Utility of Endoscopy in Hospitalized Patients with Gastrointestinal Hemorrhage and Pulmonary Hypertension

Abstract

Background

Gastrointestinal hemorrhage (GIH) has been reported as one of the most common GI complications in patients with pulmonary hypertension (PH). There is paucity of data on the national burden of GIH in patients with PH. We aimed to assess the prevalence, trends and outcomes of endoscopic interventions in patients with PH who were admitted with GIH.

Method

We queried National Inpatient Sample (NIS) database from 2005 to 2014 and identified the patients hospitalized with primary or secondary discharge diagnosis of PH (ICD 9 CM Code: 416.0, 416.8, and 416.9). Using Clinical Classification Software Coding system (153) patients with concurrent diagnosis of GIH were then identified. We studied the prevalence and trends of GIH in PH, factors associated with GIH, use of endoscopy, factors associated with utilization of endoscopic interventions, endoscopy outcomes including mortality, and overall healthcare burden.

Results

Out of 7,586,973 PH hospitalizations 3.2% (N = 246,358) had concurrent GIH, with a rising prevalence of GIH in PH patients during the last decade. Clinical predictors for GIH in PH included older age, congestive heart failure, anticoagulation therapy and concurrent alcohol abuse. Mean length of stay (LOS) in PH patients hospitalized with GIH was significantly higher than without GIH (8.6 vs. 6.4 days, p < 0.01) along with a significant increase in hospitalization cost ($20,189 vs. $14,807, p < 0.01). Similarly, odds of in-hospital mortality increase by ~ 1.5 times in PH patients with GIH than those without it (adjusted odds ratio [aOR: 1.45, 95%CI: 1.43–1.47]). Endoscopic interventions were performed in 48.6% of patients with PH and GIH during their hospitalization. Older patients were more likely to undergo endoscopy, as well as the patients who received blood transfusion, and those with hypovolemic shock. Patients with acute respiratory failure and acute renal failure were less likely to get endoscopy. Mean LOS in patients undergoing endoscopic intervention was significantly higher than those who did not receive any intervention (8.7 vs. 8.4 days, p < 0.01), without a substantial increase in hospitalization cost ($20,344 vs. $20,041, p < 0.01). Also, there was a significant decrease in in-hospital mortality in patients undergoing endoscopic interventions.

Conclusion

Concurrent GIH in patients with PH increases length of stay; healthcare costs and increases in-hospital mortality. Use of endoscopic interventions in these patients is associated with reduced length of stay, in-hospital mortality without significantly increasing the overall health care burden and should be considered in hospitalized patients with PH who are admitted with GIH. Future studies comparing GIH patients with and without PH should be done to assess if PH is a risk factor for worse outcomes.

Introduction

Pulmonary hypertension (PH) is a spectrum of diseases affecting pulmonary vasculature and is defined as the elevation of pulmonary arterial [PA] pressure ≥ 25 mm Hg [1]. Systemic and perivascular inflammation, insulin resistance, and oxidative stress in patients with PH lead to vascular remodeling. Obesity is associated with low-grade inflammation, insulin resistance and oxidative stress that may exacerbate the vascular remodeling process in patients with PH [2]. Therefore, it can be speculated that with the rising prevalence of obesity, the prevalence of PH is on the rise as well. Medical treatment of PH typically comprises of management of heart failure (HF), lowering of PA pressure and use of anticoagulation (AC) therapy to prevent thromboembolic events [3–5]. Gastrointestinal hemorrhage (GIH) has been suggested to be the major GI manifestation in patients with PH. Patients with underlying connective tissue disorders, like systemic sclerosis or porto-pulmonary hypertension may be at higher risk of GIH [5]. Presence of varices, thrombocytopenia, deficiencies of clotting factors and von Willebrand factor (vWF) have been suggested to be the underlying mechanisms to cause GIH in these patients. Certain PH specific medications like prostanoids (e.g., epoprostenol) and AC with Vitamin K antagonist further increase the risk of GIH.

