Incidence, Risk Factors, and Clinical Effects of Recurrent Diverticular Hemorrhage: A Large Cohort Study

Abstract

Background & Aims:

Although recurrent diverticular hemorrhage is common, its incidence and risk factors have not been measured outside of small institutional cohorts. We analyzed the incidence of and risk factors for recurrent diverticular hemorrhage and whether discontinuing anticoagulation after diverticular hemorrhage is associated with ischemic stroke.

Methods:

We performed a retrospective cohort study of patients enrolled in the OptumInsight Clinformatics database from 2000 through 2016. Incidence rates for initial and recurrent diverticular hemorrhage were calculated by identifying patients who had hospitalizations with a primary discharge diagnosis consistent with diverticular hemorrhage. The hazard ratios of second diverticular hemorrhage associated with anticoagulants or platelet aggregation inhibitors were calculated using Cox proportional hazards regression adjusted for demographics, comorbidities, and medication use. The hazard ratio for ischemic stroke among patients who discontinued anticoagulation after diverticular hemorrhage was calculated similarly.

Results:

In the cohort analyzed, 14,925 patients had an initial diverticular hemorrhage; 1368 of these patients had a second episode. The unstandardized incidence rates of initial and second diverticular hemorrhage were 10.9 per 100,000 person-years (95% CI, 10.7 – 11.0) and 3625.6 per 100,000 person-years (95% CI, 3436.0 – 3823.0). Platelet aggregation inhibitors were associated with second episodes of diverticular hemorrhage (hazard ratio, 1.47; 95% CI, 1.151.88), whereas all classes of anticoagulation agents were not associated. Among patients with a potential indication for stroke prophylaxis, those who discontinued anticoagulation after the diverticular hemorrhage had an increased hazard of ischemic stroke (hazard ratio, 1.93; 95% CI, 1.17–3.19).

Conclusions:

In this retrospective cohort study, platelet aggregation inhibitors, but not anticoagulants, were associated with recurrent diverticular hemorrhage. Discontinuing anticoagulation was associated with increased hazard for ischemic stroke.

Graphical Abstract

Introduction

Colonic diverticular hemorrhage is the most common cause of gastrointestinal bleeding distal to the stomach.¹ Diverticular hemorrhage develops when the artery at the base of the diverticulum ruptures due to erosion of the vessel wall.² Overall, diverticular hemorrhage is common, with a prevalence rate of 23.9 to 32.5 per 100,000 persons^1,3 based on cross-sectional analyses of hospital-level discharge data in the United States.

Patients with an initial episode of diverticular hemorrhage are presumed to be at increased risk for recurrent episodes of hemorrhage, but the incidence of a second episode has not been systematically evaluated. Current estimates of the proportion of patients who experience recurrent hemorrhage are derived from institutional cohorts and vary widely from 13.8 – 47.4%.^4–12 Several risk factors, including aspirin use, non-steroidal anti-inflammatory drug (NSAID) use,^13–17 obesity,¹⁸ hypertension,¹⁷ and atherosclerosis,¹⁹ have been associated with initial diverticular hemorrhage. There are limited data on the effect of platelet aggregation inhibitors (PAIs)⁸ and anticoagulants¹⁷ on diverticular hemorrhage, but they are thought to increase risk, like they do in other causes of gastrointestinal bleeding.²⁰

A limitation of the existing literature is that it is difficult to isolate the effect of the identified risk factors for diverticular hemorrhage from their association with the development of diverticulosis. Disentangling this relationship is challenging because diverticular hemorrhage cannot occur without pre-existing diverticulosis, diverticulosis is clinically silent until complications develop, and the relationship between risk factors, such as atherosclerosis and prescription drug use, with diverticulosis is confounded by aging.²¹ Some studies have attempted to overcome these challenges by matching patients with diverticular hemorrhage to controls with silent diverticulosis, but given that they are derived from institutional cohorts, they have not had enough power to make definitive conclusions about the range of potential risk factors.¹⁷

It is important to better understand the risk factors of diverticular hemorrhage because hospitalization for diverticular hemorrhage is associated with decompensation of comorbid conditions and increased 30-day mortality.²² For example, given the significant burden of cardiovascular comorbidities among patients with diverticular hemorrhage,^{17, 22} precisely calculating the relative hazard of PAIs and anticoagulant medications for recurrent diverticular hemorrhage is necessary to inform decision making about whether the increased risk of hemorrhage outweighs the risk of thromboembolism.^{23, 24} To address this question, we conducted a cohort study among patients with a history of a first diverticular hemorrhage using OptumInsight Clinformatics (Optum), a patient-level claims database from a national health insurer in the United States. Using this cohort of patients with known diverticulosis, we determined the incidence of recurrent diverticular hemorrhage, estimated the relative hazard of PAIs and anticoagulants for recurrent diverticular hemorrhage, and assessed the risk of ischemic stroke among patients who discontinue anticoagulation secondary to initial diverticular hemorrhage.

