A Simple Emergency Department Based Score Predicts Complex Hospitalization in Patients with Inflammatory Bowel Disease

Abstract

Background and Aims:

Thirty percent of inflammatory bowel disease (IBD) patients hospitalized with flare require salvage therapy or surgery. Additionally, 40% experience length of stay (LOS) > 7 days. No emergency department (ED)-based indices exist to predict these adverse outcomes at admission for IBD flare. We examined if clinical, laboratory, and endoscopic markers at presentation predicted prolonged LOS, inpatient colectomy, or salvage therapy in IBD patients admitted with flare.

Methods:

Patients with ulcerative colitis (UC) or colonic involvement of Crohn’s disease (CD) hospitalized with flare and tested for Clostridioides difficile infection (CDI) between 2010 and 2020 at two urban academic centers were studied. The primary outcome was complex hospitalization, defined as: LOS > 7 days, inpatient colectomy, or inpatient infliximab or cyclosporine. A nested k-fold cross-validation identified predictive factors of complex hospitalization.

Results:

Of 164 IBD admissions, 34% (56) were complex. Predictive factors included: tachycardia in ED triage (odds ratio [OR] 3.35; confidence interval [CI] 1.79–4.91), hypotension in ED triage (3.45; 1.79–5.11), hypoalbuminemia at presentation (2.54; 1.15–3.93), CDI (2.62; 1.02–4.22), and endoscopic colitis (4.75; 1.75–5.15). An ED presentation score utilizing tachycardia and hypoalbuminemia predicted complex hospitalization (area under curve 0.744; CI 0.671–0.816). Forty-four of 48 (91.7%) patients with a presentation score of 0 (heart rate < 99 and albumin >=3.4 g/dL) had noncomplex hospitalization.

Conclusions:

Over 90% of IBD patients hospitalized with flare with an ED presentation score of 0 did not require salvage therapy, inpatient colectomy, or experience prolonged LOS. A simple ED-based score may provide prognosis at a juncture of uncertainty in patient care.

Introduction

Hospitalization for flare of inflammatory bowel disease (IBD) is associated with significant morbidity and high resource utilization [1,2]. Though corticosteroids are the mainstay of acute treatment, nearly 40% of patients will experience a worsening clinical course and require surgery or salvage therapy with either infliximab or cyclosporine [3–6]. Not only are hospitalization rates for IBD flare rising at a rate of 4.3% annually [7–9], rates of in-hospitalization surgery are rising in those with severe disease [9]. Though biologics help avoid urgent surgery, they may not mitigate long-term surgical risk [10], indicating the higher morbidity surrounding patients who require inpatient second-line therapy. Therefore, early identification of patients who will require rescue therapy or surgery during admission is critical to directly improving IBD care.

Several scores utilize a combination of clinical, biochemical, and/or endoscopic data to predict those patients whose clinical course will worsen [11–19]. Truelove and Witts in the 1950s employed frequency of bloody stools, elevated inflammatory markers, fever, anemia, and tachycardia to define severe ulcerative colitis (UC) [11]. Recently, the ACE (Albumin, C-Reactive Protein [CRP], and Endoscopy) score developed by Grant, et. al. used albumin, CRP, and the endoscopic Mayo score [20] to predict which UC patients would require second line therapy or colectomy [19]. Existing scores, however, are limited to patients with UC, do not account for Clostridioides difficile infection (CDI), and do not predict prolonged length of stay (LOS).

Objective

The objective of this study was to create a simple, emergency department-based index to predict complex hospitalization, defined as: a) prolonged length of stay, b) inpatient tumor necrosis factor (TNF) inhibitor or cyclosporine, or c) inpatient colectomy in all IBD patients admitted with flare symptoms.

