Factors Associated With Unplanned Admissions Among Patients With Chronic Conditions

ABSTRACT

Objectives: Value-based care models, such as the Medicare Shared Savings Program, have placed increasing emphasis on unplanned admissions among patients with multiple chronic conditions (UAMCCs) as a quality metric. However, there are limited data on which factors are associated with the highest risk of UAMCCs. This study sought to determine which factors were associated with increased risk of UAMCCs.

Study Design: Retrospective study conducted among all adult patients with 2 or more chronic conditions defined by CMS presenting to 2 hospitals within a major Midwest health care system from November 1, 2022, to October 31, 2023.

Methods: Demographics, chronic conditions, primary care physician (PCP) visit utilization, and annual wellness visit (AWV) utilization were analyzed using multivariable logistic regression to identify associations with UAMCCs.

Results: Among 18,448 patients (55.8% women) included in the study, 3842 (20.8%) had at least 1 UAMCC. Patients with UAMCCs were more likely to have Medicare or Medicaid insurance; be widowed; speak Spanish; have a higher Charlson Comorbidity Index score; and have Alzheimer disease, atrial fibrillation, heart failure, chronic kidney disease, depression, chronic obstructive pulmonary disease, and/or stroke. When examining PCP visit measures, AWVs and having 1 or more PCP visits were associated with fewer UAMCCs.

Conclusions: Higher Charlson Comorbidity Index scores, several health conditions, and Spanish language were associated with increased UAMCCs. AWVs and having 1 or more PCP visits were associated with fewer UAMCCs.

Am J Manag Care. 2025;31(Spec. No. 6):SP312-SP321

Accountable care organizations (ACOs) were designed to align health care efforts to improve the quality of care while limiting the growth in health care costs.¹ As part of the Medicare Shared Savings Program, institutions focus on improving several key health care metrics, including quality measures and health care spending. Private insurance policy often follows CMS,² emphasizing the broader implications of these metrics beyond CMS.

Patients with multiple chronic illnesses have a higher likelihood of hospitalization, providing an ideal group for focused interventions to improve clinical outcomes and reduce unnecessary health care utilization and overall health care costs.^3-5 As such, CMS has identified unplanned admissions among patients with multiple chronic conditions (UAMCCs) as an important quality metric.⁶ Despite the importance, there remains relatively limited research on factors associated with UAMCC to inform focused and actionable interventions. This study sought to determine which patient-specific factors were associated with an increased risk of UAMCCs.

METHODS

This was a retrospective, multicenter, case-control study assessing factors associated with UAMCCs. This study was deemed exempt by the Rush University Institutional Review Board. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology guideline.⁷

Data Collection and Participants

Data were collected via electronic health records (EHRs; Epic Systems, Inc) at 2 hospitals in the Midwest from November 1, 2022, to October 31, 2023. Rush University Medical Center is a 664-bed urban, tertiary care hospital in Chicago, Illinois, with an annual emergency department volume of approximately 70,000 patients. Rush Oak Park Hospital is an 87-bed community hospital in Oak Park, Illinois, with an annual emergency department volume of approximately 50,000 patients. Both hospitals are part of ACOs.