Methods

Study design and data source

The Optum database was used for this retrospective cohort study. Optum is a patient-level medical claims database consisting of the inpatient, outpatient, pharmacy, procedure, and laboratory claims of approximately 77.2 million unique patients. Data from Optum have previously been used to study acute conditions²⁵ and medication use.^{26, 27} Optum contains inpatient diagnoses and procedures from 2004 – 2016 in addition to outpatient diagnoses, procedures, and prescriptions from 2000 – 2016. For this study, records after September 30, 2016 were excluded due to the possibility of incomplete claims at the time of data download (July 2017). Optum contains claims from patients co-insured with Medicaid from 2000 – 2010 and Medicare from 2000 – 2005. For encounters before October 1,2015, diagnoses are classified using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes, and ICD-10 codes are used thereafter. Research using Optum has been classified as exempt by the Institutional Review Board of the University of Pennsylvania.

Study population

Patients who had experienced an initial episode of diverticular hemorrhage during follow-up in the database were included in the cohort if they all of met the following criteria:

1. ICD diagnosis for diverticulosis of colon with hemorrhage as primary inpatient hospitalization discharge diagnosis at least 120 days after entry into the database. (ICD-9-CM code 562.12 - diverticulosis of colon with hemorrhage or ICD-10 code K57.31 - diverticulosis of the large intestine without perforation or abscess with bleeding).

2. No ICD diagnosis for diverticulosis of colon with hemorrhage in any inpatient or outpatient claim in any diagnostic position prior to date of service in criterion A

3. No ICD diagnosis for inflammatory bowel disease or colorectal cancer before or within two years of the date of initial diverticular hemorrhage. These diagnoses were excluded to reduce the probability of including cases of lower gastrointestinal bleeding that may have been misclassified as diverticular hemorrhage (ICD codes listed in Supplementary Table 1).

4. At least one-year of follow-up in Optum without any gaps in enrollment.

Diverticular hemorrhage is often a presumed clinical diagnosis because active diverticular bleeding is rarely witnessed during endoscopy. Thus, there is no gold standard for validation studies. To reduce the possibility of misclassification of the outcome of diverticular hemorrhage we excluded ICD-9-CM code 562.13 (diverticulitis of colon with hemorrhage) and the corresponding ICD-10 code K57.33 (diverticulitis of the large intestine without perforation or abscess with bleeding) from the main analysis. ICD-9-CM code 562.12 (diverticulosis of the colon with hemorrhage) has previously been shown to have 97% positive predictive value for lower gastrointestinal hemorrhage when it is the primary hospital diagnosis.²⁸ The minimum 120-day interval between enrollment and initial diverticular hemorrhage was incorporated into the inclusion criteria provide an opportunity for a patient’s pre-existing comorbidities at the time of enrollment in the database to be recorded.²⁹

Exposure definition

The primary exposures of interest were prescriptions for anticoagulants and platelet aggregation inhibitors (PAIs). Anticoagulant use was recorded as a categorical variable with levels of (1) no anticoagulation, (2) direct-acting oral anticoagulants (apixiban, dabigatran, and rivaroxaban), (3) warfarin, and (4) subcutaneous heparin derivatives (heparin, dalteparin, and enoxaparin). The three adenosine diphosphate receptor inhibitors widely available in the United States (clopidogrel, prasugrel, and ticagrelor) were included in the PAI class. Exposures were identified using the pharmacy claims file. Prescription use at the time of initial diverticular hemorrhage was determined by presence of a prescription claim with days of medication supplied that overlapped with the admission date. Thereafter in follow-up, prescription status was incorporated into the analysis as a time-varying exposure that updated daily based on the days of medication supplied and refill status. For example, if a patient claimed a 30-day supply of a medication on March 1, he or she was considered exposed to the medication until March 31. If he or she next claimed a 30-day supply of the medication on April 3, he or she was considered unexposed on April 1 and April 2, but exposed from April 3 to May 3. If there were prescriptions for two anticoagulants dispensed on a single date, the medication recorded in the database second was excluded. This accounted for 576 instances of a total of 102,732 prescriptions (0.56%). Use of aspirin, NSAIDs, and selective serotonin reuptake inhibitors (SSRIs) were recorded as confounding variables.^30–32 We used sensitivity analyses (described below) to assess the impact of over-the-counter aspirin and NSAID use.

Outcome definition

The main study outcome was second episode of diverticular hemorrhage, identified by an ICD diagnosis code²⁸ for diverticulosis of the colon with hemorrhage in the primary position during an inpatient hospitalization at least 30 days after the initial diverticular hemorrhage hospitalization. The 30-day interval was specified to identify distinct episodes of diverticular hemorrhage as opposed to identifying recrudescence of the initial episode. A sensitivity analysis was performed to assess the impact of this definition on the estimated incidence rates (described below).

Other covariates

Demographic covariates collected included patient age, sex, and race (documented in Optum as White, Black, Hispanic, Asian, and unknown). Diagnoses of interest were selected due to their association with diverticular hemorrhage or with medication exposures and were identified by the presence of at least one ICD diagnosis code in any position in any inpatient or outpatient claim. The full list of comorbidities is presented in Supplementary Methods and Supplementary Table 1. All covariates were incorporated into the Cox proportional hazards model as time-updating variables. Age updated yearly, and diagnoses updated unidirectionally daily. New comorbid diagnoses were recorded so long as they occurred at least one day prior to the episode of recurrent diverticular hemorrhage. See Supplementary Figure 1 for a directed acyclic graph illustrating the relationships between the study exposures, outcome, and covariates.