Methods

Selection Criteria

All patients admitted through the ED for IBD flare and tested for CDI were considered for inclusion in this retrospective dual-center study. Only those tested for CDI were considered in order to examine CDI as a risk factor for complex hospitalization. IBD diagnosis (UC or Crohn’s disease [CD]) was confirmed by International Statistical Classification of Diseases and Related Health Problems 10^th revision and by manual review of gastroenterology consultant documentation during admission. Admissions between January 2010 to September 2019 at New York-Presbyterian Columbia University Irving Medical Center, and between January 2013 to June 2020 at NYU Langone Health were considered. Inclusion criteria were limited to the following: age over 18 at admission, index hospitalization for IBD flare, hospitalization for at least 48 hours, and presentation with clinical symptoms, laboratory, imaging, or endoscopy findings consistent with IBD flare. Only hospitalizations of at least 48 hours in duration were considered in order to exclude observation visits. Patients with ileocolonic and colonic Crohn’s disease were included, but Crohn’s patients with isolated ileal, small bowel, or perianal disease were excluded. Additionally, Crohn’s patients admitted with small bowel obstructions, fistulizing disease, or intra-abdominal abscesses were excluded. Other exclusion criteria were incomplete documentation and admission for another primary diagnosis or for elective surgical procedures. Patient selection occurred prior to the extraction of exposure variables from all considered charts.

Data Collection

Demographic data and laboratory values were collected by manual chart review; all laboratory values were recorded from the first blood draw performed within the ED. Hemodynamics at presentation were recorded from vital signs in ED triage. It was noted whether colitis was read by a radiologist on in-hospital axial imaging. It was also noted whether endoscopy performed during admission demonstrated colitis, defined as a Mayo endoscopic subscore of 1 or greater for patients with UC [20] and the presence of colitis for patients with CD.

Data on IBD-related outcomes that were collected included: LOS, new inpatient TNF inhibitor or cyclosporine, and inpatient colectomy for refractory colitis. Any new IBD medication initiated while inpatient was also recorded: steroids, 5-aminosalicylates, immunomodulators (6-mercaptopurine, azathioprine, methotrexate), TNF inhibitors, anti-interleukin (IL)-12/IL-23, anti-integrin, cyclosporine, and Janus kinase inhibitor.

Outcomes and Definitions

The primary outcome of this study was complex hospitalization, defined as a composite of the following outcomes: 1) prolonged LOS (7 days or longer), 2) inpatient colectomy, or 3) requiring either inpatient TNF inhibitor or cyclosporine. Each component of the primary outcome was analyzed separately as a secondary outcome. ACE score at ED presentation (1 point each for albumin <=3.0 g/dL and CRP>=50 mg/L) and during hospitalization (an additional 1 point for colitis on endoscopy) was also calculated for all patients [19].

Statistical Analysis

Continuous variables are presented as medians with interquartile ranges (IQR). Categorical variables are presented as frequency using percentages. Regression-derived statistics are presented using odds ratio (OR) and confidence intervals (CI).

A nested k-fold cross-validation approach was used for the final data analysis to allow for simultaneous validation of the study findings. In the internal cross-validation, lasso regression was performed on a subset of data to identify variables predictive of complex hospitalization. Significant variables were then included in a reduced multivariate logistic regression using all available data.

Multiple indices were created using combinations of variables predictive of complex hospitalization. Cutoff analyses were performed for all linear variables; each variable was allocated a score of 0 for a value below cutoff and a score of 1 for a value above cutoff. ‘Presentation’ indices were created using significant variables only immediately available at ED presentation. For each presentation index, a companion ‘early admission’ index was created which also included significant variables whose results would be available shortly after admission, including results of C. difficile testing and endoscopy. Receiver operator characteristic (ROC) curves were generated using the Youden index for all indices. Area under the curve (AUC) is presented with 95% confidence intervals (CI).

Study data were collected and managed using REDCap electronic data capture tools hosted at NYU [21,22]. Statistical analyses and graphic generation were performed using R version 3.3.3 [23]. A P-value of <0.05 was considered statistically significant. This study was approved by the Institutional Review Board of New York University Langone Health.

