Langenbecks Arch Surg (2014) 399:989–1000 DOI 10.1007/s00423-014-1241-3
SYSTEMATIC REVIEWS AND META-ANALYSES
The benefits of liver resection for non-colorectal, non-neuroendocrine liver metastases: a systematic review Timothy L. Fitzgerald & Jason Brinkley & Shannon Banks & Nasreen Vohra & Zachary P. Englert & Emmanuel E. Zervos
Received: 30 April 2014 / Accepted: 11 August 2014 / Published online: 23 August 2014 # Springer-Verlag Berlin Heidelberg 2014
Abstract Purpose Defining the benefits of resection of isolated noncolorectal, non-neuroendocrine (NCRNNE) liver metastases is difficult. To better understand the survival benefit in this group of patients, we conducted a systematic review of the previous literature. Methods Medline, Web of Knowledge, and manual searches were performed using search terms, such as “liver resection” and “primary tumor.” Inclusion criteria were year>1990, >five patients, and median survival reported or derived. An expected median survival was calculated from weighted averages of median survivals, and differences were assessed using a permutation test. Results A total of 7,857 references were identified. Overall 4,735 abstracts were reviewed; 120 manuscripts evaluated and of these, 73 met the study inclusion criteria. The final population consisted of 3,596 patients with renal (n=234), ovarian (n=119), testicular (n=153), adrenal (n=90), small bowel (n= 28), gallbladder (n=21), duodenum (n=38), gastric (n=481), pancreatic (n=55), esophageal (n=23), head and neck (n=15), and lung (n=36) cancers, gastrointestinal stromal tumors (GISTs) (n=106), cholangiocarcinoma (n=13), sarcoma (n= 189), and melanoma (n=643). The greatest expected median was 63 months for genitourinary (GU) primaries (n=549; range 5.4–142 months) followed by 44.4 months for breast cancer (n=1,013; range 8–74 months), 22.3 months for gastrointestinal cancer (n = 549; range 5–58 months), and T. L. Fitzgerald (*) : S. Banks : N. Vohra : Z. P. Englert : E. E. Zervos Division of Surgical Oncology, Department of Surgery, Brody School of Medicine, East Carolina University, 4S24 600 Moye Boulevard, Greenville, NC 27834, USA e-mail: [email protected] J. Brinkley Division of Biostatistics, School of Allied Health, East Carolina University, 4S24 600 Moye Boulevard, Greenville, NC 27834, USA
23.7 months for other tumor types (n=1,082; range 10– 72 months). Using a permutation test, we observed that survival was best for patients with GU primaries followed by that for breast cancer patients. Additionally, we also observed that survival was similar for those with cancer of the GI tract and other primary sites. Conclusions There appears to be a benefit to resection for patients with NCRNNE liver metastases. The degree of survival advantage is predicated by primary site. Keywords Liver metastases . Non-colorectal non-neuroendocrine . Liver surgery . Systematic review
Introduction The clinical utility of resection for hepatic metastasis has engendered significant controversy. Although stage IV colorectal cancer is generally considered incurable, in selected patients with isolated hepatic metastasis there appears to be a survival benefit to surgical resection [1, 2]. Based on that data from large case series, such as the Registry of Hepatic Metastases, there has been a progressive liberalization of resection criteria for isolated liver metastases [3]. Even though no prospective randomized data exist, most hepatic surgeons would consider the ability to render a patient disease-free with an adequate liver remnant is an indication for resection and would expect a 5-year survival in near 60 % cases [4]. Similar results have also been reported for resection of neuroendocrine liver metastases [5]. Defining the benefits of resection of isolated non-colorectal, non-neuroendocrine (NCRNNE) liver metastases has been difficult because outcome data published using tertiary institutional and multi-institutional registries include a myriad of primary sites with little data on specific tumor types [6–10]. Data on individual histologic subtypes frequently come from
990
small case series [11–19]. Therefore, defining benefits of liver resection for individual patients is difficult [6]. A systematic review of the current literature on liver resection for NCRNNE hepatic metastases would assist in filling this knowledge gap. To clarify, a structured review of published studies on liver resection for NCRNNE hepatic metastases was conducted.