Statistical analysis

Stata version 15 (College Station, TX, USA) was used for all statistical analyses. Continuous data are reported as medians with interquartile range (IQR). Categorical data are reported as counts and percentages. Results with fewer than 20 subjects are not reported to avoid re-identification.

Rates of diverticular hemorrhage.

For the incidence rate of the initial episode of diverticular hemorrhage, the total follow-up time was calculated from the entire population of Optum that had at least one-year follow-up without gaps in enrollment. Incidence rates are primarily reported as unstandardized rates stratified by decade of life because the prevalence of diverticulosis increases with age and patients under the age of 50 are at low risk for diverticular complications.²¹ Incidence rates directly age- and sex-standardized to the 2010 United States population³³ are also presented for comparison to other studies. Unadjusted cumulative incidence rates for the second episode of diverticular hemorrhage were calculated using Kaplan-Meier survival analysis.

To compare the rate of diverticular hemorrhage with the two prior studies that assessed the occurrence of diverticular hemorrhage using administrative hospital-level discharge data,^1,3 prevalence rates were calculated using the total number of patients enrolled in Optum from January 1, 2000 - September 30, 2016. The prevalence rates are reported as both unstandardized rates and rates directly age- and sex-standardized to the 2010 United States population.

Risk factors for recurrent diverticular hemorrhage.

Adjusted hazard ratios for risk factors for recurrent diverticular hemorrhage were calculated using Cox proportional hazards regression. The model was adjusted for demographics and comorbid diagnoses using a disease risk score for diverticular hemorrhage to balance covariates among the exposure groups. Details of the disease-risk score creation and its incorporation into the Cox model are presented in Supplementary Methods. The Cox model was adjusted for use of aspirin, NSAIDs, SSRIs, PAIs, DOACs, warfarin, heparin, the type of anticoagulant at initial diverticular hemorrhage, and the disease risk score. The initial Cox model included an interaction term for anticoagulation type and chronic kidney disease, because chronic kidney disease may influence the pharmacologic effect of anticoagulants due to altered pharmacokinetics. The interaction term was not statistically significant by the Wald test, so it was removed from subsequent models. The Cox proportional hazards model was stratified with homogeneous covariate effects before and after the date of first DOAC approval by the Food and Drug Administration (October 19, 2010). Censoring occurred if the patient was no longer enrolled in the database or underwent colectomy (codes in Supplementary Table 2, Panel 1). As the goal of the analysis was to create an explanatory model adjusting for all potential confounders, covariates that were not statistically significant in the main model were not eliminated. Violations of the proportional hazards assumption were assessed using Schoenfeld residuals. The hazard of recurrent diverticular hemorrhage when using multiple medications simultaneously was assessed through linear combinations from the Stata lincom command.

Sensitivity analyses

To assess the robustness of the study results, several additional analyses were performed. First, because the two previously published studies that calculated the rates of diverticular hemorrhage using administrative data used different ICD-9-CM criteria for the definition of diverticular hemorrhage (562.12 only³ versus 562.12 and 562.13¹), incidence rates and risk factors for recurrent hemorrhage were also calculated in a sensitivity analysis where diverticular hemorrhage was identified by an expanded list of ICD codes (Supplementary Table 3). Second, to determine the impact of excluding episodes of recurrent diverticular hemorrhage that occurred within 30 days of the prior diverticular hemorrhage, we calculated incidence rates in which these episodes were included in the analysis. Third, to determine if anticoagulation or PAIs were associated with severe episodes of diverticular hemorrhage, an analysis where the outcome was diverticular hemorrhage requiring a red blood cell transfusion during the recurrent hemorrhage was conducted (ICD codes for transfusion in Supplementary Table 4). In this analysis, requirement of blood transfusion was considered a proxy for severity of hemorrhage. Fourth, to reduce the impact of other causes of lower gastrointestinal hemorrhage misclassified as diverticular hemorrhage, an analysis where diverticular hemorrhage with documented lower intestinal endoscopy was considered the outcome (ICD-procedure codes for lower intestinal endoscopy in Supplementary Table 5). Finally, to assess whether the hazard ratio estimates for anticoagulation were sensitive to misclassification of aspirin and NSAID use, simulation studies varying aspirin and NSAID exposure were performed (detailed in Supplementary Methods).

Thrombotic sequelae of anticoagulation discontinuation

To assess the clinical impact of discontinuing anticoagulation after the initial episode of diverticular hemorrhage, an analysis assessing the hazard of ischemic stroke among patients who were taking anticoagulation and had a potential indication for stroke prophylaxis at the time of initial diverticular hemorrhage was performed. The cohort was formed from the group of patients who were identified using the expanded diverticular hemorrhage ICD list. Indications for stroke prophylaxis were defined as ICD codes for atrial fibrillation, atrial flutter, hypercoagulable state, and heart valve replacement (ICD codes in Supplementary Table 1). The outcome was a diagnosis of ischemic stroke in the primary diagnostic position (ICD-9 codes 434.x and ICD-10 codes I63.x) during an inpatient encounter. The validity of these codes has been previously described.³⁴ This cohort was selected to study the impact of discontinuing anticoagulation because its members often have strong and persistent indications for anticoagulation.^35–37 Indications for anticoagulation such as deep venous thrombosis, pulmonary embolism, and post-surgical venous thromboembolism prophylaxis were not considered because the duration of therapy is variable and difficult to ascertain in administrative data.