PH patients with underlying cardiac dysfunction are volume sensitive. Therefore, occurrence of an otherwise mild hemorrhage could lead to catastrophic events in these patients. Acute volume depletion and acute blood loss anemia with GIH can worsen baseline hypoxemia. Further, lactic acidosis in these patients can increase PA pressure leading to right-sided HF which could be life threatening [5]. Therefore, early and aggressive treatment of any GIH is imperative in these patients. Endoscopic modalities are now the mainstay of GIH diagnosis and offer the potential for therapy. But, there is paucity of data on the national burden of GIH in patients with PH, utilization and role of endoscopy in this patient population. Using the largest publically available national inpatient sample (NIS) database in Unites States, we aimed to assess the prevalence, trends and outcomes of endoscopic interventions in patients with PH who were admitted with GIH.

Methods

Patient Selection

In this retrospective large database study, using International Classification of Diseases, 9^th revision, Clinical Modification (ICD-9-CM) codes, we queried all patients from the NIS database who were admitted to an acute care hospital with the primary or secondary diagnosis of PH (CM Code 416.0, 416.8, 416. 9). We then identified patients with concurrent diagnosis of GIH by using Clinical Classification Software Coding system (153). For a uniform patient selection we used NIS database between 2005 and 2014 as the NIS data published for the year 2015 utilized the codes for both ICD9 and ICD10 coding system. Using the American Hospital Association annual region survey, hospitals are stratified according to ownership and/or control, bed size, teaching status, urban and/or rural location, and geographic region. A random 20% sample of all hospitals within each stratum is then collected. Data about patient’s demographics, diagnoses, length of stay (LOS), procedures, and hospital charges and all the discharge information from these hospitals are entered into NIS database. To make it a nationally representative sample each discharge is then weighted (weight = total number of discharges from all acute care hospitals in United States divided by number of discharges included in the 20% sample).

The NIS database includes both hospital-level and patient-level information. Using the ICD-9-CM coding system, up to 25 discharge diagnosis and 15 procedures are collected on each hospitalized patient. The NIS dataset has been used previously in various studies to provide reliable burden of GI disease [6, 7]. As this database is a publicly available de-identified national data sample, this study was exempt from the Institutional Review Board at our institution.

Study Variables

Demographic variables included age, gender and race. Comorbidities including alcohol use, congestive heart failure (CHF) and region of hospitalization, hospital bed size, LOS, teaching status, primary payer, hospital LOS and cost of hospitalization are provided in the NIS for each hospitalization. Information related to blood transfusions, hypovolemic shock, and acute respiratory or renal failure is also available. Data regarding inpatient endoscopic or surgical procedures and in-hospital mortality is also provided by NIS.

Outcomes

The primary outcome was in-hospital mortality associated with GIH in PH patients. The secondary outcomes were prevalence of GIH in PH patients, proportion of patients who underwent endoscopy, factor associated with use of endoscopy, LOS and total admission charges. Multiple potential confounders were collected and accounted for during analysis including age, gender, race, number of diagnoses during the hospitalization, number of procedures, need for blood transfusion, concurrent hypovolemic shock, acute renal failure, acute respiratory failure and hospital factors which included location, teaching status and the size of the hospital.

Statistical Analysis

We performed a descriptive statistical analysis to study the demographic variables associated with the outcomes of interest in cases hospitalized with PH and GIH and those undergoing endoscopic evaluation. Continuous variables such as hospital LOS and cost were reported as mean (± standard deviation). We used the discharge weight files provided by the NIS to perform the weighted analysis and to estimate the overall annual hospitalizations. We then used the trend weight (TRENDWT) files provided by healthcare cost and utilization project (HCUP) to verify the trends over the study period. A Pearson’s Chi square test was used to analyze the statistical differences in the categorical outcomes of interest among the study groups. The overall all-cause in-hospital mortality was reported in either percentage (%) or the number of deaths per 1,000 hospitalizations. We performed a multivariate logistic regression analysis to study the differences in the in-hospital mortality associated with PH and GIH, as well as, for the mortality associated with the hospitalized cases who underwent an endoscopic evaluation. We adjusted for the confounding effect of demographic factors such as age, gender, race, primary payer for the hospitalization and the insurance status, hospital factors which included region of the hospital, urban or rural location and teaching status. We adjusted for patient co-morbidities by utilizing Elixhauser’s comorbidity software which generates the Elixhauser comorbidity index (ECI) by utilizes 29 independent patient co-morbidities identified based on the ICD 9 coding system [8]. Indicators of severity of bleeding, such as hypovolemic shock, ARF, need for blood transfusion, respiratory failure, as well as various surrogate markers of hemodynamic status, such as endotracheal intubation, were also used as covariates in multivariable analysis of the outcomes, and the threshold of p value for inclusion in the multivariate regression model was set at 0.05. We calculated a Pearson correlation coefficient to study the trend and reported the p-value. An independent sample t test was used to analyze the differences in LOS (days) and cost of care (U.S. $) among the study groups. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY: IBM Corp. was used for all analyses and a 2-sided p value of less than 0.05 was considered to be statistically significant.

