A predictive model for increased hospital length of stay following geriatric hip fracture

Abstract

Background

The purpose of this study was to identify the risk factors that are significantly associated with hospital length of stay (LOS) following geriatric hip fracture and to use these significant variables to develop a LOS calculator.

Materials and methods

This was a retrospective study examining 614 patients treated for geriatric hip fracture between January 2000 and December 2009 at an urban, Level 1 trauma center. A negative binomial regression analysis was used to identify perioperative variables associated with hospital LOS.

Results

614 patients met the inclusion criteria, presenting with a mean age of 78 (±10) years. The most common pre-operative comorbidity was hypertension, followed by diabetes and COPD. After controlling for all collected comorbidities as well as demographics and operative variables, hypertension (IRR: 1.10, p = 0.029) and disseminated cancer (IRR: 1.24, p = 0.007) were found to be significantly associated with LOS. In addition, two demographic/presenting variables, admission to the medicine service (IRR: 1.48, p < 0.001) and male sex (IRR: 1.09, p = 0.034), were shown to be independent risk factors for prolonged LOS. These variables were synthesized into a LOS formula, which estimated LOS to within 3 days of the true length of stay for 0.758 of the series (95% confidence interval: 0.661 to 0.855).

Conclusions

This study identified several comorbidity and perioperative variables that were significantly associated with LOS following geriatric hip fracture surgery. The resulting LOS model may have utility in the risk stratification of orthopaedic trauma patients presenting with hip fracture.

1. Introduction

Hip fractures continue to remain a significant source of morbidity and high economic cost among the elderly population in the United States, with over 250,000 patients hospitalized each year.¹ Indeed, over 60% of the 13.8 billion dollars associated with osteoporotic fractures were attributed to hip fracture care.² In addition, as the geriatric population of the US continues to rapidly grow, the incidence and costs associated with hip fractures will similarly increase.³

Hospital length of stay (LOS) is known to be a significant driver of the costs of inpatient hip fracture care. A variety of studies have demonstrated a significant association between LOS and the costs of hip fracture treatment, while reductions in LOS were associated with substantial savings.^{4, 5, 6} As emerging systems of healthcare reimbursement begin to encompass outcome metrics and shift the costs of additional care onto providers, it is becoming particularly important to identify and thoroughly assess the factors responsible for prolonged LOS.^{7, 8, 9}

Despite this, while several studies have attempted to elucidate variables contributing to greater LOS in their hip fracture cohorts, few have attempted to synthesize these into a model to predict LOS. Existing investigations have identified factors such as admission service, ASA score, sex, and certain pre-operative workups that contribute to increased LOS.^{10, 11, 12, 13} Other studies have utilized large, national databases to explore the relationship between perioperative factors and LOS, with the eventual goal of employing these variables to risk stratify hip fracture patients. While these studies have identified factors such as anesthesia type and procedure classification, it is currently unclear whether these relationships can be extrapolated to individual centers with large volumes of hip fracture patients or be used in stratification.¹⁴ In addition, there is a dearth of research aimed at synthesizing various risk factors into a model that could be utilized to risk stratify patients based on predicted LOS. The purpose of our study was thus to identify perioperative variables associated with prolonged LOS following hip fracture surgery at a large level 1 trauma center and synthesize significant predictors into a LOS calculator. By creating a model incorporating variables associated with greater LOS, it may be possible to risk stratify hip fracture patients and subsequently streamline inpatient resource utilization.

2. Materials and methods

2.1. Data collection

Current procedural terminology (CPT) codes were used to identify all patients who were admitted to our level 1 trauma center for a low-energy hip fracture between January 2000 and December 2009. Inclusion criteria consisted of patients over the age of 60 who underwent operative fixation as well as individuals treated via a hemiarthroplasty or total hip arthroplasty (THA) following a hip fracture (including both intertrochanteric and femoral neck fractures). Patients undergoing closed reduction and percutaneous pinning, cephalomedullary nailing, THA, hemiarthroplasty, or open reduction and internal fixation (ORIF) were thus identified via the CPT codes: 27130, 27130A, 27235, 27236, 27244, 27245, 27248, 27254, 27506, 27507, and 27509. This study was approved by the Institutional Review Board (IRB) of our institution. For this type of study formal consent is not required.

Retrospective chart review of patient records was conducted in order to document demographics, pre-operative comorbidities, and perioperative characteristics for each patient meeting inclusion criteria. Demographic variables included age, race, sex, American Society of Anesthesiologists (ASA) classification, cigarette use, and alcohol use. Pre-operative comorbidities consisted of diabetes, obesity, hypertension, myocardial infarction (MI), congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), liver disease, disseminated cancer, paralysis, peripheral vascular disease (PVD), pulmonary circulation disorders (PCD), thyroid disease, significant weight loss in the 6 months preceding surgery, and renal failure. Perioperative characteristics included the type of surgical intervention and admitting service (medicine or orthopaedics). The primary outcome variable of interest was hospital LOS, defined as the time between preliminary admission and subsequent discharge.