Materials and methods Search strategy Original studies published were identified in the following electronic data bases: Medline, Scopus, and Web of Knowledge. The following “free text,” medical subject heading (MeSH), and EMBASE subject headings were used: “liver resection” and “non-colorectal non-neuroendocrine,” “renal cell carcinoma,” “ovarian cancer,” “testicular cancer,” “adrenal cancer,” “small bowel cancer,” “gallbladder cancer,” “duodenal cancer,” “gastrointestinal stromal tumor,” “cholangiocarcinoma,” “gastric cancer,” “pancreatic cancer,” “esophageal cancer,” “sarcoma,” “melanoma,” “head and neck cancer,” or “lung cancer.” In addition, manual searches were also conducted using the bibliography from identified articles. The search was restricted to modern series, defined as the studies conducted from 1990 to February of 2013. No unpublished data were included. Studies were restricted to those in English and on human subjects. Study inclusion criteria Prospective or retrospective studies reporting outcomes on at least five patients were included. The primary goal of this study was better evaluation of survival after resection of hepatic metastases. Due to inconsistent reporting of survival time in various studies (1-, 3-, or 5-year survival), median survival was chosen as the primary endpoint. This is an important point because median survivals are often reported without estimates of variation (that is, standard deviation) or confidence intervals. This information is necessary for a traditional meta-analysis. Given this primary endpoint, studies were excluded if median survival was not reported or could not be extracted. Data extraction and clinical endpoints All studies meeting selection criteria were reviewed by the first and third author (TF and SB) to determine eligibility. Discrepancies were resolved by discussion and consensus. As mentioned above, the primary endpoint of this study was to report median survival after hepatic metastasectomy based on primary tumor type. Median survival was obtained directly from the article or estimated using reported survival. If
Langenbecks Arch Surg (2014) 399:989–1000
outcomes for resection of hepatic metastases were reported for several different primary tumor types, each subtype was treated as an individual cohort. For example, a manuscript on NCRNNE liver metastases might report multiple primary sites for which median survival may be reported for only some groups. Each tumor type for which median survival was reported or derived was treated as a separate cohort. Given this, the number of individual cohorts reported exceeded the number of articles meeting the inclusion criteria. Statistical analysis Survival was calculated using median survival from individual reports. To understand survival after resection, based upon the number of patients in each cohort, an overall weighted average of median survivals was derived (with sample sizes from each study group used as weights). This reflects more accurately what the expected median could have been had all patients been part of one study and also that weighting allows for studies with larger number of patients to contribute more to the expected median survival. This approach was selected because there was a significant variation in the median survival between and within tumor types (a range from 5 to 140 months). To understand variation within and between groups, we provided a minimum and maximum study median (as opposed to traditional confidence bounds). To move beyond simple descriptive statistics and understand whether true differences in outcomes exist, variability was assessed using a permutation test that considered a large random sample of permutations (10,000) of the studies and examined what the difference in expected medians would have been in each permutation. From there, p values were determined based on the proportion of permutations that had more extreme differences than what was observed. Using this statistical approach, differences between the broad groupings of genitourinary (GU), breast, gastrointestinal, and other cancer types were analyzed. Analyses were performed using a combination of SAS and JMP software (versions 9.3 and 11Pro, respectively).
Results A total of 7,857 citations were identified. Out of which, 3,122 citations were excluded on initial review and 4,735 abstracts were fully reviewed; additional 4,621 citations were excluded, and 120 articles were reviewed in full, out of which, 73 met the inclusion criteria (Fig. 1). Some articles reported more than one primary site. To study individual primary sites, each primary site reported in a published study from which median survival could be derived was treated as an individual data point or cohort for statistical analysis. As a result, there were
Langenbecks Arch Surg (2014) 399:989–1000 Fig. 1 Outline of systematic literature review
991 7857 potenal arcles idenfied via Medline, Scopus, Web of Knowledge and manual search 3122 excluded upon review citaon 4742 abstracts fully reviewed 4621Excluded 121 arcles idenfied for full-text reviewed
73 usable studies for analysis
48 Excluded secondary to lack of extractable data, non-reseconal therapy, no median survival, inadequate follow up reported, number of paents or duplicate
112 individual data points