Results

A total number of 371,776,860 hospitalizations were found between year 2005 and 2014. Out of these 7,590,975 were admitted with either the primary or a secondary diagnosis of PH. After excluding cases with missing data mortality data (n = 4,002) our final sample consisted of 7,586,973 hospitalized cases with PH (Supplemental flow chart 1). We extracted NIS trend weight files and used them for our trend analysis. Our sample size, as well as the results for all other descriptive and year-wise trend analysis using discharge weight were exactly identical to when the analysis was performed by using the trendweight files.

GIH and In‑Hospital Mortality

Out of 7,586,973 PH admissions, 246,358 (3.2%) had concomitant GIH at or during hospitalization. In 2005, there were 25 GIH/1000 PH hospitalizations which has slowly been increasing and was 37 GIH/1000 PH admissions in 2014 indicating a ~ 1.5 fold increase in prevalence of GIH in PH patients over the last decade (p < 0.01) (Fig. 1).

Fig. 1.

Graphical representation of rate of gastrointestinal hemorrhage in pulmonary hypertension for each year from 2005 to 2014

In terms of mortality 21,254 of PH patients with GIH died (8.6%) as compared to 331,579 (4.5%) without GIH indicating ~ 2-time higher mortality in PH patients with concomitant GIH (Supplemental flow chart 1). Even on multinomial logistic regression analysis, after adjusting for age, gender, race, primary payer/insurance status, hospital factors (region, location/teaching status, and hospital bed size) and ECI the odds of in-hospital motility were 1.45 times higher (adjusted odd ratio [aOR]: 95% confidence interval {95% CI}: 1.45 (1.43–1.47, p < 0.01) in PH patient with GIH than without GIH (Table 1).

Table 1.

In-Hospital mortality on multinomial logistic regression analysis

	In-hospital mortality associated with GIH in PH	In-hospital mortality associated with endoscopy in GIH and PH
Adjusted odds ratio	1.45	0.32
95% Confidence interval	1.43–1.47	0.31–0.33
P value	p < 0.01	p < 0.01

Adjusted for confounding effects of Age, gender, race, primary payer/insurance status, hospital factors (region, location/teaching status, hospital bed size) and Elixhauser Comorbidity Index

Factors Associated with GIH in PH

In hospitalized patients with PH, as compared patients younger than 35, the odds of GIH in 36–50, 51–65 and > 65 age groups were 1.21, 1.52 and 1.97, respectively (p < 0.01). There was slightly higher risk of GIH in males (aOR: 1.055, 95% CI 1.04–1.06, p < 0.01). In terms of race, compared to whites, Hispanics, Asian/pacific islanders, Native Americans and others had higher risk of GIH (aOR: 1.13, 1.4, 1.18, and 1.15, respectively, p < 0.001 for all), but there was no statistically significant difference among Blacks. Being on long-term AC, alcohol abuse and presence of CHF also increased the risk of having GIH (aOR: 1.13, 1.20, and 1.85, respectively, p < 0.01). Association of insurance status, hospital location, size and status (teaching, rural, urban, etc.) with GIH in hospitalized PH patients is shown in Table 2.

Table 2.

