Hospital‐Acquired Gastrointestinal Bleeding

Study Patients

The institutional review board of the Cleveland Clinic Foundation (Cleveland, OH) approved this study. All patients admitted to the General Internal Medicine service between January 1, 1999, and December 31, 2002, were eligible for inclusion. Two types of cases were included: 1) patients admitted for nongastrointestinal illnesses who developed bleeding at least 24 hours after admission and required esophagogastroduodenoscopy (EGD) during hospitalization (designated in‐hospital bleeding), and 2) patients admitted with gastrointestinal bleeding (requiring EGD) who had been hospitalized on the General Medical service during the preceding 4 weeks for a nongastrointestinal illness (designated out‐of‐hospital bleeding). This second group was included to identify risk factors for delayed bleeding that might not be obvious during hospitalization.

Medical records of all General Medicine patients who underwent EGD were reviewed in a standardized fashion (Fig. 1). We excluded patients with documented gastrointestinal complaints (including bleeding) at the time of the index admission or within 24 hours of admission, bleeding in the intensive care unit (ICU) or in another hospital prior to transfer to the General Medicine service, or a history of gastrointestinal bleeding during the month prior to admission. ICU stay prior to General Medicine admission, if not associated with GI bleeding, was not an exclusion criterion for our study.

Figure 1

Controls, also without any acute gastrointestinal symptoms at admission, were randomly matched to cases in a 1:1 ratio by date of admission. We used this liberal matching strategy because any factors matched for would no longer be eligible to be risk factors for bleeding. If more than one control was admitted on the same day as a case, then a random number was used to select the control.

Definition of Prophylactic Acid Suppression

We defined prophylactic acid suppression as in‐hospital de novo treatment with histamine‐2 receptor antagonists and/or proton pump inhibitors received prior to the onset of any symptoms that would suggest GI bleeding (for cases) or any time during hospitalization (for controls). Patients taking these drugs prior to admission were deemed ineligible for in‐hospital prophylactic acid blockade and were excluded from the related analyses.

Data Collection

We extracted demographic information, medical history, medication usage, and laboratory data by chart review. For those patients readmitted for gastrointestinal bleeding following discharge, data from the initial (nongastrointestinal illnessassociated) hospitalization were recorded. Bleeding symptoms triggering endoscopy were grouped into four categories: 1) melena or hematochezia; 2) hematemesis (frank blood in vomitus or coffee‐grounds emesis); 3) melena or hematochezia plus hematemesis (both 1 and 2); 4) stool positivity for occult blood or unexplained drop in hemoglobin in the absence of overt bleeding. Endoscopic findings were categorized by the nature of the visualized lesions, and if multiple lesions were noted, the endoscopist's impression of the most likely bleeding site was used to define the source of bleeding. We recorded colonoscopy findings for patients undergoing this evaluation.

Statistical Analysis

We analyzed data utilizing JMP 5.1 (SAS Institute, Cary, NC). Random controls were chosen using computer‐generated random numbers. The proportions of patients with various categorical characteristics were compared using the chi‐square test or Fisher's exact test as appropriate. We used the Student t test or Wilcoxon's test to compare continuous variables. Odds ratios and adjusted odds ratios were calculated by logistic regression. Two‐tailed P values less than .05 were considered statistically significant.

RESULTS: Identification of Cases and Controls

Of 17,707 patients admitted to the General Medicine service, 1327 (7.5%) underwent EGD during hospitalization or within 1 month of discharge. Only 73 (0.41%) of the total number of patients met the case definition (Fig. 1). Of these cases, 62 (84.9%) had developed gastrointestinal bleeding during the index hospitalization, whereas 11 (15.1%) were readmitted for bleeding within 4 weeks of hospital discharge. The remaining 1254 patients who underwent EGD were excluded based on exclusion criteria, including an absence of documented bleeding prompting the EGD.

Clinical Risk Factors for Bleeding

In univariate analysis, as shown in Table 1, predictors of GIB included: 1) age (P = .02); 2) admission diagnosis (P = .01); 3) preexisting coronary artery disease (P = .004); 4) treatment with blood‐thinning medications, including warfarin (P = .0004), intravenous heparin (P = .0003), and clopidogrel (P = .02); and 5) treatment with PPIs (P = .02). After adjusting for the use of full‐dose anticoagulation and/or clopidogrel, the only of these risk factors that remained significantly associated with GIB was treatment with PPIs prior to hospitalization (adjusted OR = 2.1; 95% CI 1.17.0; P = .04), suggesting that PPI treatment in the outpatient setting may be a marker for GI vulnerability.