114 cohorts obtained from the 73 articles meeting the inclusion criteria as listed in Tables 1, 2, 3, and 4. Site-specific data points Tables 1, 2, 3, and 4 report the individual cohorts included in this study under the categories of breast, gastrointestinal, genitourinary, and other cancer types. Author last name, year of publication, years spanned in the study, total number of patients, and median survival are listed. In Table 1, median survival for patients undergoing resection of hepatic metastases from breast cancer is reported. A total of 27 individual cohorts were extracted representing 1,281 patients. These studies spanned the years 1982–2010 and were published between 1994 and 2012. The results for patients undergoing resection of liver metastases from primary gastrointestinal tumors are reported in Table 2. There were 31 individual cohorts identified in articles published from 1994 to 2012 spanning the years of 1963– 2011. This specific tumor category has the longest time span over which studies were conducted. A total of 684 patients with primary tumors originating from the esophagus (n=23), gastroesophageal junction (n=25), stomach (n=481), pancreas (n=55), duodenum (n=38), bile duct (n=13), gallbladder (n=21), and small bowel (n=28) were reported. For GU tumors, a total of 16 individual cohorts were identified in the articles published from 1990 to 2012 spanning the years of 1976–2010 (Table 3). A total of 549 patients were reported with the primary tumor types, such as testicular cancer (n=153), ovarian cancer (n=119), renal cancer (n= 234), and uterine cancer (n=43). The final category “other,” including cancer sites and other histologies that were not included in the main three categories (breast, gastrointestinal, and GU), is described in Table 4. There were a total of 35 cohorts representing 1,082 patients in articles published from 1998 to 2012 spanning the years of
1981–2010. Individual tumor types were gastrointestinal stromal tumor (GIST) (n=106), sarcoma (n=189), melanoma (n= 646), lung cancer (n=36), head and neck cancer (n=15), and adrenal carcinoma (n=90). Overall survival To understand the impact of resection for NCRNNE liver metastases on long-term survival, data analysis of all reported or derived median survival was performed. Using the individual median survival and the sample size, an expected median survival was derived (Table 5) for each of the four categories outlined above: breast, gastrointestinal, GU, and others. In each category, data were further reported by primary site (Fig. 2). GU tumors had the longest expected median survival of 63 months. There is a fairly wide range of reported median survivals from a minimum of 5.4 months to a maximum of 142 months for individual cohorts. Subtypes of cancers reported in this category included testicular, ovarian, renal, and uterine carcinoma, with expected median survival of 61.65, 67.8, 67.9, and 32 months, respectively. The category with the next best expected median survival of 44.4 months was breast cancer, with a minimum of 8 months to a maximum of 74 months for individual cohorts. Patients with gastrointestinal primaries and other sites had similar expected median survival of 22.3 and 23.7 months, respectively. Survival differences between categories To determine the impact of hepatic resection on survival across the categories of breast, GU, gastrointestinal and others, a comparison was made, as noted in the “Materials and methods” section, using a permutation test. When comparing GU to breast, gastrointestinal, and other primary sites, we observed a significant improvement in survival with pvalues of 0.002, 0.014, and 0.002, respectively. A survival
992 Table 1 Resected breast cancer liver metastases, manuscripts published from 1990 to 2013
Langenbecks Arch Surg (2014) 399:989–1000
Author
Year published
Time span
Patients (number)
Median survival
Schneebaum [46] Elias [11] Raab [67] Elias [34] Seifert [68]
1994 1995 1998 1998 1999
1982–1990 1986–1994 1983–1994 1984–1996 1985–1997
6 21 34 35 15
42 26 27 40 57
Pocard [50] Selzner [47] Yoshimoto [69] Maksan [42] Kondo [70] Carlini [16] Vlastos [71] Sakamoto [72] Yedibela [35] Yedibela [35] Adam [36] Adam [43] Lendoire [33] Thelen [73] Lubrano [74]
2000 2000 2000 2000 2000 2002 2004 2005 2005 2005 2006 2006 2007 2007
1988–1997 1987–1999 1985–1998 1984–1998 1990–1999 1990–1999 1991–2002 1985–2003 1978–1989 1990–2001 1983–2004 1984–2004 1989–2006 1988–2006
49 17 25 9 6 17 31 34 7 17 454 85 19 39
42 24 34.3 60 28 53 25 36 8 40 45 32 75 53.6
Reddy [8] Caralt [75] Hoffmann [49]
2007 2007 2008 2010
1989–2004 1995–2005 1988–2006 1999–2008
16 20 12 41
42 67 35.9 58
Bresadola [7] Abbott [76] Groeschl [6] Walsum [48]
2011 2011 2012 2012
1989–2006 1997–2010 1990–2009 1991–2010
13 59 115 32
44 57 52 55
advantage was also noted when comparing breast cancer to gastrointestinal (p = 0.001) and other primary sites (p = 0.0312). There was no difference in expected median survivals when comparing gastrointestinal cancers to other types of cancers (p=0.3118).
Discussion Resection of the liver metastases has engendered a fair degree of controversy. In the early 1990s, there was a considerable debate regarding the efficacy of resection for colorectal liver metastases [20]. There is a little debate now that patients benefit from resection for isolated hepatic disease [4]. Similar data exist for resection of neuroendocrine liver metastases [21]. Controversy, however, continues regarding resection of NCRNNE liver metastasis. Small institutional series reporting outcomes for resection of NCRNNE liver metastases generally include an array of primary sites making it difficult to define the benefit of resection for individual patients [6–10, 22–36]. Despite the limitations of these data, it is generally seen that
outcomes are poor for patients with non-colorectal gastrointestinal primaries [32, 37]. This is further highlighted by a recent case-controlled study that found no difference between observation and resection for this group [38]. The authors did, however, report an advantage to resection versus observation for patients with non-gastrointestinal primaries. In this analysis, patients with GU primaries (Tables 3 and 5) had the greatest expected median survival compared to all other subsets, followed by patients with breast, gastrointestinal, and other primary sites of disease; gastrointestinal and other primaries had a similarly poorer expected median survival. There were 549 individuals with resected GU liver metastases reported in 16 cohorts with primary sites, including testicular (70 to 80 % of patients cured of their disease), ovarian, renal, and uterine cancer. The expected median survival was fairly high at 63.4 months with a high degree of variation (5–142 months). Patients with ovarian, renal, and testicular metastases appeared to have superior survival duration as compared to those with uterine cancer metastases (Table 5). The findings of this analysis are in agreement with those of the published