Results

Patient Characteristics and Outcomes

Of the 258 patients hospitalized for IBD flare and tested for CDI in the study period, 164 patients met inclusion criteria (Figure 1). Of these 164 patients, 41.5% (68) had CD and 57.9% (95) had UC. The median age at hospitalization was 34.4 years (IQR 25.8–49.8). Complex hospitalization was present in 34% (56 of 164) of patients studied. While the median LOS was 3.74 days (IQR 2.43–6.61), 21.3% of patients (35 of 164) had a prolonged LOS of > 7 days. Inpatient TNF inhibitor therapy was initiated in 11.6% (19 of 164) of patients and inpatient cyclosporine was initiated in 3.0% of patients (5 of 164). Ten (6.1%) patients underwent inpatient surgery during hospitalization. Demographics, medications prior to admission, and laboratory results on admission are outlined in Table 1.

Figure 1:

Patient selection for the study. Supplementary Table 4 details the non-flare primary hospitalization diagnoses. Abbreviations: IBD (Inflammatory Bowel Disease); CDI (Clostridioides difficile Infection); CD (Crohn’s Disease)

Table 1:

Baseline Demographics of the Study Population

Characteristic		Total (n=164)
Median age, years (IQR)		34.4 (25.8–49.8)
Female, % (n)		54.3 (89)
Race, % (n)
	White	50.6 (83)
	Other	25.6 (40)
	Black/African-American	18.3 (30)
	Asian	3–7 (6)
	Not Reported	2–4 (4)
	More than One Race	0.61 (1)
Body Mass Index, median (IQR)		23.4 (20.2–26.8)
Charlson Comorbidity Index, median (IQR)		0 (0–2)
Tobacco use, % (n)		13.4 (22)
IBD Subtype, % (n)
	Crohn’s Disease	41.5 (68)
	Ulcerative Colitis	57.9 (95)
	Indeterminate Colitis	0.61 (1)
Disease duration at admission (years), median (IQR)		7.56 (2.65–14.7)
IBD medications prior to admission
	Oral 5-aminosalicylates, % (n)	39.6 (65)
	Oral steroids, % (n)	27.4 (45)
	Immunomodulators (6-MP, AZA, MTX), % (n)	13.4 (22)
	TNF inhibitor, % (n)	19.5 (32)
	Cyclosporine, % (n)	0(0)
	JAK inhibitor, % (n)	0.61 (1)
New IBD medications initiated during admission
	Oral 5-aminosalicylates, % (n)	12.8 (21)
	IV Steroids, % (n)	65.2 (107)
	Immunomodulators (6-MP, AZA, MTX), % (n)	0 (0)
	TNF inhibitor, % (n)	11.6 (19)
	Cyclosporine, % (n)	3.0 (5)
	JAK inhibitor, % (n)	1–3 (2)
Temperature (C), median (IQR)		37.3 (37.1–37.7)
Heart Rate (/min), median (IQR)		100 (90–113)
Systolic blood pressure (mmHg), median (IQR)		96 (91–105)
Diastolic blood pressure (mmHg), median (IQR)		53 (47–57)
Admission Laboratory Values
	White blood count (10^9/L), median (IQR)	9.64 (7.50–13.5)
	Hematocrit (%), median (IQR)	36.5 (31.7–40.8)
	Platelet Count(10^9/L), median (IQR)	330 (254–451)
	Albumin (g/dL), median (IQR)	3.6 (3.1–4.0)
	Creatinine (mg/dL), median (IQR)	0.80 (0.70–0.99)
	CRP (mg/L), median (IQR)	33.0 (9.10–82.0)
	Erythrocyte Sedimentation Rate (mm/hr), median (IQR)	38.0 (22.8–63.3)
	Lactate (mmol/L), median (IQR)	1.14 (0.86–1.46)
Cross-axial imaging during admission
	Cross-axial imaging with colitis, % (n)	51.4 (75)
	Cross-axial imaging not performed, % (n)	40.2 (66)
Endoscopy during admission
	Endoscopy with colitis, % (n)	56.7 (93)
	Endoscopy not performed, % (n)	37.2 (61)
C. difficile positive, % (n)		26.8 (44)
Complex hospitalization, % (n)		34.1 (56)
	Length of stay (days), median (IQR)	3.74 (2.43–6.61)
	Prolonged length of stay (> 7 days), % (n)	21.3 (35)
	Inpatient surgery, % (n)	6.1 (10)
	New TNF inhibitor or cyclosporine, % (n)	14.6 (24)

