.0105) 10.65% (0.0105) 1.12% (0.0070) 0.00% (0.0002) 52.17% (0.0532) 61.75% (0.0546)
0.01% (0.0004) 0.00% (0.0004) 33.05% (0.2374) 0.00% (0.0003) 95.62% (0.0204) 20.49% (0.0278)
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
4
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
Table 3 Descriptive statistics of latent classes.
Average age % female % White % Medicaid % Medicare % Medicare and less than age 65 Median count of chronic diseases Median number of inpatient visits Average percent of ED visits that result in inpatient admission Chronic ED frequent Majority of ED visits in 2014 were for mental disorders Majority of ED visits in 2014 were for diseases of the circulatory system Majority of ED visits in 2014 were for diseases of the musculoskeletal system and connective tissue
The third group, “Long-term Frequent ED Users” was primarily characterized by the high prevalence of chronic ED frequent users (89% of group members). This group had the highest proportion of females (67%) and group members insured by Medicaid (50%) and Medicare while being b65 years of age (15%). On average, 41% of ED visits resulted in an inpatient admission for this group. This group demonstrated the highest percentage of people making a majority of 2014 ED visits for mental health diagnoses (10%). Lastly, the fourth group was named “Minor Care Frequent ED Users”. This group had the lowest average percentage of ED visits result in an inpatient admission (9%). Across the groups, this group was the least likely to have members insured by Medicaid (39%) and Medicare while being b65 years of age (9%). All members (100%) had visits that resulted in musculoskeletal discharge diagnoses which are inclusive of many diagnostic codes for sprains and strains. 3.2. Cost analysis A cost analysis was conducted to determine which group would generate the greatest cost-saving opportunity if offered an effective intervention to reduce utilization. As shown in Table 4, the Long-term Frequent ED Users had the largest total cost ($4,501,810) but the Heart-related Frequent ED Users had the largest per-person cost ($5609). The Short-term ED Frequent User group was the largest in terms of people (3383 members) but had lower total and per-person costs relative to other groups of ED frequent users. 4. Discussion This analysis used an objective statistical methodology to identify four unique latent classes, or subgroups, of ED frequent users—a Short-term, Long-term, Heart-related, and Minor Care ED Frequent User groups. The Heart-related group was the smallest group, and tended to be older, frequently discharged for heart-related conditions, and had the highest rate of ED visits result in inpatient or observation hospital admissions. Members of this group had the highest average cost per person. This echoes the focus by the Centers for Medicare and Medicaid Services (CMS) on reducing readmissions for patients with heart failure. Given the small size of the Heart-related group (249 patients), effective interventions could lead to relatively fast cost-avoidance benefits. However, given the advanced age and medical
Short-term (n = 3383)
Heart-related (n = 249)
Long-term (n = 1713)
Minor care (n = 386)
43 57% 63% 41% 31% 10% 0 0 14% 0% 4% 0% 0%
70 59% 62% 46% 33% 13% 1 3 66% 47% 0% 100% 0%
52 67% 58% 50% 27% 15% 0 1 41% 89% 10% 0% 1.0%
47 56% 66% 39% 29% 9% 0 0 9% 19% 0% 0% 100%
complexity of this group, some level of ED and inpatient hospital utilization will likely be unavoidable. The Minor Care ED Frequent User group presented most often for musculoskeletal complaints such as sprains and strains and was rarely admitted for inpatient or observation admission. This group has the lowest costs (total and per person) and had a low percentage of chronic ED frequent users (19%) suggesting that most of these patients resolve on their own within 12-months, as is generally true of most ED frequent users [12,21]. As such, these patients may not be an optimal group to target with intervention given their relatively low costs and high prevalence of self-resolution in a short period of time. This contrasts with recent policies by insurance companies focused on reducing low acuity ED utilization to reduce healthcare expenditures. These data clearly demonstrate that intervening with this group offers the smallest costavoidance benefit. Similarly, the Short-term group exhibited low per person costs, similar to the Minor Care group. However, as the largest group (3383 patients) it had the highest total costs. These individuals were primarily characterized by a complete lack of chronic ED frequent users in the group (0%), suggesting that intervention with this group may not be a good use of resources as they appear to naturally resolve on their own within one year. In sum, these data suggest that intervening with groups other than those presenting for minor illness or short-term bouts of frequent ED use will offer greater cost-avoidance benefits. Conversely the Long-term ED Frequent User group was primarily characterized by a high frequency of persistent ED use, and encompassed the highest percentage of ED frequent users who made the majority of their ED visits for mental health disorders. Furthermore, the Long-term group was the highest total cost group, and exhibited the second highest per-person costs. This combination of chronic use and high costs suggests that it would be the optimal subgroup to target with intervention(s). This is consistent with conclusions drawn by Smulowitz et al. in their analysis of ED cost-reduction strategies [22]. The present study builds on the body of literature examining characteristics of ED frequent users by determining how characteristics cooccur, and how this information can be used to create tailored and targeted interventions. This study takes the next logical step in creating effective interventions for this population. The present findings can inform future interventions in EDs or elsewhere, by providing information about the unique needs of distinct subgroups of ED frequent users to
Table 4 Cost analysis of ED frequent user subgroups.