Multinomial logistic regression analysis to assess risk factors associated with GIH in hospitalized patients with PH

Factor	Adjusted odds ratio	95% Confidence interval for adjusted-OR		P value
		Lower limit	Upper limit
Age groups
18–35	R	–	–	–
36–50	1.211	1.159	1.265	P < 0.01
51–65	1.524	1.463	1.587	P < 0.01
>65	1.972	1.894	2.055	P < 0.01
Gender
Men	1.055	1.046	1.064	P < 0.01
Women	R	–	–
Race
White	R	–	–	–
Black	0.996	0.984	1.009	P = 0.549
Hispanic	1.131	1.112	1.152	P < 0.01
Asian or Pacific Islander	1.370	1.330	1.412	P < 0.01
Native American	1.177	1.111	1.248	P < 0.01
Others	1.149	1.117	1.181	P < 0.01
Long-term anticoagulation
Yes	1.127	1.112	1.141	P < 0.01
No	R
Alcohol abuse
Yes	1.996	1.953	2.036	P < 0.01
No	R	–	–	–
CHF
Yes	1.862	1.845	1.876	P < 0.01
No	R	–	–	–
Primary expected payer
Medicare	R	–	–	–
Medicaid	0.955	0.936	0.975	P < 0.01
Private Insurance	0.893	0.880	0.907	P < 0.01
Self-Pay	0.883	0.853	0.916	P < 0.01
No charge	0.698	0.627	0.778	P < 0.01
Others	0.869	0.837	0.903	P < 0.01
Hospital region
Northeast	1.007	0.993	1.021	P = 0.319
Midwest	1.045	1.030	1.059	P < 0.01
South	1.018	1.005	1.030	P < 0.01
West	R
Location and teaching status
Rural	0.902	0.888	0.916	P < 0.01
Urban non-teaching	0.972	0.963	0.981	P < 0.01
Urban teaching	R
Hospital bed-size
Small	0.955	0.942	0.968	P < 0.01
Medium	0.971	0.961	0.981	P < 0.01
Large	R

R Reference category

Factors Associated with Endoscopy

Compared to patients between age group 18 and 35 admitted with PH and GIH, patients in age groups 36–50, 51–65 and > 65 had higher odds of endoscopic evaluation (aOR: 1.41, 1.5 1and 1.25, respectively, p < 0.01 for all). Similarly, males had slightly higher chances of undergoing endoscopy than females (aOR: 1.10, p < 0.01). In terms of ethnicity, Hispanics and Asian/Pacific islanders had higher (aOR: 1.11 for both) and, blacks and others had lesser odds (aOR: 0.97 and 0.84, p < 0.05) of undergoing endoscopic evaluation than whites. There were no statistically significant differences for endoscopic evaluation among Native American and whites (p > 0.05). Requirement of blood transfusions and presence of hypovolemic shock were associated with 2- and 1.5 times higher chances of getting endoscopy (aOR: 2.04 and 1.50, p < 0.01). But, odds of getting endoscopic evaluation were lesser in patients with acute respiratory and renal failure (aOR: 0.81 and 0.60, p < 0.01). Differences in endoscopy based on insurance status, geographical location, size and type of hospital are shown in Table 3.

Table 3.

Multinomial logistic regression analysis to identify clinical factors associated with Endoscopic evaluation in hospitalized patients with PH and GIH

Factor	Adjusted odds ratio	95% Confidence interval for adjusted-OR		P value
		Lower limit	Upper limit
Age groups
18–35	R	–	–	–
36–50	1.418	1.297	1.550	P < 0.01
51–65	1.518	1.397	1.650	P < 0.01
>65	1.249	1.150	1.358	P < 0.01
Gender
Men	1.103	1.083	1.122	P < 0.01
Women	R	–	–
Race
White	R	–	–	–
Black	0.975	0.950	1.000	P = 0.04
Hispanic	1.110	1.070	1.151	P < 0.01
Asian or Pacific Islander	1.112	1.047	1.181	P < 0.01
Native American	0.997	0.887	1.121	P > 0.05
Others	0.842	0.795	0.891	P < 0.01
Blood transfusion
Yes	2.040	2.004	2.074	P < 0.01
No	R	–	–	–
Hypovolemic shock
Yes	1.501	1.443	1.564	P < 0.01
No	R	–	–	–
Acute renal failure
Yes	0.812	0.796	0.827	P < 0.01
No	R
Acute respiratory failure
Yes	0.603	0.586	0.662	P < 0.01
No	R
Primary expected payer
Medicare	R	–	–	–
Medicaid	1.018	0.977	1.060	P = 0.399
Private insurance	1.106	1.073	1.141	P < 0.01
Self-pay	1.259	1.171	1.354	P < 0.01
No charge	1.028	0.826	1.280	P = 0.802
Others	0.882	0.815	0.953	P < 0.01
Hospital region
Northeast	0.867	0.843	0.892	P < 0.01
Midwest	1.894	1.161	1.228	P < 0.01
South	0.960	0.937	0.984	P < 0.01
West	R
Location and teaching status
Rural	0.594	0.576	0.613	P < 0.01
Urban non-teaching	0.959	0.941	0.977	P < 0.01
Urban teaching	R
Hospital bed-size
Small	0.796	0.775	0.818	P < 0.01
Medium	0.913	0.895	0.932	P < 0.01
Large	R