Abbreviations: CRP (C-reactive Protein); IBD (Inflammatory Bowel Disease); IQR (Interquartile Range); JAK (Janus Kinase); TNF (Tumor Necrosis Factor)

Variable Selection

In the internal k-fold cross validation on 134 patients, lasso regression was performed on the following variables: age, body mass index, race, sex, Charlson comorbidity index, temperature, heart rate, systolic blood pressure, diastolic blood pressure, white blood count, hematocrit, platelet, albumin, creatinine, C-reactive protein, CDI, cross-axial imaging demonstrating colitis, and endoscopy demonstrating colitis. Prior biologic use did not correlate with the primary outcome. In the multivariate logistic regression on all 164 patients, the following variables predicted complex hospitalization: tachycardia (OR 3.35; CI 1.79–4.91), hypotension (OR 3.45; CI 1.79–5.11), hypoalbuminemia (OR 2.54; CI 1.15–3.93), CDI (OR 2.62; CI 1.02–4.22), and colitis seen on endoscopy (OR 4.75; CI 1.75–5.15) (Table 2).

Table 2:

Significant Variables from the Multivariable Logistic Regression

Variable	Odds Ratio	Confidence Interval
Tachycardia (HR>=99 BPM)	3.35	1.79–4.91
Hypoalbuminemia (dL/g)	2.54	1.15–3.93
C. difficile positive	2.62	1.02–4.22
Endoscopy showing colitis	4.75	1.75–5.15
Hypotension (SBP<100 mmHg)	3.45	1.79–5.11

All variables demonstrated statistical significance (p<0.05). Abbreviations: BPM (Beats Per Minute); HR (Heart Rate); SBP (Systolic Blood Pressure)

These variables remained significant in analyses of a subset of 141 patients who had C. difficile testing performed within 48 hours of ED presentation to confirm that delays in C. difficile testing were not a confounding factor for the primary outcome.

The optimal cutoff for heart rate was 99 BPM (AUC 0.716; CI 0.631–0.801; sensitivity 78.6%, sensitivity 58.8%), and 100 mm Hg for systolic blood pressure (AUC 0.689; CI 0.605–0.774; sensitivity 85.7%, specificity 43.3%). The optimal cutoff for albumin was 3.4 g/dL (AUC 0.696; CI 0.611–0.780; sensitivity 67.3%, specificity 63.1%).

Index Development and Selection

Two ‘presentation’ indices were generated using the significant variables that would reasonably be available at ED presentation: heart rate, systolic blood pressure, and albumin. Systolic blood pressure was omitted in the second presentation index; this was done to augment index simplicity given that the covariate tachycardia is an earlier sign of clinical dehydration. Companion ‘early admission’ indices were created by sequentially adding endoscopy and the presence of CDI. Sensitivity, specificity, and area under the curve for each index are shown in Table 3. There were no significant differences in the area under the curve among the indices created (Table 3). Consequently, the simplest ‘presentation’ index of heart rate and albumin (AUC 0.744; CI 0.671–0.816; sensitivity 54.5%; specificity 80.2%) and its companion ‘early admission’ index adding CDI and endoscopy results (AUC 0.795; CI 0.727–0.864; sensitivity 58.2%; specificity 85.1%) were chosen for further analysis. The ideal cutoff for a positive presentation score was a score of 2. The ideal cutoff for a positive early admission score was a score of 3 or greater. Figure 2 demonstrates ROC curves for the two indices. Supplementary Table 1 outlines the performances at each cutoff for the presentation and early admission indices. There were no differences in score performance by AUC between study centers (Supplementary Table 2).