Unlike the main analysis of recurrent diverticular bleeding, anticoagulation status was recorded as exposure to any anticoagulant. Patients were categorized into the following two groups to describe the sequence of their anticoagulant use before and after the initial episode of diverticular hemorrhage: anticoagulated to none (i.e. discontinued) and anticoagulated to anticoagulated (i.e. continued). Anticoagulation group was a time-varying exposure that updated daily based on the date of anticoagulation prescription and the days of medication dispensed. Cox proportional hazards regression was used to compare the hazard ratios of patients who discontinued versus continued anticoagulation, adjusted for aspirin use, NSAID use, age, and CHADS₂-VASc score.³⁸ Exposure to PAIs was not used for adjustment because no ischemic strokes occurred during exposure to PAIs. Patients were censored upon second diverticular hemorrhage or leaving the database. We also calculated the hazard ratio of recurrent diverticular hemorrhage for use of any anticoagulant among this cohort to compare this risk to the risk of ischemic stroke from discontinuing anticoagulation.

To help ensure that the results of the analysis were not due to residual confounding, a falsification test was performed to assess the hazard of discontinuing anticoagulation with new diagnosis of osteoarthritis after the initial diverticular bleeding episode. There is no known association between anticoagulation and osteoarthritis.

Results

Cohort characteristics

14,925 patients who met the inclusion criteria and experienced an initial episode of diverticular hemorrhage during follow-up in the database were identified (Figure 1). The characteristics of the cohort at the time of initial diverticular hemorrhage are presented in Table 1. The median age was 78 (IQR 70 – 82), and 50.4% were female. The median follow-up time was 2.0 years (IQR 0.9 – 4.0). The proportion of ICD codes used to identify the cohort are presented in Supplementary Tables 6 and 7. Medication refilling patterns among patients in the cohort are presented in Supplementary Table 8. Relative to the proportion of patients using each type of medication prior to the initial diverticular bleeding episode, the proportion using each medication after the episode was nearly unchanged (Supplementary Table 9). Of the 14,925 patients who experienced initial diverticular hemorrhage, 26 (0.2%) had arterial embolization during hospitalization and 310 (2.1%) underwent colectomy.

Figure 1.

Study cohort creation

Table 1.

Cohort characteristics at date of initial diverticular hemorrhage (n = 14,925)

	median	IQR
Age	78	70 - 82
	n	%
Sex
Female	7,748	51.9
Male	7,177	48.1
Total	14,925	100.0
Race
White	9,739	65.2
Black	2,385	16.0
Hispanic	1,258	8.4
Asian	277	1.9
Unknown	1,266	8.5
Total	14,925	100.0
Anticoagulation
None	13,286	84.9
DOACs	237	1.4
Warfarin	1,185	12.2
Heparin	217	1.5
Total	14,925	100.0
Prescriptions
Aspirin	167	1.1
NSAIDs	3,184	21.3
SSRIs	1,762	11.8
Platelet aggregation inhibitors	847	5.7
Comorbidities
Atrial fibrillation	4,095	27.4
Atrial flutter	733	4.9
Carotid artery stenosis	2,340	15.7
Chronic kidney disease	3,520	23.6
Cirrhosis	198	1.3
Congestive heart failure	4,389	29.4
Coronary artery disease	6,131	41.1
Heart valve replacement	447	3.0
Hypercoagulable state	158	1.1
Hypertension	12,914	86.5
Leukemia	126	0.8
Lymphoma	175	1.2
Osteoarthritis	6,966	46.7
Peripheral artery disease	4,072	27.3
Pulmonary embolism	535	3.6
Rheumatoid arthritis	706	4.7
Stroke	2,702	18.1
Venous thromboembolism	1,110	7.4

Abbreviations: DOACs - direct oral anticoagulants, NSAIDs - non-steroidal anti-inflammatory drugs, PAIs - platelet aggregation inhibitors, SSRIs - selective serotonin reuptake inhibitors

Rates of diverticular hemorrhage

In Optum, the unstandardized and age- and sex-standardized incidence rates (IR) of initial diverticular hemorrhage were 10.9 (95% CI 10.7 – 11.0) and 18.9 (95% CI 18.9 – 18.9) per 100,000 person-years respectively. The unstandardized IR of initial hemorrhage increased as the population aged, ranging from 0.3 (95% CI 0.2 – 0.3) per 100,000 person-years in those aged 30 – 39 to 49.6 (95% CI 48.4 – 50.9) per 100,000 person-years in those 80 and older (Table 2). To compare the rates of diverticular hemorrhage derived from Optum with those from the other studies that used administrative data,^{1, 3} the prevalence rate of initial diverticular hemorrhage was also calculated. The unstandardized prevalence of initial diverticular hemorrhage was 19.6 (95% CI 19.2 – 19.9) per 100,000 persons and the age- and sex-standardized prevalence was 36.3 (95% CI 35.7 – 37.0) per 100,000 persons.

Table 2.

