Screen-film arteriography can demonstrate bleeding at rates as low as 0.5 mL/min in a canine model (5), but in vitro studies suggest that digital subtraction arteriography is five to nine times more sensitive than screen-film arteriography in detecting hemorrhage (6), which seems to be equivalent to detection rates of scintigraphic images (7). The major limitation of angiography relates to the intermittent nature of bleeding, which can result in a negative study if the bleeding has temporarily stopped at the time of contrast injection.

Glucagon and Buscopan may be given before the procedure to decrease bowel motility and motion artifacts during digital subtraction angiography (8). Aortograms are usually not performed, since visualization of contrast extravasation into the GI tract requires more selective injection (4). The vessel selected first should be based on the suspicion of the likely source of bleeding according to history, clinical signs, as well as localization provided by scintigraphic images or computed tomography scans. For UGI bleeding, the celiac and superior mesenteric arteries (SMAs) are the primary targets. If there is no sign of active bleeding on injection of the main trunks, more selective injection may be needed. For duodenal or gastric fundus bleeding, the gastroduodenal artery (GDA), or left gastric arteries, respectively, should be studied (4). Obtaining images should be continued until the venous phase has cleared out to help distinguish contrast extravasation from persistent venous opacification.

The only direct angiographic sign of UGI bleeding is extravasation of contrast medium into the bowel lumen (2). Indirect signs include aneurysms/pseudoaneurysms, vessel irregularity, vessel cutoff and arteriovenous/arterioportal shunting, neovascularity, and increased vascularity from dilated arterioles (as seen in angiodysplasia). It is important to image into the venous phase.

When a bleeding source is not identified because GI bleeding is often intermittent, provocative angiography may improve the rate of positive angiographic findings. Infusion of tolazoline (vasodilators), heparin, or even thrombolytics like tissue plasminogen activator, stimulate bleeding to allow the pathology to be localized. In one study based on 16 patients with nonlocalized lower GI bleeding from previous endoscopic and angiographic studies, these infusions identified the bleeding site in 37.5% of cases without causing any hemodynamic instability (9).

The choice of embolic agent depends on a combination of the vascular anatomy, angiographic findings, the achievable catheter position, and the operator preference (Table 1). The most common embolic agents are metallic coils and polyvinyl alcohol (PVA) particles (usually 300-500 µm). Use of coils as the only embolic agent is significantly associated with early rebleeding, compared with the use of PVA or gelatin sponge with coils (10, 11). Use of N-butyl cyanoacrylate (NBCA) is advantageous for massive bleeding that requires urgent hemostasis, especially in patients with coagulopathy because of rapid embolization (12).

Types of embolization can be classified into localized, proximal, and segmental embolizations (Fig. 1). Localized embolization is possible when the bleeding point is superselected. Proximal embolization is defined when a microcatheter could not enter the bleeding point and embolization is done in its parent artery, while the bleeding point and its distal part are not embolized. Segmental embolization is when an adjacent branch artery or arteries are included to be embolized. With proximal embolization, recanalization of the bleeding point can occur due to distal back flow. With excessive segmental embolization, ischemic complications of the involved bowel can occur.

One of the useful maneuvers in localizing culprit vessels requires clips to be placed around the area of bleeding during pre-embolization endoscopy. The clips remain in position for several hours and allow for an educated guess of the location of the culprit vascular branch (13, 14). If no extravasation is seen despite the injection of contrast, then the branches terminating at the clip are superselected using microcatheter techniques and embolized. Rotational angiography can be useful to determine the relationship between the metallic clip and a culprit vessel (14).

Vasopressin infusion works by constricting the mesenteric vessels and contraction of the smooth muscle in the bowel wall, thus reducing the blood flow to the site of bleeding and inducing the formation of a stable clot at the bleeding site. Patients should be on continuous cardiac monitoring, since vasopressin can induce coronary vasoconstriction. Infusion is started at 0.1 units/min. Repeat angiography is performed after 20 minutes to ensure that bleeding has stopped and that the vessels are not overconstricted. If bleeding persists, the dose is increased to 0.2 units/min. Infusion is then increased to 0.4 units/min if bleeding persists. Excessive vasoconstriction can lead to bowel infarction. It is known to be effective in patients with microcatheter-inaccessible bleeding or diffuse bleeding (such as hemorrhagic gastritis). However, disadvantages include the rate of rebleeding after discontinuation of the infusion, cardiovascular complications, and the difficulty in maintaining a selective catheter position. In one study, the success rate of 52% for vasopressin was inferior than that of 88% for embolization in major GI bleeding (15).

The GDA has a dual supply from both the hepatic artery and SMA. Back pressure may allow continued bleeding if embolization is performed proximally. A microcatheter is advanced beyond the bleeding site and embolization is done to prevent backflow to the bleeding site. Then, the microcatheter is withdrawn, depositing more embolic agent, until the embolization takes place proximal to the bleeding site, using the so-called sandwich technique. Likewise, bleeding from the pancreaticoduodenal arcade should be treated by embolizing the arteries both proximal and distal to the lesion. GI bleeding from GDA stump is quite common after pancreaticobiliary surgery (Fig. 2). Embolization both proximal and distal to the GDA stump or stent-graft placement could be an option. Although hepatic ischemia or infarction develops in 30-40% of patients with embolization of both proximal and distal to the GDA stump (Fig. 2), patent portal vein and extensive collateral pathways, including the inferior phrenic arteries and gastric arteries, usually serve to protect the liver from ischemic insult (16). Stent-graft placement is better than embolization for preserving intrahepatic arterial flow. Postembolization 'check' angiograms from both the celiac axis and SMA sides are required to ensure that no collateral supply to the bleeding site is present.

Embolization of pseudoaneurysms needs to be mentioned. Packing the sac of the pseudoaneurysm with coils should be avoided because this rebleeding is almost inevitable (16, 17). Because the wall of the pseudoaneurysm is weak, the breakdown or rupture of the weaker pseudoaneurysm (than normal adjacent arterial wall) seems to occur after only sac packing with embolic materials. Therefore, the sandwich technique of both proximal and distal to the bleeding site should be done. In those cases with microcatheter-inaccessible bleeding point, particulate or liquid embolic agents should be utilized. In cases requiring deeper penetration, liquid embolic materials such as NBCA could be an option (Fig. 3).