Table 3:

Comparing Multiple Severity Indices

	HR + albumin	HR + SBP + albumin	HR + albumin + endoscopy	HR + SBP + albumin + endoscopy	HR + albumin + CDI + endoscopy	HR + SBP + albumin + CDI+ endoscopy
Sensitivity	54.5%	85.5%	85.5%	76.3%	58.2%	83.6%
Specificity	80.2%	59.4%	61.4%	72.2%	85.1%	66.3%
PPV	60%	53.4%	54.7%	60%	68.1%	57.5%
NPV	76.5%	88.2%	88.6%	84.9%	78.9%	88.2%
AUC	0.744 (0.671–0.816)	0.775 (0.704, 0.846)	0.773 (0.701–0.845)	0.793 (0.723–0.862)	0.795 (0.727–0.864)	0.817 (0.75–0.883)

A comparison of the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the curve (AUC) of multiple severity indices. Abbreviations: CDI (Clostridioides difficile Infection); HR (Heart Rate); SBP (Systolic Blood Pressure).

Figure 2:

Receiver Operator Characteristic (ROC) Curves for (A) Presentation and (B) Early Admission indices. Dotted lines demonstrate the optimal cutoff. Area Under Curve (AUC) +/− confidence interval included.

Index Performance

Of all patients with a presentation score of 0 – heart rate below 99 and albumin of at least 3.4 g/dL – 91.7% (44 of 48) had a noncomplex hospitalization. Similarly, 91.5% (54 of 59) of all patients with an early admission score of 1 or below had a noncomplex hospitalization. Conversely, 30 of 50 (60%) patients with a maximum presentation score of 2 and 5 of 6 (83.3%) patients with a maximum early admission score of 4 had a complex hospitalization (Figure 3).

Figure 3:

The number of patients at each prediction score for the presentation (A) and early admission (B) indices. Percentage of patients with each score who have a noncomplex stay are shown.

Secondary Outcomes

Of all patients with a presentation score of 0, 97.9% (47/48) had a LOS less than 7 days, 100% (48 of 48) did not have surgery, and 93.8% (45 of 48) did not receive a TNF inhibitor or cyclosporine. Of all patients with an early admission score of 0, 100% (16 of 16) had a LOS less than 7 days, 100% (16 of 16) did not receive inpatient surgery, and 93.8% (15 of 16) did not receive a TNF inhibitor or cyclosporine. Complete sensitivities, specificities, positive predictive values, and negative predictive values are demonstrated in Supplementary Table 3.

Comparison to ACE Score

Optimal cutoffs were a score of 1 for ACE at presentation (AUC 0.683; CI 0.599–0.767; sensitivity 72.0%; specificity 62.6%), and a score of 2 for the full ACE score (AUC 0.710; CI 0.627–0.794; sensitivity 60%; specificity 73.3%). Of all patients with an ACE at presentation score of 0, 80.3% (57 of 71) had a noncomplex stay, while 76.7% (66 of 86) patients with an ACE at hospitalization score of 1 or below had a noncomplex stay (Supplementary Figure 1). The Early Admission score trended towards a higher AUC than the ACE score; however, this difference did not cross the statistical significance threshold (0.795 [Early Admission Score; 95% CI 0.727–0.864] vs. 0.710 [ACE Score; 95% CI 0.627–0.794], p=0.065; Supplementary Figure 2).

Restriction to Patients Treated with Intravenous Steroids

A sub-analysis was performed solely on patients who received intravenous (IV) steroids at admission (n=104) to study score performances in this subgroup of patients. Of all patients initially treated with IV steroids with a presentation score of 0, 92.9% (26 of 28) had a noncomplex stay. Of all patients initially treated with IV steroids with an early admission score of 1 or less, 91.9% (34 of 37) had a noncomplex stay (Supplementary Figure 3). Complete performance metrics can be seen in Supplementary Table 5.