Incidence rates (IR) of initial and second episodes of diverticular hemorrhage per 100,000 person-years stratified by age-group at time of initial episode of diverticular hemorrhage

	Initial diverticular hemorrhage (n = 38.5 million, 142.1 million person- years)			Second diverticular hemorrhage (n = 14,925, 37,705.8 person- years)
Age group	Events	IR	95% CI	Events	IR	95% CI
30 – 39	37	0.3	0.2 – 0.3	--	--	--
40 – 49	267	1.4	1.2 – 1.6	--	--	--
50 – 59	978	4.5	4.2 – 4.8	94	2,614.4	2,135.9 – 3,200.1
60 – 69	2,475	12.3	11.8 – 12.8	316	3,233.5	2,895.9 – 3,610.4
70 – 79	5,382	36.4	35.4 – 37.4	711	3,932.9	3,654.1 – 4,232.8
80 and older	6,164	49.6	48.4 – 50.9	229	4,792.1	4,210 – 5,454.8

Note: Cells with counts less than 20 are suppressed to avoid re-identification.

Among those with an initial hemorrhage, the unstandardized IR for a second episode was 3,625.6 (95% CI 3,436.0 – 3,823.0), and the unstandardized IR increased from 2,614.4 (95% CI 2,135.9 – 3,200.1) per 100,000 person-years to 4,792.1 (95% CI 4,210.0 – 5,454.8) per 100,000 person-years as age-group at initial hemorrhage increased from 50 – 59 years to 80 years and older (Table 2). The unstandardized IR of a third episode of diverticular hemorrhage was 10,848.3 (95% CI 9,625.2 – 12,183.7) per 100,000 person-years. The IRs of each subsequent episode were progressively higher (Table 3).

Table 3.

Incidence rates (IR) of diverticular hemorrhage per 100,000 person-years

		Unstandardized		Standardized
Episode	Events	IR	95% CI	IR	95% CI
1	14,925	10.9	10.7 - 11.0	18.9	18.9 - 18.9
2	1,368	3,625.6	3,436.0 - 3,823.0	109.1	103.4 - 115.1
3	285	10,848.3	9,625.2 - 12,183.7	314.0	278.6 - 352.6
4	76	16,441.3	12,953.8 - 20,578.7	486.2	383.1 - 608.6
5	28	24,970.1	16,592.4 - 36,088.7	718.4	477.4 - 1038.3

Note: The IR for each episode is conditional on having a prior episode of diverticular bleeding. For example, among patients who have previously experienced 2 episodes of diverticular bleeding, the unstandardized IR of a 3^rd episode is 10,848.3 per 100,000 person-years.

Characteristics of second episode of diverticular hemorrhage

Among patients experiencing an initial episode of diverticular hemorrhage, 1,368 (9.2%) experienced a second episode. The median time between the first and second hemorrhage was 1.2 years (IQR 0.5 – 2.5). In the unadjusted Kaplan-Meier analysis, the cumulative incidence of a second episode of diverticular hemorrhage was 4.7% (95% CI 4.3 – 5.0%) at one year. At two and five years, the cumulative incidences were 8.3% (95% CI 7.8 – 8.9%) and 15.7% (95% CI 14.8 – 16.6%) respectively (Figure 2). During the hospitalization for the second episode of diverticular hemorrhage, 707 (51.7%) received a transfusion. Of the 1,368 patients who had a second episode of diverticular hemorrhage, 978 (71.5%) underwent colonoscopy or flexible sigmoidoscopy during the second hospitalization. Among the patients who had a second hemorrhage, 719 (52.5%) underwent lower endoscopy during the first and second hospitalizations and 232 (17.0%) underwent lower endoscopy during the second admission only. Of the 1,368 patients who experienced a second diverticular hemorrhage, 8 (0.6%) underwent arterial embolization and 64 (4.7%) underwent colectomy.

Figure 2.

Unadjusted Kaplan-Meier survival curve for second episode of diverticular hemorrhage among patients with an initial episode of diverticular hemorrhage.

Risk factors for second diverticular hemorrhage

The overall exposure time to anticoagulants was 522.9 person-years for DOACs, 3,349.9 person-years for warfarin, 680.7 person-years for subcutaneous heparin, and 33,152.2 person-years for no anticoagulation. The overall exposure time to SSRIs and PAIs was 4,733.2 and 1,174.0 person-years respectively. PAI exposure was significantly associated with second diverticular hemorrhage (HR 1.47, 95% CI 1.15 – 1.88), whereas each category of anticoagulation was not (Table 4).

Table 4.

Risk factors for second diverticular hemorrhage

	HR	95% CI
Exposures of interest
Platelet aggregation inhibitors	1.47	1.15 – 1.88 *
Anticoagulants
DOACs	0.68	0.33 – 1.41
Warfarin	0.95	0.74 – 1.22
Heparins	1.28	0.78 – 2.08
Confounding covariates
Aspirin	1.25	0.82 – 1.91
NSAIDs	1.02	0.90 – 1.16
SSRIs	0.93	0.78 – 1.10
Anticoagulants at 1st bleed
DOACs	0.23	0.05 – 0.96
Warfarin	0.95	0.76 – 1.20
Heparins	1.08	0.60 – 1.95
Disease risk score	1.07	1.05 – 1.08 *

Note:

^*indicates p < 0.05.

