Hepatic arterial infusion (HAI) pump placement requires careful assessment of the arterial anatomy of the liver, suitability of the abdominal wall, and the assessment of extrahepatic disease. The initial evaluation of a patient with mCRC should include cross-sectional imaging of the chest, abdomen, and pelvis, usually via computed tomography (CT) to look for radiographically evident extrahepatic disease. HAI pump placement is generally not indicated in patients with apparent lung or peritoneal involvement. However, in carefully selected patients with minimal extrahepatic disease and a substantial burden of CLM, HAI treatment can be considered [27]. For patients with unresectable disease, a staging laparoscopy should be considered, as up to 1/3 of patients will have evident extrahepatic disease [28]. When extrahepatic disease is encountered and the judgment is that it is sufficient to preclude HAI pump placement, intraoperative frozen section is of obvious importance.

The preoperative evaluation should also consist of a CT arteriography to evaluate the hepatic arterial anatomy. Given standard anatomy, the preferred conduit for placement of the catheter is the gastroduodenal artery (GDA), as this is the side-branch immediately proximal to the proper hepatic artery. However, up to 34% of patients will have variant anatomy that requires special consideration [29]. The hepatic arterial anatomic variants are summarized in Table 8.2, and include replaced or accessory left and right hepatic arteries and combinations of multiple variants. Determination of the exact nature of the aberrant anatomy via careful review with the radiologist is imperative, as these findings impact the operative plan.

Suitability of the abdominal wall is also a key consideration, as patients with large ventral hernias or prior operations may have attenuated musculofascial layers of the abdominal wall. The operative plan usually consists of pump placement in the lower abdomen, typically on the left side to avoid the potential use of a future right subcostal incision. In obese patients with large subcutaneous spaces and in patients with large hernias, placement of the pump itself on the lower chest wall can enable location and access to the pump, as well as minimize the risk of flipping. Any one of a number of incisions can be employed for HAI pump placement, including an upper midline incision, right subcostal incision, or a limited hockey-stick incision. Of note, the pump itself should be placed in a subcutaneous pocket via a separate incision with tunneling of the catheter into the peritoneal cavity. Regardless of the incision type chosen, preoperative antibiotics are important in this setting, as are other standard preoperative precautions.

Intraoperatively, the hepatic artery and its branches should be carefully dissected and skeletonized. The right gastric artery should be divided, and the distal CHA, proximal proper hepatic artery (PHA), and GDA identified, encircled and freed from surrounding attachments. Proper identification and mobilization of these structures, including the entire extrapancreatic GDA, are critical. During this dissection, consideration should be given to removing portal lymph nodes in the vicinity of the CHA and the porta hepatis, as these can occasionally be interpreted as sources of extrahepatic perfusion. A cholecystectomy is also performed, as HAI therapy delivered to an in-situ gallbladder (via the cystic artery) will cause chemical cholecystitis. All branches of the CHA, PHA, and GDA are divided and ligated to minimize perfusion of the pancreas, duodenum, or stomach by the pump. The left and right hepatic artery are similarly dissected for approximately 2 cm from the PHA origin to ligate any branches that may serve as conduits for extrahepatic perfusion [32]. Finally, a hepatic arterial pulse is palpated, while the GDA is temporarily occluded to ensure there is not retrograde flow in the GDA owing to celiac stenosis. If there is retrograde flow, an attempt to release the arcuate ligament may re-establish normal flow. If this is not successful, one can consider placing the catheter in the CHA, allowing flow to the liver through the GDA into the PHA.