R Reference category

Role of Endoscopy

Of the 246,358 PH hospitalizations with concurrent GIH, 48.9% received GI endoscopy, which included 71,748 (59.5%) esophagogastroduodenoscopies (EGD) and 12,861 (10.7%) colonoscopies, while remaining 35,888 (29.8%) patients underwent both EGD and colonoscopy. The in-hospital mortality in patients who received endoscopy was ~ 3 times less than the patients who did not receive endoscopy (4.5% vs. 12.6%) (Supplemental flow chart 2). In terms of trend of GI endoscopy among PH patients with GIH, there has been an overall increase in percentage of patients receiving endoscopy in the last decade, with maximum use of endoscopy in 2010 (49.9%) (Supplemental Figure 1). The in-hospital mortality trend in PH patients with and without endoscopic intervention is shown in Fig. 2.

Fig. 2.

Graphical representation of in-hospital mortality trend in pulmonary hypertension cases with gastrointestinal hemorrhage for each year from 2005 to 2014

On multinomial regression analysis, after adjusting for age, gender, race, primary payer/insurance status, hospital factors (region, location/teaching status, hospital bed size) and ECI in PH patients with GIH undergoing endoscopic evaluations odds of in-hospital mortality decreased significantly (aOR: 0.32, p < 0.01). (~ 70%) (Table 1)

LOS and Cost of Hospitalization

Compared to patients without GIH, presence of concomitant GIH in patients admitted with PH leads to a significant increase in LOS (6.4 vs. 8.6 days, p < 0.01) and cause significant increase in cost of hospitalization ($14, 807 vs. $20, 189, p < 0.01). There was minor but statistically significant increase in LOS and cost of hospitalizations in PH patients who underwent endoscopic evaluation for GIH than PH patients with GIH who did not get endoscopy (8.7 vs. 8.4 days and $20,344 vs. $20, 041, respectively, p < 0.01 for both). (Table 4).

Table 4.

Comparison of Length of Stay (LOS) and cost of hospitalizations

		LOS (Mean [SD]-Days	Hospital cost mean (US $)	P value
GIH in PH	Yes	8.6 (8.9)	$ 20, 189	P < 0.01
	No	6.4 (6.5)	$ 14, 807	P < 0.01
Endoscopy in GIH and PH	Yes	8.7 (8.5)	$ 20, 344	P < 0.01
	No	8.4 (9.3)	$ 20, 041	P < 0.01

Discussion

Our study suggests that between 2004 and 2015, in hospitalized patients with PH, the prevalence of concurrent GIH has increased by nearly 1.5-fold, with a parallel (~ 1.4-fold) rise in the in-hospital mortality. Besides older age (> 35), certain races (Hispanics, Asian or Pacific islanders, Native Americans and other), and being admitted at hospitals in specific geographical locations (Northeast, Midwest and South region), the risk of GIH among hospitalized PH patients was significantly higher in patients on long-term AC, alcohol use and with CHF (aOR: 1.13, 1.99, and 1.86, respectively, p < 0.01 for all). The mean LOS in PH patients with concomitant GIH was significantly higher than without GIH (8.6 vs. 6.4 days, p < 0.01) along with a significant increase in hospitalization cost ($20,189 vs. $14,807, p < 0.01). Similarly, odds of in-hospital mortality increase by ~ 1.5 times in PH patients with GIH than those without it (aOR: 1.45, 95%CI: 1.43–1.47]) (Table 1). Endoscopy was performed in 48.6% of patients with PH and GIH during their hospitalization. Odds of undergoing endoscopic evaluation were higher in patients > 35 years, males and Hispanic and Asian/Pacific Islanders (p < 0.05). Patients who required blood transfusions (aOR: 2.04, p < 0.01) or had hypovolemic shock (aOR: 1.50, p < 0.01) also had higher chance of receiving endoscopy. But, GIH patients with ARF or acute respiratory failure had lesser odds of getting endoscopy (aOR: 0.81 and 0.79, respectively, p < 0.01 for both). In patients undergoing endoscopy, mean LOS was marginally higher than those who did not receive any intervention (8.7 vs. 8.4 days, p < 0.01), without a substantial increase in hospitalization cost ($20,344 vs. $20,041, p < 0.01). On multinomial regression analysis, a significant decrease in in-hospital mortality was noted among PH patients with GIH who underwent endoscopic evaluations (aOR: 0.31, p < 0.01).