Abbreviations: CI - confidence interval, DOACs - direct oral anticoagulants, HR - hazard ratio, NSAIDs - non-steroidal anti-inflammatory drugs, SSRIs - selective serotonin reuptake inhibitors.

In the analysis of simultaneous exposure to more than one class of medication, the hazard ratio of using PAIs with aspirin was 1.84 (95% CI 1.13 – 3.00). Other statistically significant combinations included PAIs with heparin (HR 1.88, 95% CI 1.08 – 3.25), PAIs with aspirin and warfarin (HR 1.73, 95% CI 1.02 – 3.02), and PAIs with aspirin and heparin (HR 2.34, 95% CI 1.17 – 4.70) (Supplementary Table 10).

Sensitivity analyses

In the sensitivity analysis using the expanded ICD code list to identify episodes of diverticular hemorrhage, 17,212 patients with initial diverticular hemorrhage were identified. The unstandardized IR was 12.6 (95% CI 12.4 – 12.8) per 100,000 person-years, compared to 10.9 (95% CI 10.7 – 11.0) per 100,000 person-years in the primary analysis. The age- and sex-adjusted IR was 22.0 (95% CI 22.0 – 22.0) per 100,000 person-years, compared to 18.9 (95% CI 18.9 – 18.9) in the primary analysis (Supplementary Table 11, Panel 1). Statistically significant risk factors for recurrent diverticular hemorrhage were unchanged from the primary analysis (Supplementary Table 12).

In the sensitivity analysis including recurrent episodes of diverticular hemorrhage that occurred within 30 days of prior episode, the unstandardized IR of a second diverticular hemorrhage was 4,592.6 (95% CI 4,377.2 – 4,815.8) compared to 3,625.6 (95% CI 3,436.0 – 3,823.0) in the main analysis. The incidence rates of subsequent diverticular hemorrhage episodes in this analysis are presented in Supplementary Table 11, Panel 2. In the unadjusted Kaplan-Meier analysis, the cumulative incidence of a second episode of diverticular hemorrhage at one, two, and five years was 7.0% (95% CI 6.6 – 7.5%), 10.6% (95% CI 10.0 – 11.1%), and 17.8% (95% CI 16.8 – 18.7%) respectively. In the analysis considering second diverticular hemorrhage with blood transfusion as the outcome, only PAI therapy (HR 1.45, 95% CI 1.02 – 2.04) was associated (Supplementary Table 13). In the sensitivity analysis considering diverticular hemorrhage with endoscopy as the outcome, again, only PAI therapy (HR 1.49, 95% CI 1.06 – 2.11) was associated (Supplementary Table 14). These results were consistent with the primary analysis.

In the sensitivity analyses assessing the impact of over-the-counter aspirin and NSAID use, if the exposure time to aspirin among only the patients who experienced diverticular bleeding increased by 20% (44.5 to 53.5 person-years), the hazard ratio of aspirin would change from non-significant to significant (HR 1.57, 95% CI 1.08 – 2.28). If the exposure time to aspirin among patients who experienced diverticular bleeding increased by 2,546% (to 1,177.5 person-years), the hazard ratio of PAIs would change from significant to non-significant (HR 1.28, 95% CI 0.99 – 1.64). For NSAIDs, if the exposure time among patients who experienced diverticular bleeding increased by 8% (from 616.0 to 668.1 person-years), the hazard ratio of NSAIDs would change from non-significant to significant (HR 1.17, 95% CI 1.03 – 1.32). If the NSAID exposure time among patients who experienced diverticular bleeding increased by 176% (to 1,701.6 person-years), the hazard ratio of PAIs would change from significant to non-significant (HR 1.31, 95% CI 0.99 – 1.72). There was no level of differential misclassification of over-the-counter aspirin or NSAID use that changed the statistical significance of the hazard ratios for DOACs, warfarin, or heparin (Supplementary Tables 15 and 16).

Thrombotic sequelae of anticoagulation discontinuation

Of the 17,212 patients with a first diverticular hemorrhage identified by the expanded ICD list, 1,542 (9.2%) were prescribed anticoagulation and had a potential indication for ischemic stroke prophylaxis at the time of initial diverticular hemorrhage. Of these patients, 983 (63.7%) continued or eventually resumed anticoagulation. The median time to an anticoagulant prescription refill was 40 days (IQR 19 – 85). Subsequent to the initial hemorrhage, ischemic stroke occurred in 64 (4.1%) patients. Among those patients who experienced an ischemic stroke, the median number of days from initial diverticular hemorrhage to stroke was 392 (IQR 105 – 679). These results are similar to previously reported annualized risks of stroke in patients with atrial fibrillation.³⁹ Of the 64 patients who had an ischemic stroke, 35 (54.7%) were not on anticoagulation at the time of their stroke. Discontinuing anticoagulation was associated with increased relative hazard of ischemic stroke (HR 1.93, 95% CI 1.17 – 3.19, Table 5). Thirteen patients of these patients (37.1%) eventually re-started anticoagulation. The median number of days from the stroke to starting anticoagulation was 115 days (IQR 44 – 295).

Table 5.