Sub-analysis by Disease Subtype

In a separate sub-analysis, the performances of the scores were evaluated separately for ulcerative colitis patients (n=90) and for patients with Crohn’s disease with colonic involvement (n=65). Of UC patients with a presentation score of 0, 91% (21 of 23) had a noncomplex hospitalization. Of those with an early admission score of 1 or less, 93% (26 of 28) had a noncomplex hospitalization (Supplementary Figure 4). The performances of the presentation and early admission scores in the subgroup of CD patients were similar (Supplementary Figure 4). Complete metrics can be seen in Supplementary Table 6.

Discussion

In this multicenter study, we present an internally validated simple score that can be employed at ED presentation to determine complex hospitalization in all IBD patients admitted with flare of UC and Crohn’s with colonic involvement. At a time point of undifferentiation, this score demonstrates value in prognosticating the hospitalization course, including duration of stay, requirement of salvage therapy, or inpatient surgery. Specifically, those with a presentation score of 0 (heart rate < 99, albumin >=3.4 g/dL) have a 92% likelihood of having a noncomplex hospitalization. Additionally, we created an early admission score, factoring in the presence of CDI and colitis on endoscopy, demonstrating that 92% patients with a score of 1 or below have a noncomplex hospitalization.

Distinguishing Factors of this Study

This study addresses multiple limitations of existing scores. First, the broad inclusion criteria of this study translate to a higher real-world applicability at a critical time point in patient care. Prior scores have solely focused on acute severe UC patients as defined by the modified Truelove and Witts score (TWS) [13,14,16,24]. Many non-severe patients, however, experience adverse in-hospital outcomes: in a 2010 study, 16.4% of UC patients hospitalized with non-severe flare required infliximab, cyclosporine, or surgical therapy [9,15]. This subset limitation has also led to a paucity of guidance for patients with flare of colonic CD. In this study, not only are Crohn’s colitis patients included, but also all illness levels are considered. By not limiting the severity level of our studied population at the outset, we are able to identify relevant predictive components of severity (i.e. tachycardia, a component of the modified TWS) while maintaining broad applicability.

Another important factor that has either been excluded or controlled for in prior predictive scores [19,24] is CDI, despite the known interplay between CDI and flare. IBD patients are 5–8 times as likely as non-IBD patients to develop CDI [25,26]. Patients with concurrent CDI and flare have worse outcomes, including higher mortality, LOS, and resource utilization [27,28]. Of the 44 C. difficile positive patients in this study, 19 (38.6%) received concurrent steroids, one received infliximab, one started cyclosporine, one underwent colectomy, and 18 (40.9%) had a prolonged LOS. These findings highlight the importance of including CDI testing when evaluating for complex hospitalization, and provide early prognosis when it is unclear whether symptoms are due to flare or CDI.

Although predictive factors of prolonged LOS in hospitalized IBD patients have been evaluated separately [29,30], this study is the first to incorporate prolonged LOS in a composite of adverse outcomes. Predicting extended hospitalization duration likely helps prognosticate the known associated risks of venous thromboembolism [31] and hospital-acquired infections [32], although these associated outcomes were not directly evaluated in this study. It is unlikely that CDI was acquired during hospitalization in this study; 141 of 164 (86%) of CDI tests were sent within 48 hours of presentation. Even in this subset of patients, CDI remained a significant predictor of the primary outcome in the multivariate logistic regression. This is consistent with prior findings that over 75% of CDI in hospitalized IBD patients are community acquired [33].

Additionally, with the exception of the ACE score, the majority of indices can only be calculated at day 3 of hospitalization. The presentation score developed in this study can be calculated at ED presentation, averting the loss of valuable time for early intervention and potential escalation of therapy.