Vascular control is obtained, and the distal GDA is ligated at its most distal point. In the case of standard anatomy, a transverse arteriotomy is made in the GDA, and the catheter is inserted up to the confluence with the hepatic artery. Positioning of the catheter tip is crucial, as the proximal GDA should neither be exposed to full concentrations of chemotherapeutic agent, nor should the catheter protrude into the lumen so far as to induce thrombosis. The optimal approach when there is aberrant anatomy is ligation of the aberrant vessel(s) and placement of the catheter in the GDA, as cross-perfusion is extremely reliable. Cross-perfusion is often visible at the time of operation, and occurs in almost everyone by 4 weeks after the operation. In a series of 52 patients with variant anatomy, all but one had adequate bilobar perfusion at 4 weeks [31]. Cannulation of a vessel other than the GDA is associated with a significantly elevated incidence of catheter-related complications and limited catheter durability, and is not preferred. When the GDA is not available, we generally prefer placement in the right or left hepatic artery (with ligation); in rare situations, vascular graft placement to create a “GDA” for catheter insertion is required. In the case of variant GDA anatomy, ligation of either in situ (left or right) hepatic artery may be necessary if the GDA arises from the contralateral vessel.

When placed at the time of major hepatectomy, the technical considerations are no different, except that the stump of the ligated arterial branch may be employed to perfuse the remnant liver if the GDA is not available. Of note, ligation of aberrant left or right arteries to a remnant liver for catheter placement should be performed with caution, as it may exacerbate postoperative liver dysfunction. In the face of a remnant liver perfused by a replaced hepatic artery, pump placement (into the GDA) should probably be deferred rather than employing direct cannulation of the replaced vessel.

The catheter is secured in place with silk ties, and the pump reservoir is placed in the pump pocket. Bilobar perfusion of the liver and the absence of extrahepatic perfusion are confirmed by either fluorescein or half-strength methylene blue injection into the side port of the pump. If extrahepatic perfusion is detected (most commonly to the duodenum and head of pancreas), a search for any vessel ensues with ligation and retesting. The catheter is then flushed with heparinized saline and wounds are closed. Postoperatively, perfusion is assessed by a radionuclide pump flow study using technetium 99m (^99mTc)—sulfur colloid and ^99mTc-labeled macroaggregated albumin (MAA). This study is used to detect extrahepatic perfusion (occurs in 5–7% of cases) that can usually be salvaged by angiographic intervention [33, 34]. Incomplete hepatic perfusion can also occur, but usually resolves on a repeat scan obtained a few weeks after the index study. If resolution is not apparent, there may be a missed accessory vessel not ligated at the first operation, and consideration to angiography should be given.

While the implantable hepatic artery infusion pump is the most commonly employed device, there are other means of access to the hepatic arterial tree. One of the earliest approaches was the placement of a subcutaneous port with a catheter terminating in the hepatic artery. A large randomized MRC/EORTC study evaluating HAI 5-FU/leucovorin with systemic 5-FU/LV, however, featured a 36% rate of catheter-related complications that limited dose administration [24]. Subsequent studies exploring the role of IV oxaliplatin via HAI catheters placed in the GDA but using a subcutaneous pump showed significant improvement in the rate of catheter-associated complications of 10–15% [35].

Percutaneously placed catheters have also been explored. Arru et al. evaluated percutaneous axillary artery catheters as compared to implantable pumps, finding a 43% rate in the percutaneous group of an issue, causing either an interruption or end to treatment (versus 7% in the implantable pump group) [36]. Several studies have also attempted to develop and refine the use of intercostal artery catheters, either with a subcutaneous port or with an attached pump [37].

Recent attention has turned to minimally invasive surgical placement of implantable pumps. A number of initial case series established feasibility of a laparoscopic approach; the largest of these describes an experience with 38 patients, among whom there was one mortality and no pump-related morbidity [38]. Another series featuring 29 patients demonstrated that aberrant anatomy could be addressed safely via a laparoscopic approach and without significant perioperative morbidity [39]. Despite its widespread application, a robotic approach to HAI catheter and pump placement has yet to be studied in any systematic fashion.