Adult patients (aged ≥18 years) with at least 2 CMS-defined chronic conditions were included in the study (eAppendix 1 [eAppendices available at ajmc.com]).⁶ We included patients who had commercial, Medicare, or Medicaid insurance coverage. We collected the following key demographics: age, sex, race, ethnicity, primary language, marital status, insurance type, and chronic conditions. These chronic conditions were the same 9 used in the UAMCC algorithm (ie, acute myocardial infarction, atrial fibrillation, Alzheimer disease, chronic kidney disease, chronic obstructive pulmonary disease [COPD], depression, diabetes, heart failure, stroke). Briefly, these conditions are measured based on International Statistical Classification of Diseases, Tenth Revision diagnoses from patient encounters during a 3-year period before the year of UAMCC measurement. The definitions are complex and vary by medical condition (details are available from CMS⁸). The Charlson Comorbidity Index (CCI) score, representing patient comorbidity among 17 disease conditions, was used as a general measure of medical complexity.⁹ Poverty level was defined as the percentage of the population in the patient’s zip code that was below poverty level as estimated in the US Census Bureau’s American Community Survey for 2022.¹⁰ In addition, we collected 2 measures of primary care physician (PCP) utilization during the study interval: (1) number of total outpatient PCP encounters and (2) annual wellness visits (AWVs). AWVs are structured, in-depth PCP visits designed specifically for Medicare patients, whereas other PCP visits can be done with any patient regardless of insurance type. A patient can have only 1 AWV but several PCP visits during a given year. Therefore, AWVs were coded as dichotomous, whereas PCP visits were modeled as a count (0, 1, 2, 3, or ≥ 4). Patient characteristics were either time invariant (eg, patient demographics) or coded during the time period before the measurement year of UAMCC. For example, chronic conditions were coded using data from November 1, 2019, to October 31, 2022, following chronic condition definitions specified by CMS. PCP visit count and AWV were coded during the year prior to the UAMCC to estimate the association of PCP utilization with UAMCCs (ie, PCP utilization measured in 11/1/2021-10/31/2022 and UAMCCs measured in 11/1/2022-10/31/2023).

Outcomes

The objective was to identify which factors were associated with an increased risk of UAMCCs (the primary outcome).⁶ UAMCC was defined as a patient with at least 2 CMS-defined chronic conditions who had an unplanned inpatient admission between November 1, 2022, and October 31, 2023. As it was relatively rare for a given patient to have more than 1 admission, the outcome was dichotomized as at least 1 UAMCC vs no UAMCC. This population was then compared against all patients with at least 2 CMS-defined chronic conditions within the entire Rush University health care system (inclusive of both sites) who had no unplanned admissions. Modeling focused on the association of PCP utilization measures with unplanned admissions after statistically adjusting for other patient characteristics. These were age, sex, race, ethnicity, primary language, marital status, insurance type, poverty level, CCI score, and binary indicators of having each of the 9 CMS-defined chronic conditions.

Statistical Analysis

It was hypothesized that higher levels of PCP utilization (represented by AWV as well as count of PCP visits) would be associated with decreased risk of having unplanned admissions. However, it was speculated that the effect of PCP visits might vary by the medical complexity of patients (eg, a greater number of PCP visits may have a stronger association with reduced likelihood of admissions among more medically complex patients). Only Medicare patients can have AWVs, with a maximum of 1 annually. In the models, AWV was coded as 1 if they had an AWV vs 0 if they did not have an AWV, and Medicare was coded as 1 if they had Medicare insurance vs 0 if they did not have Medicare insurance; patients without Medicare insurance (eg, younger adults) were coded as 0 for both AWV and Medicare. Thus, the estimated AWV effect is applicable to Medicare patients only. Specifically, in a logistic regression model with all other predictors held constant at 0, the estimated probability of having a UAMCC is e^α/(1 + e^α), where α = β₀ (intercept term) + β₁ (AWV coefficient) + β₂ (Medicare coefficient) for Medicare patients who had an AWV; α = β₀ + β₂ for Medicare patients who did not have an AWV; and α = β₀ for non-Medicare patients (the estimates can be computed for any set of patient characteristics from the coefficients shown in Table 1, and odds can be computed as e^α).

A logistic regression model estimating main effects only (no interactions) was used to estimate the probability of having a UAMCC (coded as 1 if they had a UAMCC vs 0 if they did not have a UAMCC), based on PCP utilization measures (AWV and count of PCP visits), after statistically adjusting for the other patient characteristics described earlier (see Table 1 for more information about the coding of these variables used in the models). Interactions between PCP utilization and all of the other patient characteristics in the main-effects logistic regression were explored in a second logistic regression model, but few interactions were significant, and those that were did not suggest meaningful interpretations, nor did they improve model fit noticeably (area under the receiver operating characteristic curve = 0.759 in the main-effects-only model vs 0.762 in the model including interactions), therefore these analyses were dropped. Model diagnostics indicated reasonably low levels of collinearity among the patient characteristics (variance inflation factors < 5). A small number of observations (n = 219; < 2%) were excluded due to missing poverty level. Because so few data were excluded, no special techniques were used to handle missing data (the models excluded the 219 patients with missing poverty level data).¹¹