Count of patients Sum of full costs Average full costs
Short-term frequent ED users
Heart-related frequent ED users
Long-term frequent ED users
Minor care frequent ED users
3383 $3,879,519 $1196
249 $1,273,288 $5609
1713 $4,501,810 $2807
386 $344,903 $922
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
assist in the development of tailored and targeted interventions. Recent data shows this strategy is effective. One study demonstrated approximately $26 million in charge avoidance by targeting interventions tailored to distinct needs of super-utilizers [11]. Additionally, this analysis suggests a strategy for determining which group should receive intervention to promote cost-effective decision-making. The present analysis is the first-step in creating targeted and tailored interventions for ED frequent users. An additional advantage of the development of a standard typology of ED frequent users is the potential for utilization of predictive models. Predictive models could prospectively sort patients into preventative interventions to decrease the likelihood of high ED utilization at the first sign of increased use and could potentially be incorporated into existing electronic medical record technology. Based on the current understanding of factors that drive frequent ED use, interventions will likely need to address not only healthcare needs, but also basic daily living, behavioral, and psychosocial needs [23]. Such preventative interventions could be aimed at addressing the root cause of frequent ED use (such as underlying health issues, social and environmental problems), rather than the symptom (ED visits). The present research was conducted at a single adult-only emergency department which could potentially limit the generalizability of the findings. ED utilization outside of the health system where this research was conducted could not be accessed and included in the analysis which may result in underestimation of ED and IP utilization rates. Future research should examine the potential for the existence of a standard typology of ED frequent users in a large national emergency department dataset and examine potential differences between typologies for pediatric and adult populations. Additionally, conducting this style of analysis with more recent data may influence the resulting latent classes since many emergency departments have recently seen large increases in rates of ED visits for opioid-related overdoses [24]. Cost data from a more inclusive perspective, such as the societal or payer perspective may provide more insight into the cost-savings that could be achieved through effective intervention. The costs included in this analysis were limited to only hospital-based costs, thus the few physicians who were consultants billing separately from the hospital were not included due to this data being unavailable. This may explain why costs per person reported in the present analysis are slightly lower than some other estimates [4]. While mortality was not included in the present analysis, future research can examine differing mortality rates across latent classes, and may inform opportunities for early palliative care involvement. Inclusion of additional information on unmet psychosocial needs, social determinants of health, or other confounders related to population health could increase the usefulness of this type of analysis, as these factors are largely believed to influence emergency department utilization [25-29]. The strength of this study is that few analyses have endeavored to create objectively defined distinct subgroups of ED frequent users, and none to our knowledge have utilized LCA. A few recent studies have analyzed groups of super-utilizers (primarily focused on frequent inpatient hospital utilization) using structural equation modeling techniques [30-32]. These studies of super-utilizers provide some validation of the subgroups identified in the present analysis since they also identified distinct groups with high levels of cardiac conditions [30,32], and another group with high levels of ED utilization and psychiatric and substance-related disorders [31]. A sizable portion of ED revisits may be driven by factors not readily amenable to modification by EDs as they stand today [33]; instead pointing to well-recognized fragmentations of the medical system through which patients fail to receive continuous or coordinated medical services, and as a result often end up in the ED. Others have pointed to the need for the ED to have greater involvement in population health interventions like care management, or involvement with community health agencies that provide care at home, pushing the ED to function beyond its traditional scope [22,34]. In this fashion, the ED has become