Langenbecks Arch Surg (2014) 399:989–1000
993
Table 2 Resected gastrointestinal primary metastases, manuscripts published from 1990 to 2013 Tumor type
Author
Year published
Time span
Patients (number)
Median survival
Esophagus
Adam [36] Bresadola [7] Adam [36] Ochiai [56] Taniguchi [77] Miyazaki [78] Elias [34] Ambiru [58] Imamura [79] Saiura [80] Zacherl [57] Sakamoto [62] Roh [61] Adam [36] Sakamoto [55] Bresadola [7] Schildberg [60]
2006 2011 2006 1994 1997 1997 1998 2001 2001 2002 2002 2003 2005 2006 2007 2011 2012
1983–2004 1989–2006 1983–2004 1963–1991 1998–1994 1980–1994 1984–1996 1975–1999 1990–1997 1981–1998 1980–1999 1985–2001 1988–1996 1983–2004 1990–2005 1989–2006 1972–2008
20 3 25 21 6 21 11 40 17 10 15 22 11 64 37 10 31
16 18 14 58 21.2 18.6 18 12 12.6 25 8.8 21 19 15 31 16 15.2
Takemura [54] Garancini [63] Miki [81]
2012 2012 2012
1993–2011 1998–2007 1995–2009
64 21
34 11
Wang [64] Adam [36] De Jong [15] Adam [36] Adam [36] De Jong [15] Hemming [82] Elias [34] Hemming [82] Bresasola [7] Adam [36]
2012 2006 2010 2006 2006 2010 2010 1998 2010 2011 2006
2003–2008 1983–2004 1993–2009 1983–2004 1983–2004 1993–2009 1996–2009 1984–1996 1996–2009 1989–2006 1983–2004
50 30 40 15 15 12 11 13 7 9 5 28
15 11 20 13 38 34 23 28 21 42 5 58
GE Junction Gastric
Pancreas Duodenum
Cholangiocarcinoma Gallbladder
Small bowel
literature. The largest series, a multi-institutional registry by Adam, reported similar median survival with GU metastases classified as a favorable subtype for hepatic resection. These outcomes appear to be superior to what one would expect with observation alone. Although survival is excellent with testicular cancer, hepatic resection should be reserved for only those poor prognostic patients with medically refractory disease [39]. It is assumed that such patients would have progressed without intervention. Expected survival for a patient with ovarian cancer with visceral metastasis is only 10.5 months. Prognosis is similarly poor for patients with metastatic uterine cancer (12.3–15 months) and renal cell cancer (10–20 months) with modern chemotherapy regimens [40, 41]. In this systematic review, liver resection was most frequently reported for breast carcinoma metastases. A sample
of 27 cohorts representing data from 1,281 patients was identified (Table 1). The expected median survival was 44.3 months (Table 5). There was a significant variation in survival among various studies (8–75 months). As mentioned above, patients with breast cancer liver metastases have an improved survival compared to those with gastrointestinal or other tumor types. These data are consistent with other studies [42, 43]. A thoughtful workup of such patients is imperative, as the incidence of isolated hepatic recurrence is unusual [44]. It is necessary to routinely work up such patients with positron emission tomography to ensure absence of disease at other sites. In addition, these patients also appear to have an increased survival if treated with hormonal or systemic chemotherapy in addition to resection [45–47]. Other factors associated with improved survival include lymph node negative disease, hormone receptor positivity,
994 Table 3 Resected genitourinary liver metastases, manuscripts published from 1990 to 2013
Langenbecks Arch Surg (2014) 399:989–1000
Tumor type
Author
Year published
Time span
Patients (#)
Median survival
Testicular
Goulet [83] Elias [34] Adam [36] You [13] Merideth [14]
1990 1998 2006 2009 2003
1976–1989 1984–1996 1983–2004 1975–2002 1976–1999
14 20 78 4 26
64.8 23 82 55.2 26.3
Adam [36] Abood [84] Roh [85] Stief [86] Alves [12] Weitz [37] Adam [36] Staehler [87] Ruys [88] Langan [89] Adam [36]
2006 2007 2011 1997 2003 2005 2006 2010 2011 2012 2006
1983–2004 1998–2006 1991–2008 1983–1993 1982–2001 1981–2002 1983–2004 1995–2006 1990–2008 1980–2010 1983–2004
65 10 18 13 14 11 85 68 33 10 43
98 33 38 16 26 115 36 142 33 24 32
Ovary
Renal
–
response to preoperative chemotherapy, solitary liver metastases, and prolonged disease-free interval [11, 42, 47–50]. Survival after liver resection for breast cancer appears to be better than one would expect with observation alone. Marinai et al. reported an 80.7 % 3-year survival period after surgical resection versus 50.89 % in a matched cohort not undergoing resection [51]. On a multivariate analysis, the risk for death was threefold higher in the non-surgical group. Similarly, Elias et al. found that patients undergoing resection of liver metastasis form breast cancer had a threefold improvement in survival compared to standard medical treatment [11]. Multiple series report outcomes for resection of noncolorectal gastrointestinal liver metastases. Unfortunately, in a vast majority of articles, gastrointestinal primaries are reported as a group making it impossible to define benefit for a given site [6, 8–10, 22, 24–34, 37]. Our data are consistent with most series that report poor long-term survival after liver resection [6]. Similarly, Slotta et al. found no benefit to resection for gastric, pancreatic, and esophageal liver metastasis, with an approximately 20 % 5-year survival period [38]. Although some subsets, such as small bowel and duodenal cancers, have surprisingly good survival, these data need to be interpreted with great caution because small sample size is likely to consist of highly selected group of patients. Expected survival periods for metastatic pancreatic and esophageal cancer are 4.1–11.1 and 9 months, respectively [52, 53]. The survival times reported after resections in this systematic review are only slightly better (18.9 and 20.6 months). There were more patients with gastric cancer than those with any other non-colorectal gastrointestinal primary. There were a total of 18 cohorts identified representing