Thrombotic sequelae of discontinuing anticoagulation after the initial episode of diverticular hemorrhage among patients with a potential indication for ischemic stroke prophylaxis with anticoagulation (n = 1,542)

	Ischemic stroke (64 events)		Osteoarthritis (166 events)
	HR	95% CI	HR	95% CI
Discontinued anticoagulation	1.93	1.17 – 3.19 *	1.32	0.97 – 1.81
Aspirin	1.37	0.19 – 9.96	0.58	0.08 – 4.19
NSAIDs	0.83	0.44 – 1.60	0.83	0.55 – 1.26
Age	1.03	0.99 – 1.07	1.00	0.98 – 1.02
CHADS2-VASc	1.30	1.09 – 1.57 *	0.95	0.86 – 1.07

Notes:

^*indicates p < 0.05.

The hazard ratios for the risk of ischemic stroke/osteoarthritis are adjusted for aspirin use, NSAID use, age, and CHADS₂-VASc score. Estimates were not adjusted for platelet aggregation inhibitor use because no ischemic strokes occurred in 86.4 person-years of exposure. Osteoarthritis was a falsification endpoint to test for indications of residual confounding in the ischemic stroke analysis.

Abbreviations: CI - confidence interval, DOACs - direct oral anticoagulants, HR - hazard ratio, NSAIDs - non-steroidal anti-inflammatory drugs.

In the cohort of 1,542 patients on anticoagulation at the time of initial diverticular hemorrhage with a potential indication for ischemic stroke prophylaxis, the relative hazard of recurrent diverticular hemorrhage was not elevated among those who continued any anticoagulant after the first hemorrhage (HR 0.98, 95% CI 0.79 – 1.22, Supplemental Table 12). In the falsification test, discontinuing anticoagulation was not associated with new diagnosis of osteoarthritis, indicating that the results of the ischemic stroke analysis were unlikely to be due to residual confounding.

Discussion

Recurrent diverticular hemorrhage is a common complication of colonic diverticulosis. In this study, the incidence rate of a second episode of diverticular hemorrhage was estimated to be 3,625.6 per 100,000 person-years, and the risk of recurrent hemorrhage increased with older age at initial diverticular hemorrhage. Additionally, platelet aggregation inhibitors (PAIs) were demonstrated to be a pharmacologic risk factor for recurrence of diverticular hemorrhage, but anticoagulant medications were not. Patients with a potential indication for anticoagulation to prevent ischemic stroke had increased relative hazard for ischemic stroke if they discontinued anticoagulation following diverticular hemorrhage. These nationally representative estimates of the incidence of and risk factors for recurrent diverticular bleeding may serve as an important step for developing management plans for patients with indications for PAI or anticoagulation use and a history of diverticular hemorrhage.

Given that several studies^{27, 40, 41} have demonstrated that anticoagulant medications are associated with initial all-cause gastrointestinal hemorrhage, one might expect to observe a strong positive association among the classes of anticoagulants and recurrent diverticular hemorrhage. To our knowledge, prior studies have not systematically studied the association of diverticular hemorrhage and anticoagulant use. It is possible that the pathophysiology of diverticular hemorrhage may explain the lack of association between anticoagulants and recurrent diverticular hemorrhage in our study. Because diverticular hemorrhage involves arterial rupture, there is a mechanistic difference relative to other causes of gastrointestinal hemorrhage. For example, the mechanism of anticoagulant-related bleeding in upper gastrointestinal ulcers and intestinal malignancy is postulated to be related to exacerbation of bleeding from systemic and local anticoagulation at sites of existing mucosal injury.^{42, 43} Since this mucosal injury does not exist in colonic diverticula,⁴⁴ diverticular bleeding may not be as strongly related to anticoagulation as other causes of gastrointestinal hemorrhage. It is unknown whether the observed association with PAIs in this study is a result of medication-induced mucosal injury⁴⁵ at the diverticular dome or neck that exposes the artery to the colonic lumen to cause eventual rupture, but this should be explored in future studies.

The study results presented here are an initial step to guide the development of clinical pathways to manage diverticular bleeding. This study showed that anticoagulants were not associated with recurrent hemorrhage and that discontinuing anticoagulation for stroke prophylaxis increased the relative hazard of ischemic stroke. These results are consistent with recent studies that demonstrated high rates of comorbid complications related to lower gastrointestinal bleeding,²² including recurrent thromboembolism^{23, 24} and increased 90-day mortality when anticoagulation is discontinued at hospital discharge.⁴⁶ Furthermore this study demonstrated that after two prior episodes of diverticular hemorrhage, the annual risk of recurrent hemorrhage per patient was approximately 11%. Future studies should assess the impact of different surgical thresholds for colectomy for prevention of future diverticular hemorrhage on overall survival and quality of life.⁴⁷

Strengths and limitations

Prior studies of diverticular hemorrhage using administrative data used hospital-level discharge data,^{1, 3} thus preventing assessment of recurrent hemorrhage. Prior studies of recurrent diverticular hemorrhage through institutional cohorts have widely varying incidence rate estimates, likely due to small sample sizes and loss to follow-up.^4–10 This study used a patient-level administrative database to overcome the limitations of the prior literature through a large sample size of approximately 15,000 patients with longitudinal medical claims data. This allowed for granular assessment of several time-varying exposures. By focusing on patients who already had established diverticulosis, this study controls for confounding related to the development of diverticulosis. Other strengths of this study include the use of directed acyclic graphs and disease risk scores to adjust for confounding from several comorbidities associated with bleeding propensity and prescription for anticoagulants, PAIs, aspirin, and NSAIDs.