Among patients on warfarin, the peak international normalized ratio (median [IQR]) was 3.0 (1.25.0) for cases and 1.9 (1.64.8) for controls (P = .52). For those on heparin (23 cases and 6 controls), the median peak activated partial thromboplastin time (aPTT) was 67 (5082) and 128 (67180) seconds for cases and controls, respectively (P = .03), a surprising finding that was likely a result of type III error and small sample size.

Outcomes

We found no evidence of major complications from bleeding, as shown in Table 2. As expected, cases were more likely to receive blood transfusions than were controls, but clinically serious outcomes were uncommon in both groups.

Gastrointestinal Symptoms and Endoscopic Findings

Bleeding symptoms prompting EGD and associated endoscopic findings are shown in Table 3. Findings on colonoscopy (performed in 34 patients) are included. Overall, 54 (74%) patients had a detected abnormality on EGD and/or colonoscopy that was believed to be a likely source of bleeding by the endoscopist, and 19 (26%) had no apparent culprit lesions. Melena and stool positivity for occult blood were the most common manifestations of gastrointestinal bleeding (77%) and also accounted for all the normal endoscopic evaluations. Of the 21 ulcers, 18 (85.7%) had a clean base, 1 (4.8%) had a red spot, and 2 (9.5%) had an adherent clot. None had a bleeding vessel. Endoscopic treatment was performed in one patient and angiography in one patient. A possible gastric stromal tumor (not the source of bleeding) was seen in one patient, but no mucosal malignant lesions were identified. Of the 73 cases, 41 (56.2% of cases and 0.2% of the total cohort of 17,707 patients) had culprit lesions that might have been preventable with gastric acid suppression (including peptic ulcers, esophagitis, and duodenitis).

Prophylactic Gastric Acid Suppression

One hundred and two patients were not taking any acid‐suppressive prophylaxis on admission to the hospital. Of these patients, on admission 28 (27.5%) were prescribed either histamine‐2 receptor antagonists or proton pump inhibitors. We identified no clinical features associated with the prescriptions for these medications (Table 4), suggesting that physician preference, rather than perceived risk factors for bleeding, determined which patients received prophylactic acid blockade. There was no association between this prophylaxis and GI bleeding, but because of the small size of our sample, the confidence interval was wide (OR = 1.0; 95% CI 0.42.4; P = .97). In the analysis of the subgroup of patients receiving anticoagulation or clopidogrel, prophylaxis showed a nonsignificant trend toward benefit (OR = 0.71; 95% CI 0.23.9; P = .67). There was no significant interaction between the presence of anticoagulation or clopidogrel and prophylaxis (P = .61). Similarly, when we excluded those without prior GI bleeding from analysis, there was still no apparent protective effect of acid‐suppressive prophylaxis (OR = 1.0; 95% CI 0.42.5; P = .97). Finally, there was no significant association between the use of prophylaxis and lesions (theoretically) preventable by acid blockade (OR = 0.9; 95% CI 0.32.3; P = .84).

Study Patients

The institutional review board of the Cleveland Clinic Foundation (Cleveland, OH) approved this study. All patients admitted to the General Internal Medicine service between January 1, 1999, and December 31, 2002, were eligible for inclusion. Two types of cases were included: 1) patients admitted for nongastrointestinal illnesses who developed bleeding at least 24 hours after admission and required esophagogastroduodenoscopy (EGD) during hospitalization (designated in‐hospital bleeding), and 2) patients admitted with gastrointestinal bleeding (requiring EGD) who had been hospitalized on the General Medical service during the preceding 4 weeks for a nongastrointestinal illness (designated out‐of‐hospital bleeding). This second group was included to identify risk factors for delayed bleeding that might not be obvious during hospitalization.

Medical records of all General Medicine patients who underwent EGD were reviewed in a standardized fashion (Fig. 1). We excluded patients with documented gastrointestinal complaints (including bleeding) at the time of the index admission or within 24 hours of admission, bleeding in the intensive care unit (ICU) or in another hospital prior to transfer to the General Medicine service, or a history of gastrointestinal bleeding during the month prior to admission. ICU stay prior to General Medicine admission, if not associated with GI bleeding, was not an exclusion criterion for our study.

Figure 1

Controls, also without any acute gastrointestinal symptoms at admission, were randomly matched to cases in a 1:1 ratio by date of admission. We used this liberal matching strategy because any factors matched for would no longer be eligible to be risk factors for bleeding. If more than one control was admitted on the same day as a case, then a random number was used to select the control.