481 patients with a median survival of only 20.6 months, lower than that of non-colorectal gastrointestinal metastases. However, three large Japanese series demonstrated an apparent survival advantage to liver resection [54–56]. Factors associated with improved survival included tumors of size less than 5 cm and no serosal involvement. Other authors suggest a benefit for patients with metachronous presentation, solitary tumors, and peritumoral fibrous pseudocapsule on pathology [56–64]. The final category, other, is a diverse group of tumors, including GISTs, sarcoma, melanoma, lung, head and neck, and adrenal primaries. The largest group of metastatic melanoma has 646 patients in 19 studies. These patients have relatively modest survival of 21.8 months, similar to those with non-colorectal gastrointestinal liver metastases. This modest survival may be an improvement compared to non-resection as prognosis for patients with non‐pulmonary visceral metastatic melanoma is poor, with a median survival time of 4– 6 months [65]. There are five studies reporting median survival in 106 patients with GISTs. Most of these data come from an era preceding the effective tyrosine-kinase inhibitor, Imatinib; therefore, the data must be interpreted with caution. Patients with sarcoma and adrenal metastases have a favorable prognosis compared to others in this category, at 53 and 40.9 months, respectively. Survival for patients is poor with unresected metastatic sarcoma with progression commonly occurring within 6 months after treatment and survival rarely exceeding 1 year [66]. Similarly poor outcomes can be expected with metastatic adrenal cancer, with a median survival between 6 and 20 months [17]. Although the
Langenbecks Arch Surg (2014) 399:989–1000 Table 4 Resection of other primary site liver metastases, manuscripts published from 1990 to 2013
Tumor type
Author
Year published
Time span
Patients (#)
Median survival
GIST
DeMatteo [90] Shima [19] Nunobe [91] Xia [92] Bresandola [7]
2001 2003 2005 2009 2011
1982–2000 1989–2001 1984–2003 2005 1989–2006
34 9 18 20 3
28.8 39 36 32.8 40
Sarcoma
Elias [34] DeMatteo [90] Groeschl [6] Zarcherl [93] Salmon [94] Elias [34] Rose [95] Hsueh [96] Rivoire [97] Kodjikian [98] Adam [36]* Adam [36]#
1998 2001 2012 2011 1998 1998 2001 2003 2004 2005 2006
1984–1996 1982–2000 1990–2009 1987–2006 1992–1997 1984–1996 1971–1999 1971–1999 1983–1996 1983–1996 1983–2004
20 22 98 15 19 10 24 24 28 14 44
23 38.4 72 33.6 22 21 28 38 15 25 27
Herman [99] Pawlik [100] Lendoire [33] Reddy [8] Frenkel [101] Woon [102]
2006 2006 2006 2007 2007 2008 2008
1983–2004 1999–2005 1988–2004 1989–2006 1995–2005 1988–2007 1990–2002
104 10 40 6 13 35 7
19 22 28.3 37 41 24 22
Mariani [18] Bresadola [7] Groeshl [6] Du [103] Weitz [37] Adam [36] Adam [36] Weitz [37] Adam [36] Ripley [104] Gaujoux [17]
2009 2011 2012 2012 2005 2006 2006 2005 2006 2011 2012
1991–2007 1989–2006 1990–2009 1984–2010 1981–2002 1983–2004 1983–2004 1981–2002 1983–2004 1979–2009 1978–2009
225 3 31 9 4 32 15 15 28 19 28
14 10 39 30.7 17 16 18 40 63 22.8 31.5
Melanoma
Lung Head and neck Adrenal
Asterisk is cutaneous and pound sign is ocular
995
sample size is small, similar to other investigators, we also observed that patients with lung and head and neck primaries have poor survival at 16 and 18 months, respectively [36]. Although this large systematic review allows us to better understand the treatment of NCRNNE liver metastases, conclusions need to be interpreted in light of the study limitations. This is a systematic review of prospective and retrospective cohort studies of patients who were likely to be highly selected. Survival for less well-selected patients is likely to be shorter. In addition, there also remains a possibility of biased reporting favoring superior outcomes. In addition, there is a lack of granularity inherent to this review. Details regarding tumor histology,
patient comorbidities, demographics, and other clinically germane factors could not be obtained. The authors realize that the approach of comparing groups in terms of differences in expected median survival using a permutation test is not standard; however, in this case, this has been allowed for new insights. The authors recognize the limitation in this approach compared to a standard metaanalysis. A standard meta-analysis would allow for superior comparisons across groups. However, a standard meta-analysis requires similar information across studies, such as proportions, odds ratios, means (with variances), or (in the case of survival data) hazard ratios. Had that information been consistently available, then a standard meta-analysis would provide superior results. Limitations
996
Langenbecks Arch Surg (2014) 399:989–1000
Table 5 Expected median survival after resection of non-colorectal, non-neuroendocrine liver metastasis Category
Tumor type
Data points
Number of patients
Median
Minimum
Maximum
Breast Gastrointestinal
Breast
GIST
27 31 2 1 18 2 3 1 3 1 16 4 4 7 1 35 5
1281 684 23 25 481 55 38 13 21 28 549 153 119 234 43 1082 106
44.3 22.3 16.26 14 20.6 18.09 32.39 28 26.19 58 63.4 61.65 67.8 67.87 32 23.7 31.96
8.0 5 16 14 8.8 13 23 28 5 58 5.4 5.4 26.3 16 32 10.0 28.8
75.0 58 18 14 58 20 38 28 42 58 142 82 98 142 32 72.0 40
Sarcoma Melanoma Lung Head and neck Adrenal
4 19 2 1 4
189 646 36 15 90
53.81 21.77 16.11 18 40.88
23 10 16 18 22.8
72 41 17 18 63
Esophagus GE Junction Gastric Pancreas Duodenum Cholangiocarcinoma Gallbladder Small bowel Genitourinary Testicular Ovary Renal Uterine Other
of this approach include that no funnel plot could be generated, and there was no way to verify that heterogeneous groups were not combined. The authors make some assumptions here, and the largest of which is that studies with greater sample sizes make better estimates of median survival. Note that the only way to definitively define benefit would be via a prospective randomized clinical Fig. 2 Expected median survival for patients undergoing resection of non-colorectal, nonneuroendocrine liver metastasis, studies reported from 1990–2013. Line represents overall expected median survival
trial. The authors contend that this analysis is unique and deemed exploratory in nature, but faced with limited consistent data between studies, it is an important first step in exploring the results across the myriad of studies. Finally, the study period is quite long, and it must be recognized that treatment and imaging techniques have changed over that timeframe.
