Article, Orthopedics

Factors associated with hospital admission for proximal humerus fracture

a b s t r a c t

Background: The number of inpatient admissions for Proximal humerus fracture is increasing, but the factors that determine hospitalization are not well documented. We sought to identify predictors of hospital admission among individuals presenting to the emergency department (ED) with a fracture of the proximal humerus.

Methods: Using the Nationwide Emergency Department Sample for 2010 and 2011, an estimated 285661 pa- tients were identified and separated into those who were admitted to hospital (19%) and those who were discharged directly home (81%). Multivariable logistic regression modeling was used to identify independent predictors of hospital admission.

Results: Factors associated with admission included increasing age and Charlson Comorbidity Index, ED visit on a weekday, Medicare and Medicaid insurance, open fracture, injury due to motor vehicle crash, polytrauma, urban teaching hospital, and residence in the Northeast. The lowest ratio of hospital admission to home discharge was noted for Uninsured patients (0.09).

Discussion: Factors unrelated to medical complexity such as insurance status, geographic region, timing of ED visit, and Hospital type are associated with inpatient admission for proximal Humerus fracture. Interventions to reduce variation in hospital admission and the influence of nonclinical factors merit attention.

Level of evidence: Level II, prognostic study.

Introduction

Although fractures of the proximal humerus are generally managed as an outpatient, many patients are hospitalized after injury [1,2]. The number of inpatient admissions for proximal humerus fracture appears to be on the rise, driven by both an increase in emergency department (ED) visits and a growing tendency to manage these fractures opera- tively, which ultimately results in higher health care resource use [2-6]. There is currently no consensus on whom to admit for inpatient care rather than managing as an outpatient, including outpatient surgery. A recent study in patients presenting to the ED with transient ischemic at- tacks suggested that, apart from clinical factors (eg, age and comorbid- ity), there might as well be nonclinical factors implicated such as insurance status, Household income, and hospital type [7]. In times of

? This work was performed at the Orthopaedic Hand and Upper Extremity Service, Mas- sachusetts General Hospital, Boston, MA, USA.

?? Conflict of interest statement: M.M. and D.R. certify that they had nothing of value re-

lated to this study.

? Ethical review committee statement: No institutional review board approval is man-

datory for this study. The data are deidentified and commercially available for use. The study has been performed in accordance with the ethical standards in the 1964 Declara- tion of Helsinki and has been carried out in accordance with relevant regulations of the

US Health Insurance Portability and Accountability Act.

* Corresponding author. Tel.: +1 352 871 3851; fax: +1 617 726 0460.

E-mail addresses: [email protected] (M.E. Menendez), [email protected] (D. Ring).

intense scrutiny of health care costs, a better understanding of clinical and nonclinical factors affecting the decision to admit a patient to the hospital for proximal humerus fracture can help reduce variation and cost, and perhaps improve the quality, safety, and efficiency of care by informing efforts to better manage a greater number of fractures as outpatients.

Using a large administrative database, this study sought to deter- mine predictors of inpatient admission among individuals who present to US hospital-based EDs with a fracture of the proximal humerus. Specifically, we tested the null hypothesis that there are no factors asso- ciated with hospital admission.

Methods

We conducted this retrospective cross-sectional study using ED Encounter data from the Nationwide Emergency Department Sample (NEDS). The NEDS is operated by the Agency for Healthcare Research and Quality and currently represents the largest all-payer ED database in the United States [8,9]. Unweighted, each data set year contains records on approximately 30 million ED visits from more than 900 hospitals, representing a 20% stratified sample of US hospital-based EDs. Weighted, it estimates roughly 130 million ED visits. Besides collecting patient- and provider-related characteristics, the NEDS uses the International Classification of Diseases, 9th Revision, Clinical Modifica- tion (ICD-9-CM) codes to standardize the reporting of up to 15 diagnoses

http://dx.doi.org/10.1016/j.ajem.2014.10.045

156 M.E. Menendez, D. Ring / American Journal of Emergency Medicine 33 (2015) 155-158

and procedures. Since its inception in 2006, the NEDS database has been increasingly used for comparative health services research [2,10-14]. Formal approval by our institutional review board was not required as

Table 1

Characteristics of patients presenting to the ED with a proximal humerus fracture

Parameter All patients ED disposition P Ratio of

we used anonymous data.