Definition of Prophylactic Acid Suppression

We defined prophylactic acid suppression as in‐hospital de novo treatment with histamine‐2 receptor antagonists and/or proton pump inhibitors received prior to the onset of any symptoms that would suggest GI bleeding (for cases) or any time during hospitalization (for controls). Patients taking these drugs prior to admission were deemed ineligible for in‐hospital prophylactic acid blockade and were excluded from the related analyses.

Data Collection

We extracted demographic information, medical history, medication usage, and laboratory data by chart review. For those patients readmitted for gastrointestinal bleeding following discharge, data from the initial (nongastrointestinal illnessassociated) hospitalization were recorded. Bleeding symptoms triggering endoscopy were grouped into four categories: 1) melena or hematochezia; 2) hematemesis (frank blood in vomitus or coffee‐grounds emesis); 3) melena or hematochezia plus hematemesis (both 1 and 2); 4) stool positivity for occult blood or unexplained drop in hemoglobin in the absence of overt bleeding. Endoscopic findings were categorized by the nature of the visualized lesions, and if multiple lesions were noted, the endoscopist's impression of the most likely bleeding site was used to define the source of bleeding. We recorded colonoscopy findings for patients undergoing this evaluation.

Statistical Analysis

We analyzed data utilizing JMP 5.1 (SAS Institute, Cary, NC). Random controls were chosen using computer‐generated random numbers. The proportions of patients with various categorical characteristics were compared using the chi‐square test or Fisher's exact test as appropriate. We used the Student t test or Wilcoxon's test to compare continuous variables. Odds ratios and adjusted odds ratios were calculated by logistic regression. Two‐tailed P values less than .05 were considered statistically significant.

RESULTS: Identification of Cases and Controls

Of 17,707 patients admitted to the General Medicine service, 1327 (7.5%) underwent EGD during hospitalization or within 1 month of discharge. Only 73 (0.41%) of the total number of patients met the case definition (Fig. 1). Of these cases, 62 (84.9%) had developed gastrointestinal bleeding during the index hospitalization, whereas 11 (15.1%) were readmitted for bleeding within 4 weeks of hospital discharge. The remaining 1254 patients who underwent EGD were excluded based on exclusion criteria, including an absence of documented bleeding prompting the EGD.

Clinical Risk Factors for Bleeding

In univariate analysis, as shown in Table 1, predictors of GIB included: 1) age (P = .02); 2) admission diagnosis (P = .01); 3) preexisting coronary artery disease (P = .004); 4) treatment with blood‐thinning medications, including warfarin (P = .0004), intravenous heparin (P = .0003), and clopidogrel (P = .02); and 5) treatment with PPIs (P = .02). After adjusting for the use of full‐dose anticoagulation and/or clopidogrel, the only of these risk factors that remained significantly associated with GIB was treatment with PPIs prior to hospitalization (adjusted OR = 2.1; 95% CI 1.17.0; P = .04), suggesting that PPI treatment in the outpatient setting may be a marker for GI vulnerability.

Among patients on warfarin, the peak international normalized ratio (median [IQR]) was 3.0 (1.25.0) for cases and 1.9 (1.64.8) for controls (P = .52). For those on heparin (23 cases and 6 controls), the median peak activated partial thromboplastin time (aPTT) was 67 (5082) and 128 (67180) seconds for cases and controls, respectively (P = .03), a surprising finding that was likely a result of type III error and small sample size.

Outcomes

We found no evidence of major complications from bleeding, as shown in Table 2. As expected, cases were more likely to receive blood transfusions than were controls, but clinically serious outcomes were uncommon in both groups.

Gastrointestinal Symptoms and Endoscopic Findings

Bleeding symptoms prompting EGD and associated endoscopic findings are shown in Table 3. Findings on colonoscopy (performed in 34 patients) are included. Overall, 54 (74%) patients had a detected abnormality on EGD and/or colonoscopy that was believed to be a likely source of bleeding by the endoscopist, and 19 (26%) had no apparent culprit lesions. Melena and stool positivity for occult blood were the most common manifestations of gastrointestinal bleeding (77%) and also accounted for all the normal endoscopic evaluations. Of the 21 ulcers, 18 (85.7%) had a clean base, 1 (4.8%) had a red spot, and 2 (9.5%) had an adherent clot. None had a bleeding vessel. Endoscopic treatment was performed in one patient and angiography in one patient. A possible gastric stromal tumor (not the source of bleeding) was seen in one patient, but no mucosal malignant lesions were identified. Of the 73 cases, 41 (56.2% of cases and 0.2% of the total cohort of 17,707 patients) had culprit lesions that might have been preventable with gastric acid suppression (including peptic ulcers, esophagitis, and duodenitis).