Langenbecks Arch Surg (2014) 399:989–1000
Conclusions Resection appears to be a benefit for patients with NCRNNE liver metastases. The degree of survival advantage is predicated by the primary site. The greatest survival advantage is for those with GU primaries, particularly those with testicular, ovarian, and renal cell cancer. In addition, resection also has a significant survival advantage for isolated breast metastases to the liver. This benefit is positively impacted by treatment with chemotherapy, response to chemotherapy, and hormone receptor status. Although not clear from the literature, it is likely that similar multimodal modal management may also be beneficial for patients of cancer of other primaries. Patients with sarcoma and adrenal metastases also appear to derive clinical benefit from resection, although the sample size is relatively small. There is a minimal benefit of resection for non-colonic gastrointestinal, lung, and head and neck cancers metastatic to the liver. This has been substantiated in previously published literature. An exception may be well-selected patients with gastric cancer (no serosal involvement,
SYSTEMATIC REVIEWS AND META-ANALYSES
The benefits of liver resection for non-colorectal, non-neuroendocrine liver metastases: a systematic review Timothy L. Fitzgerald & Jason Brinkley & Shannon Banks & Nasreen Vohra & Zachary P. Englert & Emmanuel E. Zervos
Received: 30 April 2014 / Accepted: 11 August 2014 / Published online: 23 August 2014 # Springer-Verlag Berlin Heidelberg 2014
Abstract Purpose Defining the benefits of resection of isolated noncolorectal, non-neuroendocrine (NCRNNE) liver metastases is difficult. To better understand the survival benefit in this group of patients, we conducted a systematic review of the previous literature. Methods Medline, Web of Knowledge, and manual searches were performed using search terms, such as “liver resection” and “primary tumor.” Inclusion criteria were year>1990, >five patients, and median survival reported or derived. An expected median survival was calculated from weighted averages of median survivals, and differences were assessed using a permutation test. Results A total of 7,857 references were identified. Overall 4,735 abstracts were reviewed; 120 manuscripts evaluated and of these, 73 met the study inclusion criteria. The final population consisted of 3,596 patients with renal (n=234), ovarian (n=119), testicular (n=153), adrenal (n=90), small bowel (n= 28), gallbladder (n=21), duodenum (n=38), gastric (n=481), pancreatic (n=55), esophageal (n=23), head and neck (n=15), and lung (n=36) cancers, gastrointestinal stromal tumors (GISTs) (n=106), cholangiocarcinoma (n=13), sarcoma (n= 189), and melanoma (n=643). The greatest expected median was 63 months for genitourinary (GU) primaries (n=549; range 5.4–142 months) followed by 44.4 months for breast cancer (n=1,013; range 8–74 months), 22.3 months for gastrointestinal cancer (n = 549; range 5–58 months), and T. L. Fitzgerald (*) : S. Banks : N. Vohra : Z. P. Englert : E. E. Zervos Division of Surgical Oncology, Department of Surgery, Brody School of Medicine, East Carolina University, 4S24 600 Moye Boulevard, Greenville, NC 27834, USA e-mail: [email protected] J. Brinkley Division of Biostatistics, School of Allied Health, East Carolina University, 4S24 600 Moye Boulevard, Greenville, NC 27834, USA
23.7 months for other tumor types (n=1,082; range 10– 72 months). Using a permutation test, we observed that survival was best for patients with GU primaries followed by that for breast cancer patients. Additionally, we also observed that survival was similar for those with cancer of the GI tract and other primary sites. Conclusions There appears to be a benefit to resection for patients with NCRNNE liver metastases. The degree of survival advantage is predicated by primary site. Keywords Liver metastases . Non-colorectal non-neuroendocrine . Liver surgery . Systematic review
Introduction The clinical utility of resection for hepatic metastasis has engendered significant controversy. Although stage IV colorectal cancer is generally considered incurable, in selected patients with isolated hepatic metastasis there appears to be a survival benefit to surgical resection [1, 2]. Based on that data from large case series, such as the Registry of Hepatic Metastases, there has been a progressive liberalization of resection criteria for isolated liver metastases [3]. Even though no prospective randomized data exist, most hepatic surgeons would consider the ability to render a patient disease-free with an adequate liver remnant is an indication for resection and would expect a 5-year survival in near 60 % cases [4]. Similar results have also been reported for resection of neuroendocrine liver metastases [5]. Defining the benefits of resection of isolated non-colorectal, non-neuroendocrine (NCRNNE) liver metastases has been difficult because outcome data published using tertiary institutional and multi-institutional registries include a myriad of primary sites with little data on specific tumor types [6–10]. Data on individual histologic subtypes frequently come from
990
small case series [11–19]. Therefore, defining benefits of liver resection for individual patients is difficult [6]. A systematic review of the current literature on liver resection for NCRNNE hepatic metastases would assist in filling this knowledge gap. To clarify, a structured review of published studies on liver resection for NCRNNE hepatic metastases was conducted.