Over the last decade along with increase in diagnostic/imaging modalities to detect PH, rising prevalence of obesity and other lung diseases have led to increase in overall PH prevalence [2, 3]. Most of these patients have underlying cardiac and other medical co-morbidities, and anticoagulants are frequently used in PH patients to prolong survival [1, 4]. Presence of varices, low or dysfunctional platelets, relative depletion of vWF, use of prostacyclin analogues, and AC use in PH patients increase the risk of major bleeding, most common being GIH [1, 4]. In a study by Roldan and colleagues [1], the incidence rate of major bleeding in PH patients on AC was 4.7 per 100 patient-years and 72.7% of them had major GIH. Therefore, it’s not surprising that the prevalence of GIH in hospitalized PH patients in our study has increased significantly over the last decade, contrary to an overall decrease incidence and mortality of GI bleeding in general population in United States [9, 10]. Due to poor cardio-pulmonary reserves in PH patients even minor bleeding can lead to a catastrophic which explains the parallel increase in in-hospital mortality [5]. Older age is prone to have higher cardiovascular issues, stroke and malignancy, and elder patients tend to be on AC and anti-platelets/NSAIDs increasing risk of GIH and mortality [11–13]. Similar reasons could also explain the high incidence of GIH in patients on long-term AC and CHF. In addition end-stage HF patients often end up on left ventricular assist device (LVAD) as a bridge to transplantation. GIH is a significant complication of LVAD use in these patients adding to the burden of AC therapies [14]. PH patients with underlying alcohol use often have portal hypertension as a complication of alcoholic liver disease (ALD) and in some case as a result of PH itself [15] [16]. In addition to variceal bleeding, ALD patients have more frequent and more severe bleeding from peptic ulcer disease (PUD), Dieulafoy’s lesion (DL), and Mallory–Weiss syndrome (MWS) explaining higher odds of bleeding associated with alcohol in our study population [16]. Medicare patients tend to have significant percentage of uninsured adults with more chronic diseases and conditions with more medical co-morbidities putting them at higher risk of GIH [17, 18]. Urban teaching and large sized hospitals tends to have more specialized care with referral from regional, non-teaching or rural hospital corresponding to our finding of more GIH patients seen at large tertiary care teaching hospitals.