RESULTS

There were 18,448 patients included in the study, of whom 3842 (20.8%) had at least 1 unplanned admission (eAppendix 2). Patient characteristics are described in Table 2. Table 1 reports the association of patient characteristics with UAMCC in the logistic regression model estimating main effects only. As shown in Table 1, after adjusting for the other variables in the model, patients with diabetes were less likely to have a UAMCC, whereas those with Spanish language; Medicare or Medicaid insurance; widowed marital status; a history of Alzheimer disease, atrial fibrillation, chronic kidney disease, COPD, depression, heart failure, or stroke; or a higher CCI score were more likely to have a UAMCC. As an example of interpretation of the ORs in Table 1, holding the other patient characteristics constant, patients with diabetes had 0.71 times the odds of having a UAMCC compared with patients without diabetes (ie, they were less likely to have a UAMCC), whereas patients with a history of depression had 1.52 times the odds of having a UAMCC compared with patients who did not have a history of depression (ie, they were more likely to have a UAMCC).

When examining PCP utilization measures, AWVs and count of PCP visits were both significantly associated with UAMCCs (Table 1). Patients with an AWV were significantly less likely to have a UAMCC (OR, 0.88; 95% CI, 0.78-1.00; P = .0432) compared with those having no AWV. Patients with 1 or more PCP visits were also significantly less likely to have a UAMCC than those with no PCP visits (OR range, 0.64-0.72; P < .0001).

DISCUSSION

This study identified several key factors associated with increased risk of UAMCCs, which can inform efforts by institutions. A significant association was seen with PCP visits and AWVs. Having 1 or more PCP visits was associated with a lower risk of UAMCCs, highlighting the important role that outpatient PCP access can have in addressing chronic conditions before patients require hospital admission. AWVs had a weaker association but were also associated with lower risk of UAMCCs, emphasizing the potential importance of comprehensively addressing patient needs beyond their medical conditions (eg, assessing home safety, fall risk, activities of daily living). Additionally, we found that the Medicare and Medicaid populations were at higher risk of UAMCCs. Because AWVs are a function of Medicare, this may help to mitigate that risk. Further, we found that patients with higher overall CCI scores had higher risk, highlighting this population as one to which greater efforts should be dedicated.

Another important finding was a higher rate of UAMCC among those who spoke Spanish as a primary language compared with other languages. This may reflect greater difficulty with obtaining adequate outpatient access to a PCP who speaks Spanish. Alternatively, the disproportionate rate seen in Spanish vs other non-English languages may reflect disparities in use of interpreters leading to inaccurate translations among health care providers with more limited Spanish language proficiency.^12,13

Limitations

This study was limited to 2 institutions and may not fully reflect the broader population. Additionally, we relied upon EHR data and were unable to capture UAMCCs at institutions outside our system. However, EHR data offered the benefit of a wider range of patients (regardless of insurance type) and less latency than administrative claims data. This is particularly important because many CMS measures are later adopted by insurance agencies, so it is important to account for this broader population. Moreover, our study was limited to observational data, and future work should assess the impact of focused interventions to reduce UAMCCs based on these data. Although having an AWV was associated with reduced rates of UAMCC in our study, it is unclear whether patients who participate in AWVs may reflect a cohort more willing to engage in preventive screening and proactive health assessment. However, prior research has found no differences when attempting to adjust for this.^14,15 Finally, we used zip code as a surrogate for poverty level, which may not fully reflect individual income levels.

CONCLUSIONS

This study provides information on factors influencing UAMCCs, emphasizing the potential role of AWVs and PCP visits and identifying which populations are most likely to benefit in order to best steward limited resources.

Author Affiliations: Department of Emergency Medicine, Rush University Medical Center (MG, PC, GW), Chicago, IL; Department of Medicine, Rush University Medical Center (BDS), Chicago, IL; Rush Health (NG, DT), Chicago, IL.