However, there are potential caveats to consider when interpreting these study results. First, although a disease risk score was used to adjust for potential confounding from comorbidities and channeling, it is possible that there was unmeasured confounding. The most likely source of unmeasured confounding is misclassification of exposure to aspirin and NSAIDs from over-the-counter use. The potential magnitude of this confounding was assessed using simulation studies adjusting the amount of differential misclassification of aspirin and NSAID use. Although the main analysis did not demonstrate a statistically significant association of aspirin and NSAID use with recurrent diverticular bleeding, the sensitivity analyses showed that if we were able to account for small amounts of over-the-counter use, the hazard ratios for these medications would reach traditional thresholds for statistical significance. This result is consistent with prior research demonstrating that aspirin and NSAID use is associated with diverticular hemorrhage^13–17. Similarly, when accounting for over-the-counter use as an unmeasured confounder, the hazard estimates for the classes of anticoagulant drugs were not meaningfully changed for all levels of over-the-counter aspirin or NSAID misclassification and the hazard estimates for PAIs were altered only at implausibly high levels of differential misclassification (the 1,368 patients who ultimately had recurrent hemorrhage would have to take 1,133 person-years of over-the-counter aspirin while the 13,557 patients who did not have hemorrhage would have to take 0 person-years of over-the-counter aspirin). These results indicate that the hazard ratios for anticoagulants and PAIs in the main analysis were not confounded by over-the-counter aspirin and NSAID use.

Additionally, it is possible that there was misclassification of the study outcome, diverticular hemorrhage, in the claims data. To assess the impact of potential misclassification, we conducted several sensitivity analyses varying the definition of the outcome to consider hemorrhage requiring transfusion, hemorrhage requiring endoscopy, and diverticular hemorrhage identified by an expanded list of ICD codes. None of these analyses meaningfully changed the results. In clinical practice, diverticular hemorrhage is usually a presumed diagnosis when another source of bleeding is not identified and diverticulosis is present. Diverticulosis can be diagnosed with colonoscopy or radiographic tests. In this cohort, 71.5% of patients who experienced a recurrent hemorrhage underwent lower endoscopy, which is within the range of 50.3 – 80.8% reported from prior studies that used administrative data to study lower gastrointestinal bleeding.^{9, 48} Why the other patients did not undergo lower endoscopy could not be ascertained, but could include the absence of severe bleeding and recent imaging confirming diverticulosis.

Notably, the cumulative incidences of recurrent diverticular hemorrhage of 4.7% at one year, 8.3% at two years, and 15.7% at five years were at the lower end of the range of cumulative incidence rates that had been previously reported from institutional cohorts.^4–10 The strict definition for identifying recurrent diverticular hemorrhage that excluded re-hospitalizations for diverticular hemorrhage within 30 days of the initial hemorrhage may explain the lower cumulative incidence rate in this study. This definition was used to identify new hemorrhage events rather than recrudescence of hemorrhage of an incompletely healed bleeding diverticulum. Additionally, this study demonstrated that the prevalence rate of initial diverticular hemorrhage was similar to rates identified in the prior studies of diverticular hemorrhage calculated using administrative data.^{1, 3} This lends support to the reliability of the data from the Optum database.

Finally, because the inpatient data from the Optum database uses ICD procedure codes instead of the more granular Current Procedural Terminology, we were unable to assess how endoscopic interventions such as hemoclip placement or electrocautery at the bleeding diverticulum alter the risk of recurrent hemorrhage. However, because the main analysis excluded recurrent hemorrhage episodes within 30 days of the initial hemorrhage and endoscopic intervention for an actively bleeding diverticulum is rare,⁴⁹ we expect that this limitation did not profoundly alter the study results.

In conclusion, in this study of patients who had previously experienced an initial episode of diverticular hemorrhage, the incidence per 100,000 person-years of a second episode of diverticular hemorrhage was 3,625.6 (95% CI 3,436.0 – 3,823.0). The incidence rate increased with older age at initial hemorrhage and with more episodes of prior diverticular hemorrhage. PAI use was a risk factor for a second episode of diverticular hemorrhage, while anticoagulant use was not. Moreover, among patients with a potential indication for stroke prophylaxis, discontinuing anticoagulation after the initial episode of diverticular hemorrhage was associated with approximately 2-fold increased risk of ischemic stroke. These results may serve as a launching point for future clinical management algorithms for recurrent diverticular hemorrhage. The creation of such pathways is increasingly important as the prevalence of diverticular hemorrhage is expected to increase as the global population ages.⁵⁰

Abbreviations:

DOAC

Direct-acting oral anticoagulant

ICD

International Classification of Diseases

IQR

Interquartile range IR: Incidence rate

NSAID

non-selective non-steroidal anti-inflammatory drug

Optum

OptumInsight Clinformatics

PAIs

Platelet aggregation inhibitors

SSRIs

Selective serotonin reuptake inhibitors