Prophylactic Gastric Acid Suppression

One hundred and two patients were not taking any acid‐suppressive prophylaxis on admission to the hospital. Of these patients, on admission 28 (27.5%) were prescribed either histamine‐2 receptor antagonists or proton pump inhibitors. We identified no clinical features associated with the prescriptions for these medications (Table 4), suggesting that physician preference, rather than perceived risk factors for bleeding, determined which patients received prophylactic acid blockade. There was no association between this prophylaxis and GI bleeding, but because of the small size of our sample, the confidence interval was wide (OR = 1.0; 95% CI 0.42.4; P = .97). In the analysis of the subgroup of patients receiving anticoagulation or clopidogrel, prophylaxis showed a nonsignificant trend toward benefit (OR = 0.71; 95% CI 0.23.9; P = .67). There was no significant interaction between the presence of anticoagulation or clopidogrel and prophylaxis (P = .61). Similarly, when we excluded those without prior GI bleeding from analysis, there was still no apparent protective effect of acid‐suppressive prophylaxis (OR = 1.0; 95% CI 0.42.5; P = .97). Finally, there was no significant association between the use of prophylaxis and lesions (theoretically) preventable by acid blockade (OR = 0.9; 95% CI 0.32.3; P = .84).

Study Patients

The institutional review board of the Cleveland Clinic Foundation (Cleveland, OH) approved this study. All patients admitted to the General Internal Medicine service between January 1, 1999, and December 31, 2002, were eligible for inclusion. Two types of cases were included: 1) patients admitted for nongastrointestinal illnesses who developed bleeding at least 24 hours after admission and required esophagogastroduodenoscopy (EGD) during hospitalization (designated in‐hospital bleeding), and 2) patients admitted with gastrointestinal bleeding (requiring EGD) who had been hospitalized on the General Medical service during the preceding 4 weeks for a nongastrointestinal illness (designated out‐of‐hospital bleeding). This second group was included to identify risk factors for delayed bleeding that might not be obvious during hospitalization.

Medical records of all General Medicine patients who underwent EGD were reviewed in a standardized fashion (Fig. 1). We excluded patients with documented gastrointestinal complaints (including bleeding) at the time of the index admission or within 24 hours of admission, bleeding in the intensive care unit (ICU) or in another hospital prior to transfer to the General Medicine service, or a history of gastrointestinal bleeding during the month prior to admission. ICU stay prior to General Medicine admission, if not associated with GI bleeding, was not an exclusion criterion for our study.

Figure 1

Controls, also without any acute gastrointestinal symptoms at admission, were randomly matched to cases in a 1:1 ratio by date of admission. We used this liberal matching strategy because any factors matched for would no longer be eligible to be risk factors for bleeding. If more than one control was admitted on the same day as a case, then a random number was used to select the control.

Definition of Prophylactic Acid Suppression

We defined prophylactic acid suppression as in‐hospital de novo treatment with histamine‐2 receptor antagonists and/or proton pump inhibitors received prior to the onset of any symptoms that would suggest GI bleeding (for cases) or any time during hospitalization (for controls). Patients taking these drugs prior to admission were deemed ineligible for in‐hospital prophylactic acid blockade and were excluded from the related analyses.

Data Collection

We extracted demographic information, medical history, medication usage, and laboratory data by chart review. For those patients readmitted for gastrointestinal bleeding following discharge, data from the initial (nongastrointestinal illnessassociated) hospitalization were recorded. Bleeding symptoms triggering endoscopy were grouped into four categories: 1) melena or hematochezia; 2) hematemesis (frank blood in vomitus or coffee‐grounds emesis); 3) melena or hematochezia plus hematemesis (both 1 and 2); 4) stool positivity for occult blood or unexplained drop in hemoglobin in the absence of overt bleeding. Endoscopic findings were categorized by the nature of the visualized lesions, and if multiple lesions were noted, the endoscopist's impression of the most likely bleeding site was used to define the source of bleeding. We recorded colonoscopy findings for patients undergoing this evaluation.

Statistical Analysis

We analyzed data utilizing JMP 5.1 (SAS Institute, Cary, NC). Random controls were chosen using computer‐generated random numbers. The proportions of patients with various categorical characteristics were compared using the chi‐square test or Fisher's exact test as appropriate. We used the Student t test or Wilcoxon's test to compare continuous variables. Odds ratios and adjusted odds ratios were calculated by logistic regression. Two‐tailed P values less than .05 were considered statistically significant.