HAI pump chemotherapy for unresectable CLM has been extensively studied. Over the last 20 years, ten phase III trials (Table 8.3) comparing HAI with systemic chemotherapy have been conducted; three subsequent meta-analyses have evaluated these findings further still. Overall, there is relative concordance among the studies that response rates are higher with HAI. Nine of the ten studies employed FUDR as the HAI chemotherapeutic—each showed response rates of 42–62%, compared to response rates of 9–24% for systemic chemotherapy in these trials [45]. However, all of these studies employed older systemic regimens consisting of intravenous FUDR, 5-FU alone, or 5-FU/leucovorin, rather than modern regimens incorporating either irinotecan or oxaliplatin.

Despite the substantial increases in response rate, these studies have often failed to detect a difference in overall survival. Several factors have contributed to this phenomenon. Most notably, early studies at MSKCC (99 patients) [40] and in the Northern California Oncology Group (NCOG) trial [41] both allowed crossover between groups (which occurred frequently), making an intention-to-treat analysis of overall survival meaningless. Those studies that did not allow for crossover and have shown differences in overall survival—namely the Hepatic Artery Pump trial (HAPT) and a French trial—are confounded by the fact that patients in the control arms frequently received only best supportive care rather than 5-FU [42, 43].

Among these ten studies comparing HAI with systemic chemotherapy, the CALGB 9481 trial is the most recent. In this trial, no crossover was permitted, and 134 patients were randomized to either systemic 5-FU/LV (via the Mayo Clinic regimen) or HAI (consisting of FUDR, LV, and dexamethasone). Dexamethasone was added in this series because of earlier data showing decreased biliary toxicity with the addition of steroid to HAI [46]. Again, response rates were significantly higher with HAI (47% vs 24%) and there was a significant improvement in overall survival (24.4 vs 20 months; p = 0.0034) [44].

Three meta-analyses of these trials have been performed, with variable results in determining a survival advantage. This inconsistency has been driven by variable exclusion criteria among the trials for either methodological reasons or because of concerns about study design—especially for those trials where some control patients received best supportive care only, or crossover was allowed. The most recent meta-analysis, published in 2007, includes all ten trials and attempts to account for their design flaws. The authors conclude that HAI was associated with a significantly elevated response rate (42.9% vs 18.4%), but this did not translate into an improvement in overall survival (hazard ratio 0.9; p = 0.24) [47]. Given the extreme heterogeneity of these trials, it remains difficult to draw any firm conclusions from these meta-analyses.

As mentioned above, these trials predate the development of modern and more effective systemic chemotherapeutic regimens incorporating oxaliplatin or irinotecan. In addition, several of these studies detected a high frequency of extrahepatic progression (40–70%) in patients treated with HAI. More recent studies have attempted to exploit the lack of systemic exposure to chemotherapy with HAI FUDR treatment, and evaluate the efficacy of HAI chemotherapy combined with systemic chemotherapy. The first of these studies involved 95 patients randomized to HAI FUDR with or without intravenous FUDR, and showed similar response rates (~60%) but higher extrahepatic recurrence in the HAI-only group (79% vs 56%; p < .01) [48].

Several phase I and II studies have since combined HAI with systemic chemotherapy (Table 8.4). The first of these evaluated 46 patients given HAI consisting of FUDR + dexamethasone in conjunction with systemic irinotecan; response rates were 74% in these pre-treated patients, with an overall survival (OS) of 20 months following pump placement [49]. Similar results were seen with addition of systemic FOLFOX (oxaliplatin +5-FU/leucovorin); in 15 patients, a response rate of 87% with a median OS of 22% was obtained [50]. The combination of oxaliplatin and irinotecan yielded the best results, with a pooled analysis of 49 patients showing a 92% response rate and an OS of 51 months for previously untreated patients and 35 months for previously treated patients. Also of note was that 47% of these individuals converted from unresectable to resectable disease [51]. A more recent phase II study with 49 patients (two-thirds previously treated) treated with HAI and modern systemic chemotherapy (initially with bevacizumab) showed high response rates of 76% and a conversion to resectability in 47% of the 49 patients. Median survival was 38 months for the whole cohort [52].