Endoscopy remains the “gold standard” modality to assess and treat GIH. Recent studies have demonstrated that early risk stratification and timely endoscopic therapy for GIH leads to rapid achievement of hemostasis, decrease rebleeding, need for surgical interventions, and decrease in in-hospital mortality [6, 9, 10]. In our study 48.6% underwent endoscopic evaluation for GIH, with an overall increase in percentage of patients receiving endoscopy in the last decade. The in-hospital mortality in patients who received endoscopy was ~ 3 times less than the patients who did not receive endoscopy (4.5% vs. 12.6%). This difference in mortality persisted even after adjusting for age, gender, race, primary payer/insurance status, hospital factors (region, location/teaching status, hospital bed size) and ECI (aOR: 0.32, p < 0.01). In addition to achieving hemostasis, risk stratification of GI lesions and resumption of life saving anti-coagulation/anti-platelet therapies in PH patients after GIH could be responsible for low overall mortality after endoscopy. In terms of factors associated with endoscopic evaluation, presence of hypovolemic shock and need of blood transfusions more were the strongest factors suggesting need for endoscopy. Hypotension in patients with GIH is a clinical indicator of severity of blood loss [19, 20]. In a recent study by Oakland et al. systolic blood pressure was used to calculate shock index (SI) in patients with lower GI bleed and stratify them as stable vs. unstable (SI > 1), which dictates further management with endoscopy vs. imaging/interventional radiology procedures to achieve hemostasis [20]. Similarly, requirement of blood transfusion suggests significant blood loss and need for endoscopic interventions. Patients older than 35 also had high chances of getting endoscopy, with the highest odds in patients between 51 and 65 years followed by 36 and 50 years. Older adults with PH and GIH tend to be sicker with underlying co-morbidities and also need endoscopic evaluation to identify bleeding lesions, as they might need to resume anti-coagulant/anti-platelet therapies [14, 19, 20]. In contrary, patients < 35 have higher chances of quicker recovery after GIH, not being on blood thinner and getting endoscopy as outpatient. Male gender is a risk for PUD and alcoholic liver disease [16]. In a nationwide database study from Taiwan, advanced age, male gender, diabetes mellitus, chronic renal disease, prior variceal bleeding, and use of NSAIDs were risk factors for PUD in the cirrhotic patients [21]. In addition, bleeding episodes in these patients are more frequent and severe explaining higher likelihood of undergoing endoscopic evaluation. Compare to large urban teaching hospital, patients admitted at rural, non-teaching and small to medium size hospitals had lesser odds of undergoing endoscopy. Large urban teaching hospitals are well quipped in managing urgent/emergent GIH, have around the clock availability of endoscopist and get transfers from smaller centers with severe illness requiring endoscopic assessments. Similar findings have been shown in a recent study with GIH in acute stroke patients [7]. PH patients with GIH and concomitant acute respiratory and ARF were found to be less likely to undergo endoscopy. These patients appear to be sicker requiring intensive care unit admission. Endoscopic evaluation especially upper endoscopy in patients with acute respiratory failure could further compromise the respiratory status and may lead to need of mechanical ventilation. PH patients have baseline poor respiratory reserve and unless bleeding is life threatening close monitoring without endoscopy such patients with supportive measures is the best strategy.

Even though the mortality decreased significantly in patients undergoing endoscopy, there was a slight increase in LOS (8.7 vs. 8.4 day, p < 0.01). Increased LOS could be from more severe illness in patients with GIH who received endoscopy. Overall, a decrease in mortality without substantial increase in cost of hospitalization ($20,344 vs. $20,041, p < 0.01) suggests that all PH patients with GIH should be assessed with endoscopy.

Our study has a large national database which makes the results generalizable. Use of a national database also eliminates selection bias seen in single or even multicenter studies. Even though we did not directly adjust for obesity and diabetes, these are components of ECI providing indirect adjustment for these two major co-morbid conditions associated with PH. We also used both patient and hospital related factors for statistical analysis to eliminate various confounders. Despite these strengths our study has several limitations. The NIS database does not provide data on the medications used, severity of GIH, parameters used to blood transfusion, factors leading to decision making regarding endoscopy, type of intervention done and experience of the endoscopist. We also did not quantify GIH into lower and upper GIH and did not include type of lesions. Lastly, we did not assess baseline tobacco use, which could have contributed to complications associated with PH. Despite these limitations, this is the first study to assess the prevalence of GIH in PH hospitalizations and role of endoscopic interventions. With the rising prevalence of PH we will encounter more GIH in PH patients and endoscopy during hospitalization can provide definitive therapy, provide information on the type of lesion to help stratify these patients for antiplatelet and AC therapies, which could be lifesaving.

In conclusion, the overall prevalence of GIH in PH patients is on the rise. Concurrent GIH in patients with PH increases LOS; healthcare costs and in-hospital mortality. Use of endoscopy in these patients is associated with reduced LOS, in-hospital mortality without significantly increasing the overall health care burden and should be considered in all hospitalized PH patients with GIH. Future studies comparing GIH patients with and without PH should be done to assess if PH is a risk factor for worse outcomes.

Abbreviations

ANOVA

Analysis of variance

AC

Anticoagulation

aOR

Adjusted odds ratio

ARF

Acute renal failure

CHF

Congestive heart failure

ECI

Elixhauser comorbidity index

GIH

Gastrointestinal hemorrhage

HCUP

Healthcare cost and utilization project

ICD-9 CM

International Classification of Diseases, 9^th revision, Clinical Modification

LOS

Length of stay

LVAD

Left ventricular assist device

NIS

National inpatient sample

PH

Pulmonary hypertension

PA

Pulmonary artery

TRENDWT

Trend weight

vWF

Von willebrand factor