Comparison of Outcomes

Our study had a similar median LOS [30] and surgery rate [10] compared to prior studies, though rates of TNF inhibitor and cyclosporine administration varied. In our study, 11.6% of patients received TNF inhibitors and 3.0% received cyclosporine. In prior studies, the TNF inhibitor rate has ranged from 5.5% to 21.1% [10,19], while cyclosporine administration rates have been as high as 22% [19]. These differences in administration rates are likely due to differing institutional practices and different patient populations.

Comparison of Predictive Factors

The variables predictive of complex hospitalization in this study were tachycardia at presentation, albumin at presentation, endoscopic inflammation, and CDI. This is largely consistent with prior findings; multiple studies have found day 3 albumin to be predictive of inpatient colectomy in UC patients [14,15,19]. Tachycardia approached significance for predicting prolonged LOS in a study by Kelso, et. al [30]. The majority of studies examining surgery and second-line therapy outcomes had likely already selected for tachycardic patients by limiting their evaluation to severe patients by TWS.

Endoscopic inflammation is a strong predictor in multiple other studies: the Mayo score had an OR of 3.17 in one study [19]. Additionally, the Ulcerative Colitis Endoscopic Index of Severity was an independent predictor of needing rescue therapy or colectomy [17,18], even outperforming the endoscopic Mayo score [24].

Temperature [12] and CRP [12,13,19] were significant factors in other studies; although included in our study, these were not significant in the multivariate analysis. Though Jain, et. al. found fecal calprotectin of over 1000 as an independent predictor of complex hospitalization [18], the clinical utilization of fecal calprotectin during admission in our study was not high enough to perform meaningful analyses.

Validation of ACE score

Our analyses provide external validation to the ACE score (Supplementary Figure 1), albeit not for its original purpose [19] of predicting non-response to IV steroids in UC patients hospitalized with flare. Here we show that the ACE score is also able to predict those patients who will have complex hospitalization with an AUC of 0.710 (CI 0.627–0.794). As previously discussed, our indices have overlapping components of albumin and inflammation on endoscopy. CRP, however, was not a significant predictor in our study. Although our equivalent early admission score had a higher specificity (85.1%) compared to the ACE score (73.3%), the higher AUC compared to that of the ACE score did not cross the statistical significance threshold (p=0.065).

Limitations

This study has certain limitations. First, it is a retrospective evaluation of electronic medical record data. As such, the IBD diagnoses in the medical record may not always be accurate. However, diagnosis was manually confirmed for all patients using gastroenterology consultant documentation to mitigate this possibility. Another limitation is that patients who were not tested for CDI were excluded; though it is rare for IBD patients admitted with flare symptoms to not be tested for CDI, it is possible we may be missing a portion of hospitalized IBD patients. Although we did not collect data on symptoms – such as frequency and characteristic of bowel movements – these are known to be difficult to characterize with high subjectivity in interpretation.

It may be argued that including both UC and CD patients with colonic disease creates excess heterogeneity within our study population. However, even when patients were separated by subtype, the scores performed well in both the UC and CD subgroups. We believe that including Crohn’s colitis patients lends insight into a previously unstudied population, and is a strength of our study.

Although this study was not limited solely to patients who were initially treated with IV steroids, the sub-analysis focusing on such patients demonstrates the persistent ability of the proposed scores to predict noncomplex hospitalization well. A limitation of the Early Admission Score is that patients who have endoscopy will inevitably have longer hospitalization than those who do not, thus potentially introducing bias in predicting prolonged LOS. Finally, we have a relatively small sample size; though k-fold cross validation decreases the effect of small sample size, our findings will need external validation.

Conclusion

Here we present a simple, ED-based score to recognize those IBD patients presenting with flare who are at risk for prolonged length of stay, requiring salvage therapy, or inpatient surgery. Additionally, we report an early admission score that includes the factors of CDI and endoscopy. These presentation and early admission scores may provide prognostic information and guide subsequent management in IBD patients who require hospitalization.