We identified all patients presenting to the ED with an ICD-9-CM

primary diagnosis code of closed (812.00-812.03, 812.09) or open

Home Admitted to hospital

admissions to home discharges

(812.10-812.13, 812.19) fracture of the proximal humerus [2]. Records

Weighted, n (%) 285661 (100) 231095 (81) 54566 (19) 0.24

for patients who were transferred to another hospital or died in the ED were excluded from analysis [7]. Between January 1, 2010, and

Age (y), mean +- SD

Age group (y), %

December 31, 2011, an estimated 285661 patients met the inclusion	b60	36	40	20	b.001 0.12
criteria for our study.	60-79	38	38	37	0.23

62 +- 23 60 +- 24 72 +- 18 b.001 -

Patient-level variables included age (both continuous and cate- gorized into the age groups: b 60, 60-79, and >= 80 years), sex, primary ^S health insurance (Medicare, Medicaid, private, uninsured, and other), and household income of the patient’s zip code of residence C ($1-$38 999, $39 000-$47 999, $48 000-$63999, and >=$64 000).

>=80	26	22	43	0.46
ex (%) Male 28 29 26 b.001 0.22
Female	72	71	74	0.24
harlson comorbidity 0.52 +- 1.2 0.34 +- 0.89 1.3 +- 1.8 b.001 - index, mean +- SD
Insurance status (%) Private 28		30	17	b.001 0.13
Medicare 53		48	72	0.35

Baseline comorbidity status was quantified using the Charlson comor- bidity index [15,16]. We also assessed fracture type (closed and open),

whether a patient was admitted to hospital or was discharged directly to home. We performed bivariate analyses using independent-

Metropolitan teaching

Hospital geographic

trauma type (single trauma and polytrauma), mechanism of injury	Medicaid	8.1	8.8	4.9	0.13
(fall, motor vehicle traffic crash, struck by person or object, and other),	Uninsured	7.1	8.0	3.0	0.09
and whether or not the ED visit occurred on a weekend. The study	Other	4.5	4.7	3.5	0.17
population was aged 62 +- 23 years and predominantly comprised female patients (72%). Most patients sustained an isolated proximal	Median household income (%) $1-$38999	24	24	22	b.001 0.21
humerus fracture (88%), and the most frequent mechanism of injury	$39000-$47999	26	26	26	0.23
was falls (79%; Table 1).	$48000-$63999	25	25	26	0.25
Hospitals were classified according to their location and teaching status (nonmetropolitan, metropolitan nonteaching, and metropolitan	>=$64000 Hospital location and teaching status (%)	25	24	27	0.26
teaching), geographic region (Northeast, Midwest, South, and West),	Nonmetropolitan	18	20	11	b.001 0.13
and trauma level (trauma levels I-III and nontrauma).	Metropolitan	45	46	45	0.23
Our primary outcome of interest was ED disposition, specifically	nonteaching

36 35 44 0.30

samples t test for continuous data and Pearson ?² test for categorical region (%)
data to evaluate the association between each explanatory variable to Northeast		19	18	21	b.001 0.27
ED disposition. Ratios of hospital admissions to home discharges were ^Midwest		23	23	24	0.24
calculated for all explanatory variables. Multivariable logistic regression ^South		38	38	36	0.22
West		20	21	20	0.22
modeling was then used to determine factors independently associated _Trauma _center _(%)
with hospital admission among patients presenting to the ED with a	Nontrauma center	65	67	56	b.001 0.20
proximal humerus fracture. All predictor variables were included simul-	Trauma levels I-III	35	33	46	0.32
taneously in the multivariable regression model [17]. Results were re- ported as odds ratios (ORs) with 95% confidence intervals (CIs). We	Visit on weekend (%) No	69	69	72	b.001 0.25
evaluated the model’s discriminatory performance using the area	Yes	31	32	28	0.21
under the receiver operating characteristic (ROC) curve, which mea-	Fracture type (%)

sured the ability of our model to assign a high probability of admission to those patients who were actually admitted to hospital. Area under curve values range from 0.50 to 1.0, with higher values meaning better discrimination. In general, values less than 0.70 can be considered as poor discrimination, between 0.70 and 0.80 as acceptable, between

0.80 and 0.90 as excellent, and above 0.90 as outstanding [18]. We also assessed model performance using the Nagelkerke pseudo R-square, a measure of the proportion of the variation of hospital ad- mission risk explained by our model. Hence, as to set stricter stan- dards owing to multiple testing and the large weighted sample size, the statistical threshold for ? error was set at .001.

Results

Approximately 1 (19%) in 5 patients presenting to the ED with a frac- ture of the proximal humerus were admitted to hospital (Table 1).