5
a center of care for entire communities, connecting patients at high-risk to community health resources and other agencies. 5. Conclusions A greater understanding of subgroups of ED frequent users will be indispensable in helping population health practitioners tailor interventions to the distinct needs of subgroups. Tailored interventions have demonstrated greater effectiveness than universal, non-targeted interventions [9-11,35]. This analysis identified four homogeneous groups of ED frequent users that have different needs and user patterns and will likely require different interventions or resources. Interventions that are tailored and targeted to distinct homogeneous groups of ED frequent users, such as those identified here, may help drive improved results with respect to costs, utilization, and health status in the frequent ED user population. The present analysis also provides a framework for those providing financial resources for interventions to consider how they will achieve the greatest cost-savings with frequent ED user interventions. Declarations of interest None. References [1] Cook LJ, Knight S, Junkins Jr EP, Mann NC, Dean JM, Olson LM. Repeat patients to the emergency department in a statewide database. Acad Emerg Med 2004;11(3): 256–63. [2] LaCalle E, Rabin E. Frequent users of emergency departments: the myths, the data, and the policy implications. Ann Emerg Med 2010;56(1):42–8. [3] Hunt KA, Weber EJ, Showstack JA, Colby DC, Callaham ML. Characteristics of frequent users of emergency departments. Ann Emerg Med 2006;48(1):1–8. [4] South Carolina Public Health Institute. A report on frequent users of hospital emergency departments in South Carolina; 2011. [5] Martin GB, Stokes-Buzzelli SA, Peltzer-Jones JM, Schultz LR. Ten years of frequent users in an urban emergency department. West J Emerg Med 2013;14(3):243–6. [6] Althaus F, Paroz S, Hugli O, Ghali WA, Daeppen JB, Peytremann-Bridevaux I, et al. Effectiveness of interventions targeting frequent users of emergency departments: a systematic review. Ann Emerg Med 2011;58(1):41–52 [e42]. [7] Soril LJ, Leggett LE, Lorenzetti DL, Noseworthy TW, Clement FM. Reducing frequent visits to the emergency department: a systematic review of interventions. PLoS One 2015;10(4):e0123660. [8] Brown RS, Peikes D, Peterson G, Schore J, Razafindrakoto CM. Six features of Medicare coordinated care demonstration programs that cut hospital admissions of high-risk patients. Health Aff (Millwood) 2012;31(6):1156–66. [9] Issel LM. Health program planning and evaluation: a practical, systematic approach for community health. Boston, MA: Jones and Barlett Publishers; 2004. [10] Cutrona SL, Choudhry NK, Fischer MA, Servi AD, Stedman M, Liberman JN, et al. Targeting cardiovascular medication adherence interventions. J Am Pharm Assoc (2003) 2012;52(3):381–97. [11] Durfee J, Johnson T, Batal H, Long J, Rinehart D, Everhart R, et al. The impact of tailored intervention services on charges and mortality for adult super-utilizers. Healthc (Amst) 2018;6(4):253–8. [12] Johnson TL, Rinehart DJ, Durfee J, Brewer D, Batal H, Blum J, et al. For many patients who use large amounts of health care services, the need is intense yet temporary. Health Aff 2015;34(8):1312–9. [13] Krieg C, Hudon C, Chouinard MC, Dufour I. Individual predictors of frequent emergency department use: a scoping review. BMC Health Serv Res 2016;16(1):594. [14] Kanzaria HK, Niedzwiecki MJ, Montoy JC, Raven MC, Hsia RY. Persistent frequent emergency department use: core group exhibits extreme levels of use for more than a decade. Health Aff 2017;36(10):1720–8. [15] Sacco P, Bucholz KK, Spitznagel EL. Alcohol use among older adults in the National Epidemiologic