2.2. Statistical analysis

In order to identify the factors that were significantly associated with prolonged LOS, a negative binomial regression model (NBM) was constructed, incorporating all of the pre-operative comorbidities, demographics, and operative characteristics, with LOS as the dependent variable of interest. A NBM is utilized when assessing overdispersed outcome variables, such as LOS, and has been successfully employed in prior studies analyzing LOS following orthopaedic surgery.^11,15 To check that the data set was overdispersed, a Poisson regression model was also constructed. In comparing the negative binomial and Poisson regression models, it was observed that the data was significantly overdispersed (P = 8 × 10⁻³⁰). Therefore, data analysis proceeded utilizing the negative binomial regression model.

Beta (β) variable coefficients were computed for each perioperative factor while controlling for comorbidities and surgical characteristics. Backwards, stepwise elimination was subsequently used to identify significant predictors of LOS following orthopaedic trauma. Incidence rate ratios (IRR) were calculated to quantify the difference in length of stay associated with significant predictors.

These significant factors were then synthesized into a LOS calculator via which inclusion of comorbidity and admission data would yield the predicted patient LOS. In order to assess the predictive capacity of this model, we computed the mean percent error in the LOS estimates for the entire series. A priori statistical significance was established at P < 0.05. All analyses were performed using IBM SPSS Statistics 22 (IBM, Armonk, NY).

3. Results

3.1. Case series characteristics

A total of 614 hip fracture patients meeting inclusion criteria were incorporated into the analysis. Of these, 202 were male and 553 presented with an ASA score of 3 or 4. The mean age of the patients in the series was 78 (±10) and the mean BMI was 25 (±6). The most common pre-operative comorbidities were hypertension, present in 403 (65.6%) individuals, diabetes, seen in 146 (23.8%) people, and COPD, present in 138 (22.5%) patients. The most performed intervention was ORIF, which was undergone by 235 (38.3%) hip fracture patients, followed by arthroplasty, performed on 194 (31.6%) individuals in the series. The demographic, comorbidity, and perioperative characteristics for the patients in the series are presented in Table 1.

Table 1.

Case series characteristics.

Demographics	N = 614
Age (mean ± SD)	78 ± 10 years
Female	51%
American Society of Anesthesiologists Score 3-4	553 (90.1%)
Body Mass Index (mean ± SD)	25 ± 6
Select Comorbidities
Hypertension	403 (65.6%)
Diabetes	146 (23.8%)
COPD	138 (22.5%)
Congestive Heart Failure	110 (17.9%)
Thyroid Disease	102 (16.6%)
Renal Failure	61 (9.9%)
Peripheral Vascular Disease	44 (7.2%)
Intervention Type
Open Reduction and Internal Fixation	235 (38.3%)
Arthroplasty	194 (31.6%)
Intramedullary Nail	82 (13.4%)
Percutaneous	61 (9.9%)
Other	42 (6.8%)
Length of Stay (LOS)
LOS (mean ± SD)	6.92 ± 4.44

3.2. LOS distribution

The mean hospital LOS for the hip fracture cohort was 6.92 (±) 4.44 days. The median LOS was 6.0 days, with a range of 28 (minimum of 1 and a maximum of 29). A histogram depicting the distribution of LOS among the hip fracture patients in the series is highlighted in Fig. 1.

Fig. 1.

A histogram depicting the distribution of LOS among the entire geriatric hip fracture series.

3.3. Significant predictors of LOS

Two comorbidities including hypertension (IRR: 1.10, p = 0.029) and disseminated cancer (IRR: 1.24, p = 0.007) were significantly associated with LOS (see Fig. 2). In addition, two demographic/presenting variables, admission to the medicine service (IRR: 1.48, p < 0.001) and male sex (IRR: 1.09, p = 0.034) were independent risk factors for prolonged LOS. The coefficients and IRR values associated with each significantly predictive variable are presented in Table 2.

Fig. 2.

A flowchart highlighting the transition of beta value coefficients of significant risk factors to IRRs and the LOS calculator formula components.

Table 2.

Factors significantly associated with LOS.