When compared with patients discharged directly home, those admitted to hospital were more likely (P b .001) to be older (72 +- 18 vs 60 +- 24 years), be female (74% vs 71%), be Medicare insured (72% vs 48%), be injured in a motor vehicle crash (5.8% vs 2.8%), to present with an open fracture (1.6% vs 0.4%) or polytrauma (27% vs 8.7%), to reside in higher-income neighborhoods (27% vs 24%) or in the Northeast (21% vs 18%), to have more medical comorbidities (Charlson

Closed	99	100	98	b.001 0.23
Open	0.60	0.40	1.6	0.90

Mechanism of injury (%)

or object

Fall 79		79	79	b.001 0.23
Motor vehicle 3.4		2.8	5.8	0.49
traffic crash Struck by person 2.6		3.0	0.80	0.06
Other 15		15	15	0.23
Trauma type (%) Single trauma	88	91	73	b.001 0.19
Polytrauma	12	8.7	27	0.73

comorbidity index: 1.3 +- 1.8 vs 0.34 +- 0.89), and to visit the ED on a weekday (72% vs 69%). In addition, patients presenting to metropolitan teaching hospital EDs (44% vs 35%) and trauma-level centers (46% vs 33%) were admitted for inpatient care more often.

In multivariable modeling (Table 2), factors independently associat- ed with hospital admission included increasing age and Charlson co- morbidity index (OR, 1.7 per 1-unit increase; 95% CI, 1.7-1.7; P b .001), ED visit on a weekday (OR, 1.1; 95% CI, 1.1-1.1; P b .001), Medicare (OR, 1.6; 95% CI, 1.6-1.7; P b .001) and Medicaid (OR, 1.1; 95% CI, 1.1-

1.2; P b .001) insurance, injury due to motor vehicle crash (OR, 3.5;

M.E. Menendez, D. Ring / American Journal of Emergency Medicine 33 (2015) 155-158 157

Table 2 Multivariable regression modeling: factors associated with hospital admission among pa- tients presenting to the ED with a proximal humerus fracture

95% CI

Predictor Coefficient (?) OR Lower Upper P

human bias and error. Because the NEDS data are de-identified, valida- tion through cross-referencing medical records was not possible. How- ever, validation of the NEDS data is regularly performed by the Agency for Healthcare Research and Quality. Second, the retrospective nature of the NEDS does not allow ascertainment of the exact reasons (clinical

Age (reference: b 60 y) 60-79	0.33	1.4	1.4	1.4	and nonclinical) for which patients were admitted to hospital from the b.001 ED. Third, we were unable to adjust for radiographic severity and time
>=80	0.97	2.6	2.5	2.7	b.001 from injury to ED visit, both of which may influence ED disposition
Female sex (reference: male)	0.0060	1.0	1.0	1.0	.62 [19]. Fourth, another limitation was our inability to account for pa-
Charlson comorbidity index,	0.52	1.7	1.7	1.7	b.001 tient/family preferences and hospital bed availability [20]. Fifth, we
per 1-unit increase Primary health insurance (reference: private)					were unable to compare functional and patient-reported outcomes between patients who were admitted to hospital and those
Medicare	0.47	1.6	1.6	1.7	b.001 discharged directly to home. Sixth, the NEDS does not allow for
Medicaid	0.14	1.1	1.1	1.2	b.001 long-term follow-up, and we were thus unable to evaluate the im-
Uninsured	-0.31	0.73	0.69	0.78	^b.001 pact of ED disposition on outcomes such as mortality, morbidity,
Other	0.26	1.3	1.2	1.4	^b.001 and readmissions. Seventh, as each record in the NEDS represents
Household income (reference: $1-$38999)					a single ED visit and not a patient, it is possible that there are mul-
$39000-$47999	-0.002	1.0	1.0	1.0	.89 tiple records for the same patient with frequent ED visits. Finally,
$48000-$63999	0.0030	1.0	1.0	1.0	.86 the reader should be aware that findings in large-scale studies can
>=$64000	0.028	1.0	1.0	1.1	^.097 be statistically significant yet clinically insignificant.
Hospital location and teaching status

Open fracture

(reference: closed fracture) Mechanism of injury (reference: fall)

1.6 5.1 4.6 5.7 b.001

EDs. Approximately 1 in 5 patients presenting to the ED with proximal humerus fracture were subsequently admitted for inpatient care.

In agreement with the Hip fracture literature [21,22], we found that advanced age was associated with greater odds of hospital ad- mission. Notably, patients 80 years or older were 2.6 times more likely to be admitted compared with patients younger than 60 years. Increasing medical comorbidity was independently associ- ated with risk of hospitalization, which is consistent with previous studies by Neuhaus and colleagues [5,6] indicating that preinjury infirmity plays an important role in health care resource use after