RESULTS: Identification of Cases and Controls

Of 17,707 patients admitted to the General Medicine service, 1327 (7.5%) underwent EGD during hospitalization or within 1 month of discharge. Only 73 (0.41%) of the total number of patients met the case definition (Fig. 1). Of these cases, 62 (84.9%) had developed gastrointestinal bleeding during the index hospitalization, whereas 11 (15.1%) were readmitted for bleeding within 4 weeks of hospital discharge. The remaining 1254 patients who underwent EGD were excluded based on exclusion criteria, including an absence of documented bleeding prompting the EGD.

Clinical Risk Factors for Bleeding

In univariate analysis, as shown in Table 1, predictors of GIB included: 1) age (P = .02); 2) admission diagnosis (P = .01); 3) preexisting coronary artery disease (P = .004); 4) treatment with blood‐thinning medications, including warfarin (P = .0004), intravenous heparin (P = .0003), and clopidogrel (P = .02); and 5) treatment with PPIs (P = .02). After adjusting for the use of full‐dose anticoagulation and/or clopidogrel, the only of these risk factors that remained significantly associated with GIB was treatment with PPIs prior to hospitalization (adjusted OR = 2.1; 95% CI 1.17.0; P = .04), suggesting that PPI treatment in the outpatient setting may be a marker for GI vulnerability.

Among patients on warfarin, the peak international normalized ratio (median [IQR]) was 3.0 (1.25.0) for cases and 1.9 (1.64.8) for controls (P = .52). For those on heparin (23 cases and 6 controls), the median peak activated partial thromboplastin time (aPTT) was 67 (5082) and 128 (67180) seconds for cases and controls, respectively (P = .03), a surprising finding that was likely a result of type III error and small sample size.

Outcomes

We found no evidence of major complications from bleeding, as shown in Table 2. As expected, cases were more likely to receive blood transfusions than were controls, but clinically serious outcomes were uncommon in both groups.

Gastrointestinal Symptoms and Endoscopic Findings

Bleeding symptoms prompting EGD and associated endoscopic findings are shown in Table 3. Findings on colonoscopy (performed in 34 patients) are included. Overall, 54 (74%) patients had a detected abnormality on EGD and/or colonoscopy that was believed to be a likely source of bleeding by the endoscopist, and 19 (26%) had no apparent culprit lesions. Melena and stool positivity for occult blood were the most common manifestations of gastrointestinal bleeding (77%) and also accounted for all the normal endoscopic evaluations. Of the 21 ulcers, 18 (85.7%) had a clean base, 1 (4.8%) had a red spot, and 2 (9.5%) had an adherent clot. None had a bleeding vessel. Endoscopic treatment was performed in one patient and angiography in one patient. A possible gastric stromal tumor (not the source of bleeding) was seen in one patient, but no mucosal malignant lesions were identified. Of the 73 cases, 41 (56.2% of cases and 0.2% of the total cohort of 17,707 patients) had culprit lesions that might have been preventable with gastric acid suppression (including peptic ulcers, esophagitis, and duodenitis).

Prophylactic Gastric Acid Suppression

One hundred and two patients were not taking any acid‐suppressive prophylaxis on admission to the hospital. Of these patients, on admission 28 (27.5%) were prescribed either histamine‐2 receptor antagonists or proton pump inhibitors. We identified no clinical features associated with the prescriptions for these medications (Table 4), suggesting that physician preference, rather than perceived risk factors for bleeding, determined which patients received prophylactic acid blockade. There was no association between this prophylaxis and GI bleeding, but because of the small size of our sample, the confidence interval was wide (OR = 1.0; 95% CI 0.42.4; P = .97). In the analysis of the subgroup of patients receiving anticoagulation or clopidogrel, prophylaxis showed a nonsignificant trend toward benefit (OR = 0.71; 95% CI 0.23.9; P = .67). There was no significant interaction between the presence of anticoagulation or clopidogrel and prophylaxis (P = .61). Similarly, when we excluded those without prior GI bleeding from analysis, there was still no apparent protective effect of acid‐suppressive prophylaxis (OR = 1.0; 95% CI 0.42.5; P = .97). Finally, there was no significant association between the use of prophylaxis and lesions (theoretically) preventable by acid blockade (OR = 0.9; 95% CI 0.32.3; P = .84).