Risk Factor	β coefficient	P-value	IRR
Intercept	1.301	0.05	–
Admitted to Medicine	0.394	<0.001	1.48
Male Sex	0.094	0.034	1.09
Hypertension	0.098	0.029	1.10
Cancer	0.218	0.007	1.24

These significant variables were incorporated into a LOS calculator by employing the beta coefficients associated with each risk factor. The calculator was designed based on the formula ${\text{e}}^{(intercept+{\beta }_1{X}_1+{\beta }_2{X}_2+\dots )}$, incorporating the β variable coefficients from the multivariable regression modeling. In order to illustrate the practical utility of our model, data from one patient in our series was randomly selected for inclusion into the model. This patient, an 86-year-old female admitted to the orthopaedic service and presenting with an ASA score of 4 along with a history of COPD and smoking, would be predicted to stay 3.78 days using our formula. Her actual LOS was 4.0 days. When utilizing data from the entire series, our model was found to result in a mean 0.5% underestimation of the hospital LOS for the patient population. In this assessment, the median proportion of LOS data estimated to within 3 days of the true length of stay was 0.758, with a 95% confidence interval ranging from 0.661 to 0.855.

4. Discussion

Trauma-induced hip fractures result in significant morbidity that has been associated with prolonged LOS, contributing to the rapidly rising costs of care. Although several studies have identified risk factors associated with greater LOS, these variables are often reported in isolation and not used in the context of risk stratification based on LOS. These data, derived from a large series of geriatric hip fracture patients presenting to a level 1 trauma center, reveal admission to medicine, male sex, hypertension, and cancer to be significantly associated with LOS. In addition, these risk factors were synthesized into a LOS calculator in the form of a model that underestimated true hospital LOS for the entire series by 0.5%.

Several studies have attempted to elucidate the causes of LOS in their respective hip fracture cohorts. Overall, the literature assessing LOS following hip fracture surgery remains heterogeneous and conflicting regarding significant risk factors. For instance, several studies have identified age as a demographic variable that is significantly associated with overall LOS,^{13,16, 17, 18} while others have concluded that age plays no role.^11,12,19,20 In addition, cumulative comorbidities have been found to be both significantly associated and not significant in predicting LOS.^12,18 The data from this study demonstrated age to not be significantly associated with increased LOS, while comorbidities such as hypertension and cancer were found to be significantly predictive. To the authors’ knowledge, this is the first study highlighting hypertension as a significant predictor of increased LOS following hip fracture. However, it has been reported to be associated with LOS in other, non-orthopaedic surgical cohorts.²¹

In examining the relationship between demographic characteristics and LOS, our data showed a significant association between male sex and increased LOS following hip fracture surgery. This effect has been corroborated in the existing literature. In a study assessing the factors affecting delay to surgery and LOS in patients with hip fractures, Ricci et al. found that male patients had a significantly greater LOS (9.8 days compared to 7.3 days).¹⁰ In a database study utilizing the American College of Surgeons National Quality Improvement Program (ACS-NSQIP), Basques et al. also identified male sex as being significantly associated with increased LOS.¹⁴ While these studies sought to determine risk factors significantly associated with greater LOS following hip fracture, our study aimed to synthesize significant elements into a predictive formula that could be utilized for risk stratification. By combining the significant results of the NBM, this formula was able to accurately predict hospital LOS to within 3 days for over 75% of geriatric hip fracture patients presenting at our institution over the course of 10 years. Further assessment of this model in the context of other patient populations will be important in refining its potential utility for risk stratification in orthopaedic trauma.

These results need to be assessed in the context of the limitations of this study design. One limitation of this study is that there may be poor predictive value with a hospital stay longer than 10 days. Another is the retrospective nature of the investigation and is thus at inherent risk for selection bias. While we attempted to mitigate this by incorporating every case performed at our center over the course of 10 years, it is not possible to account for every confounder in an observational study design. Second, it was difficult to assess the severity of each comorbidity via retrospective chart review, resulting in conditions such as diabetes and hypertension to be analyzed as single entities, regardless of the degree of management. Finally, the predictive capacity of this LOS calculator was assessed using our 10-year hip fracture series. Further research is needed to evaluate the predictive performance of this model either utilizing large series at other institutions or in the context of a prospective clinical trial.

5. Conclusion

This study used data from over 600 patients treated for geriatric hip fracture over the course of 10 years at a single level one trauma center. The data demonstrated hypertension and disseminated cancer to be pre-operative comorbidities that were significantly associated with hospital LOS. Male sex and admission to the medicine service were found to be significant predictors of greater LOS following hip fracture. These factors were incorporated into a LOS calculator that was shown to be effective at predicting LOS for a large series of hip fracture patients.

Compliance with ethical standards

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. For this type of study formal consent is not required.

Conflicts of interest

The authors declare that they have no conflict of interest.

Funding

There was no funding source for this study.