Materials and methods Search strategy Original studies published were identified in the following electronic data bases: Medline, Scopus, and Web of Knowledge. The following “free text,” medical subject heading (MeSH), and EMBASE subject headings were used: “liver resection” and “non-colorectal non-neuroendocrine,” “renal cell carcinoma,” “ovarian cancer,” “testicular cancer,” “adrenal cancer,” “small bowel cancer,” “gallbladder cancer,” “duodenal cancer,” “gastrointestinal stromal tumor,” “cholangiocarcinoma,” “gastric cancer,” “pancreatic cancer,” “esophageal cancer,” “sarcoma,” “melanoma,” “head and neck cancer,” or “lung cancer.” In addition, manual searches were also conducted using the bibliography from identified articles. The search was restricted to modern series, defined as the studies conducted from 1990 to February of 2013. No unpublished data were included. Studies were restricted to those in English and on human subjects. Study inclusion criteria Prospective or retrospective studies reporting outcomes on at least five patients were included. The primary goal of this study was better evaluation of survival after resection of hepatic metastases. Due to inconsistent reporting of survival time in various studies (1-, 3-, or 5-year survival), median survival was chosen as the primary endpoint. This is an important point because median survivals are often reported without estimates of variation (that is, standard deviation) or confidence intervals. This information is necessary for a traditional meta-analysis. Given this primary endpoint, studies were excluded if median survival was not reported or could not be extracted. Data extraction and clinical endpoints All studies meeting selection criteria were reviewed by the first and third author (TF and SB) to determine eligibility. Discrepancies were resolved by discussion and consensus. As mentioned above, the primary endpoint of this study was to report median survival after hepatic metastasectomy based on primary tumor type. Median survival was obtained directly from the article or estimated using reported survival. If
Langenbecks Arch Surg (2014) 399:989–1000
outcomes for resection of hepatic metastases were reported for several different primary tumor types, each subtype was treated as an individual cohort. For example, a manuscript on NCRNNE liver metastases might report multiple primary sites for which median survival may be reported for only some groups. Each tumor type for which median survival was reported or derived was treated as a separate cohort. Given this, the number of individual cohorts reported exceeded the number of articles meeting the inclusion criteria. Statistical analysis Survival was calculated using median survival from individual reports. To understand survival after resection, based upon the number of patients in each cohort, an overall weighted average of median survivals was derived (with sample sizes from each study group used as weights). This reflects more accurately what the expected median could have been had all patients been part of one study and also that weighting allows for studies with larger number of patients to contribute more to the expected median survival. This approach was selected because there was a significant variation in the median survival between and within tumor types (a range from 5 to 140 months). To understand variation within and between groups, we provided a minimum and maximum study median (as opposed to traditional confidence bounds). To move beyond simple descriptive statistics and understand whether true differences in outcomes exist, variability was assessed using a permutation test that considered a large random sample of permutations (10,000) of the studies and examined what the difference in expected medians would have been in each permutation. From there, p values were determined based on the proportion of permutations that had more extreme differences than what was observed. Using this statistical approach, differences between the broad groupings of genitourinary (GU), breast, gastrointestinal, and other cancer types were analyzed. Analyses were performed using a combination of SAS and JMP software (versions 9.3 and 11Pro, respectively).