Source of Funding: None.

Author Disclosures: Dr Gottlieb reports grant funding from the CDC, National Institutes of Health, GE Healthcare, Bill & Melinda Gates Foundation, Society for Academic Emergency Medicine Foundation, and the Charles J. and Margaret Roberts Fund unrelated to this work. The remaining authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (MG, NG, BDS, PC, GW, DT); acquisition of data (MG); analysis and interpretation of data (MG, NG, BDS, PC, GW, DT); drafting of the manuscript (MG, NG, BDS, PC, GW, DT); critical revision of the manuscript for important intellectual content (MG, BDS, PC, GW); statistical analysis (MG, NG, BDS, GW, DT); administrative, technical, or logistic support (MG); and supervision (DT).

Send Correspondence to: Michael Gottlieb, MD, Department of Emergency Medicine, Rush University Medical Center, 1750 W Harrison St, Ste 108 Kellogg, Chicago, IL 60612. Email: MichaelGottliebMD@gmail.com.

REFERENCES

1. Report to Congress: National Strategy for Quality Improvement in Health Care. CMS. March 2011. Accessed July 9, 2024. https://www.cms.gov/cciio/resources/forms-reports-and-other-resources/quality03212011a#es
2. Clemens J, Gottlieb JD. In the shadow of a giant: Medicare’s influence on private physician payments. J Polit Econ. 2017;125(1):1-39. doi:10.1086/689772
3. Zhang NJ, Wan TTH, Rossiter LF, Murawski MM, Patel UB. Evaluation of chronic disease management on outcomes and cost of care for Medicaid beneficiaries. Health Policy. 2008;86(2-3):345-354. doi:10.1016/j.healthpol.2007.11.011
4. Dantas I, Santana R, Sarmento J, Aguiar P. The impact of multiple chronic diseases on hospitalizations for ambulatory care sensitive conditions. BMC Health Serv Res. 2016;16(a):348. doi:10.1186/s12913-016-1584-2
5. Rodrigues LP, de Oliveira Rezende AT, Delpino FM, et al. Association between multimorbidity and hospitalization in older adults: systematic review and meta-analysis. Age Ageing. 2022;51(7):afac155. doi:10.1093/ageing/afac155
6. ACO #38 risk-standardized acute admission rates for patients with multiple chronic conditions. CMS. Accessed May 7, 2025. https://www.cms.gov/files/document/test-february-11th.pdf
7. Vandenbroucke JP, von Elm E, Altman DG, et al; STROBE Initiative. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. PLoS Med. 2007;4(10):e297. doi:10.1371/journal.pmed.0040297
8. Quality Payment Program: resource library. CMS. Accessed May 7, 2025. https://qpp.cms.gov/resources/resource-library
9. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8
10. 2022: ACS 5-year estimates subject tables: S1702. US Census Bureau. Accessed February 13, 2025. https://data.census.gov/table/ACSST5Y2023.S1702?g=010XX00US$8600000
11. Harrell FE Jr. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. Springer; 2001.
12. Nápoles AM, Santoyo-Olsson J, Karliner LS, Gregorich SE, Pérez-Stable EJ. Inaccurate language interpretation and its clinical significance in the medical encounters of Spanish-speaking Latinos. Med Care. 2015;53(11):940-947. doi:10.1097/MLR.0000000000000422
13. Rosenthal A, Wang F, Schillinger D, Pérez Stable EJ, Fernandez A. Accuracy of physician self-report of Spanish language proficiency. J Immigr Minor Health. 2011;13(2):239-243. doi:10.1007/s10903-010-9320-1
14. Beckman AL, Becerra AZ, Marcus A, et al. Medicare annual wellness visit association with healthcare quality and costs. Am J Manag Care. 2019;25(3):e76-e82.
15. Misra A, Lloyd JT. Hospital utilization and expenditures among a nationally representative sample of Medicare fee-for-service beneficiaries 2 years after receipt of an annual wellness visit. Prev Med. 2019;129:105850. doi:10.1016/j.ypmed.2019.105850