Survey on Alcohol and Related Conditions: a latent class analysis. J Stud Alcohol Drugs 2009;70(6):829–38. [16] Ahn J, McCombs JS, Jung C, Croudace TJ, McDonnell D, Ascher-Svanum H, et al. Classifying patients by antipsychotic adherence patterns using latent class analysis: characteristics of nonadherent groups in the California Medicaid (Medi-Cal) program. Value Health 2008;11(1):48–56. [17] Niu L, Luo D, Silenzio VM, Xiao S, Tian Y. Are informing knowledge and supportive attitude enough for tobacco control? A latent class analysis of cigarette smoking patterns among medical teachers in China. Int J Environ Res Public Health 2015;12(10): 12030–42. [18] Lanza ST, Collins LM, Lemmon DR, Schafer JL. PROC LCA: a SAS procedure for latent class analysis. Struct Equ Model 2007;14(4):671–94. [19] Anson O, Carmel S, Levin M. Gender differences in the utilization of emergency department services. Women Health 1991;17(2):91–104. [20] Fassler J. How doctor's take women's pain less seriously. The Atlantic 2015 [Oct 15]. [21] Billings J, Raven MC. Dispelling an urban legend: frequent emergency department users have substantial burden of disease. Health Aff 2013;32(12):2099–108.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029
6
L.E. Birmingham et al. / American Journal of Emergency Medicine xxx (xxxx) xxx
[22] Smulowitz PB, Honigman L, Landon BE. A novel approach to identifying targets for cost reduction in the emergency department. Ann Emerg Med 2013;61(3):293–300. [23] Birmingham LE, Cochran T, Frey JA, Stiffler KA, Wilber ST. Emergency department use and barriers to wellness: a survey of emergency department frequent users. BMC Emerg Med 2017;17(1):16. [24] Birmingham LE, Nielson JA. An increase in per-patient naloxone requirements in an opioid epidemic. Am J Emerg Med 2017;35(12):1958–9. [25] McCusker J, Roberge D, Ciampi A, Levesque JF, Pineault R, Belzile E, et al. Primary care organization and outcomes of an emergency visit among seniors. Healthc Policy 2009;5(1):e115–31. [26] Haggerty JL, Roberge D, Pineault R, Larouche D, Touati N. Features of primary healthcare clinics associated with patients' utilization of emergency rooms: urbanrural differences. Healthc Policy 2007;3(2):72–85. [27] Kushel MB, Gupta R, Gee L, Haas JS. Housing instability and food insecurity as barriers to health care among low-income Americans. J Gen Intern Med 2006;21 (1):71–7. [28] Bodenmann P, Velonaki VS, Ruggeri O, Hugli O, Burnand B, Wasserfallen JB, et al. Case management for frequent users of the emergency department: study protocol of a randomised controlled trial. BMC Health Serv Res 2014;14:264.
[29] Kangovi S, Barg FK, Carter T, Long JA, Shannon R, Grande D. Understanding why patients of low socioeconomic status prefer hospitals over ambulatory care. Health Aff 2013;32(7):1196–203. [30] Lee NS, Whitman N, Vakharia N, Ph DG, Rothberg MB. High-cost patients: hot-spotters don't explain the half of it. J Gen Intern Med 2017;32(1):28–34. [31] Rinehart DJ, Oronce C, Durfee MJ, Ranby KW, Batal HA, Hanratty R, et al. Identifying subgroups of adult superutilizers in an urban safety-net system using latent class analysis: implications for clinical practice. Med Care 2018;56(1):e1–9. [32] Davis AC, Shen E, Shah NR, Glenn BA, Ponce N, Telesca D, et al. Segmentation of highcost adults in an integrated healthcare system based on empirical clustering of acute and chronic conditions. J Gen Intern Med 2018;33(12):2171–9. [33] Pham JC, Bayram JD, Moss D. Characteristics of frequent users of three hospital emergency departments. Agency for Healthcare Research & Quality: Department of Health & Human Services; 2017. [34] Sarfaty M. Health affairs blog. [Internet]. Available from: Health Affairs. https:// www.healthaffairs.org/do/10.1377/hblog20140103.036069/full/; 2014. [35] Meropol SB, Schiltz NK, Sattar A, Stange KC, Nevar AH, Davey C, et al. Practice-tailored facilitation to improve pediatric preventive care delivery: a randomized trial. Pediatrics 2014;133(6):e1664–75.
Please cite this article as: L.E. Birmingham, V.K. Cheruvu, J.A. Frey, et al., Distinct subgroups of emergency department frequent users: A latent class analysis, American Journal of Emergency Medicine, https://doi.org/10.1016/j.ajem.2019.04.029