(reference: nonmetropolitan) Metropolitan nonteaching

0.43

1.5

1.6

b.001

In this study of nationally representative data, we estimated the an-

nual number of ED visits for proximal humerus fractures to be 142830,

Metropolitan teaching Hospital geographic region

0.52

1.7

1.6

1.7

b.001

an estimate consistent with a previous analysis by Kim and colleagues

[11] on the epidemiology of humerus fractures in US hospital-based

(reference: West) Northeast	0.21	1.2	1.2	1.3	b.001
Midwest	-0.039	1.0	0.9	1.0	.022
South	0.010	1.0	1.0	1.0	.53
Trauma centers I-III	0.41	1.5	1.5	1.5	b.001
(reference: nontrauma center) Visit on a weekday	0.10	1.1	1.1	1.1	b.001
(reference: weekend visit)

Motor vehicle traffic crash	1.3	3.5	3.3	3.7	b.001
Struck by person or object	-0.75	0.47	0.42	0.52	b.001

Other	0.17	1.2	1.1	1.2	b0.001	proximal humerus fracture. What remains uncertain is the degree
Polytrauma	1.3	3.7	3.6	3.8	b.001	to which hospital admission improves health and function after

(reference: single trauma) Model performance

Area under the ROC curve (95% CI) 0.79 (0.79-0.80)

Nagelkerke R-square 0.26

95% CI, 3.3-3.7; P b .001), the presence of polytrauma (OR, 3.7; 95% CI,

3.6-3.8; P b .001) or open fracture (OR, 5.1; 95% CI, 4.6-5.7; P b .001),

metropolitan nonteaching (OR, 1.5; 95% CI, 1.5-1.6; P b .001) and teach- ing (OR, 1.7; 95% CI, 1.6-1.7; P b .001) facilities, and residence in the Northeast (OR, 1.2; 95% CI, 1.2-1.3; P b .001). The area under the ROC curve derived from the multivariable model predicting hospital admission was 0.79 (95% CI, 0.79-0.80), indicating near-excellent discriminatory ability.

Discussion

A better understanding of factors associated with inpatient man- agement of fracture of the proximal humerus can inform efforts to better manage such patients without hospital admission. Anecdotal evidence suggests that the decision to hospitalize a patient with a proximal humerus fracture is influenced primarily by clinical factors such as patient age and medical complexity, but sometimes, careful study reveals nonclinical factors. We therefore sought to identify clinical and nonclinical factors associated with hospital admission after fracture of the proximal humerus.

Notwithstanding the large sample size and associated power, our study should be interpreted cautiously in light of several limitations in- herent to the analysis of administrative Claims data. First, the NEDS data set is based on billing data from ICD-9-CM codes, which are subject to

proximal humerus fractures in older, infirm patients and whether there are better or more resourceful options.

As expected, the presence of an open fracture, polytrauma, or vi- olent mechanism of injury (eg, motor vehicle crash) was linked to higher rates of hospital admission; however, we also identified sociodemographic disparities in ED disposition after proximal hu- merus fracture, statistically independent of these measures of pa- tient frailty and injury. Compared with privately insured patients, those with Medicare and Medicaid were more likely to be admitted to the hospital. Conversely, patients without insurance coverage were less likely to be hospitalized and more likely to be discharged directly home. On the one hand, the lower likelihood of admission among the uninsured might place them at risk for subOptimal care and prolonged disability [23]. On the other, it is possible that in- sured patients are being admitted unnecessarily. Reasons for the observed variation in admission rates according to insurance status merit further investigation. Among the uninsured, factors that could account for some of the variability include lower levels of health lit- eracy, fear of economic repercussions, and provider bias [24].

Patients presenting to the ED on a weekday were 10% more likely to be hospitalized than those visiting the ED during the weekend. It could be argued that patients presenting on weekdays may be older and sicker patients; however, our analysis controlled for these and other important measures of complexity, so it is likely that there is some differential rec- ommendation of inpatient admission on nonmedical grounds.

Patients in the Northeast were 20% more likely to be admitted to the hospital than those in the rest of the United States, which is in agreement with a recent study in patients presenting to the ED with urolithiasis [10]. It is well known that the Northeast has the highest per capita spending in health care [25], but reasons for the

158 M.E. Menendez, D. Ring / American Journal of Emergency Medicine 33 (2015) 155-158

increased rate of admissions remain incompletely understood. Ad- missions occurred more frequently in urban EDs than in rural EDs. Among urban EDs, hospitalizations were more common in teaching insti- tutions. A plausible explanation might be that urban hospitals-teaching hospitals in particular-are more willing to admit patients because they have more nursing staff available, access to high-technology equipment, and higher hospital bed availability. It might also be that either patient ex- pectations or physician practice habits are different in various regions of the country.

In conclusion, there is variation in hospitalization rates after proxi- mal humerus fracture attributable to factors seemingly unrelated to medical complexity such as insurance status, geographic region, timing of ED visit, and hospital type. Measures for reducing variation in admis- sion rates and the influence of nonclinical factors are merited.

References

AJEM