Results A total of 7,857 citations were identified. Out of which, 3,122 citations were excluded on initial review and 4,735 abstracts were fully reviewed; additional 4,621 citations were excluded, and 120 articles were reviewed in full, out of which, 73 met the inclusion criteria (Fig. 1). Some articles reported more than one primary site. To study individual primary sites, each primary site reported in a published study from which median survival could be derived was treated as an individual data point or cohort for statistical analysis. As a result, there were
Langenbecks Arch Surg (2014) 399:989–1000 Fig. 1 Outline of systematic literature review
991 7857 potenal arcles idenfied via Medline, Scopus, Web of Knowledge and manual search 3122 excluded upon review citaon 4742 abstracts fully reviewed 4621Excluded 121 arcles idenfied for full-text reviewed
73 usable studies for analysis
48 Excluded secondary to lack of extractable data, non-reseconal therapy, no median survival, inadequate follow up reported, number of paents or duplicate
112 individual data points
114 cohorts obtained from the 73 articles meeting the inclusion criteria as listed in Tables 1, 2, 3, and 4. Site-specific data points Tables 1, 2, 3, and 4 report the individual cohorts included in this study under the categories of breast, gastrointestinal, genitourinary, and other cancer types. Author last name, year of publication, years spanned in the study, total number of patients, and median survival are listed. In Table 1, median survival for patients undergoing resection of hepatic metastases from breast cancer is reported. A total of 27 individual cohorts were extracted representing 1,281 patients. These studies spanned the years 1982–2010 and were published between 1994 and 2012. The results for patients undergoing resection of liver metastases from primary gastrointestinal tumors are reported in Table 2. There were 31 individual cohorts identified in articles published from 1994 to 2012 spanning the years of 1963– 2011. This specific tumor category has the longest time span over which studies were conducted. A total of 684 patients with primary tumors originating from the esophagus (n=23), gastroesophageal junction (n=25), stomach (n=481), pancreas (n=55), duodenum (n=38), bile duct (n=13), gallbladder (n=21), and small bowel (n=28) were reported. For GU tumors, a total of 16 individual cohorts were identified in the articles published from 1990 to 2012 spanning the years of 1976–2010 (Table 3). A total of 549 patients were reported with the primary tumor types, such as testicular cancer (n=153), ovarian cancer (n=119), renal cancer (n= 234), and uterine cancer (n=43). The final category “other,” including cancer sites and other histologies that were not included in the main three categories (breast, gastrointestinal, and GU), is described in Table 4. There were a total of 35 cohorts representing 1,082 patients in articles published from 1998 to 2012 spanning the years of
1981–2010. Individual tumor types were gastrointestinal stromal tumor (GIST) (n=106), sarcoma (n=189), melanoma (n= 646), lung cancer (n=36), head and neck cancer (n=15), and adrenal carcinoma (n=90). Overall survival To understand the impact of resection for NCRNNE liver metastases on long-term survival, data analysis of all reported or derived median survival was performed. Using the individual median survival and the sample size, an expected median survival was derived (Table 5) for each of the four categories outlined above: breast, gastrointestinal, GU, and others. In each category, data were further reported by primary site (Fig. 2). GU tumors had the longest expected median survival of 63 months. There is a fairly wide range of reported median survivals from a minimum of 5.4 months to a maximum of 142 months for individual cohorts. Subtypes of cancers reported in this category included testicular, ovarian, renal, and uterine carcinoma, with expected median survival of 61.65, 67.8, 67.9, and 32 months, respectively. The category with the next best expected median survival of 44.4 months was breast cancer, with a minimum of 8 months to a maximum of 74 months for individual cohorts. Patients with gastrointestinal primaries and other sites had similar expected median survival of 22.3 and 23.7 months, respectively. Survival differences between categories To determine the impact of hepatic resection on survival across the categories of breast, GU, gastrointestinal and others, a comparison was made, as noted in the “Materials and methods” section, using a permutation test. When comparing GU to breast, gastrointestinal, and other primary sites, we observed a significant improvement in survival with pvalues of 0.002, 0.014, and 0.002, respectively. A survival
992 Table 1 Resected breast cancer liver metastases, manuscripts published from 1990 to 2013
Langenbecks Arch Surg (2014) 399:989–1000
Author
Year published
Time span
Patients (number)
Median survival
Schneebaum [46] Elias [11] Raab [67] Elias [34] Seifert [68]
1994 1995 1998 1998 1999
1982–1990 1986–1994 1983–1994 1984–1996 1985–1997
6 21 34 35 15
42 26 27 40 57
Pocard [50] Selzner [47] Yoshimoto [69] Maksan [42] Kondo [70] Carlini [16] Vlastos [71] Sakamoto [72] Yedibela [35] Yedibela [35] Adam [36] Adam [43] Lendoire [33] Thelen [73] Lubrano [74]
2000 2000 2000 2000 2000 2002 2004 2005 2005 2005 2006 2006 2007 2007
1988–1997 1987–1999 1985–1998 1984–1998 1990–1999 1990–1999 1991–2002 1985–2003 1978–1989 1990–2001 1983–2004 1984–2004 1989–2006 1988–2006
49 17 25 9 6 17 31 34 7 17 454 85 19 39
42 24 34.3 60 28 53 25 36 8 40 45 32 75 53.6
Reddy [8] Caralt [75] Hoffmann [49]
2007 2007 2008 2010
1989–2004 1995–2005 1988–2006 1999–2008
16 20 12 41
42 67 35.9 58
Bresadola [7] Abbott [76] Groeschl [6] Walsum [48]
2011 2011 2012 2012
1989–2006 1997–2010 1990–2009 1991–2010
13 59 115 32
44 57 52 55
advantage was also noted when comparing breast cancer to gastrointestinal (p = 0.001) and other primary sites (p = 0.0312). There was no difference in expected median survivals when comparing gastrointestinal cancers to other types of cancers (p=0.3118).
Discussion Resection of the liver metastases has engendered a fair degree of controversy. In the early 1990s, there was a considerable debate regarding the efficacy of resection for colorectal liver metastases [20]. There is a little debate now that patients benefit from resection for isolated hepatic disease [4]. Similar data exist for resection of neuroendocrine liver metastases [21]. Controversy, however, continues regarding resection of NCRNNE liver metastasis. Small institutional series reporting outcomes for resection of NCRNNE liver metastases generally include an array of primary sites making it difficult to define the benefit of resection for individual patients [6–10, 22–36]. Despite the limitations of these data, it is generally seen that
outcomes are poor for patients with non-colorectal gastrointestinal primaries [32, 37]. This is further highlighted by a recent case-controlled study that found no difference between observation and resection for this group [38]. The authors did, however, report an advantage to resection versus observation for patients with non-gastrointestinal primaries. In this analysis, patients with GU primaries (Tables 3 and 5) had the greatest expected median survival compared to all other subsets, followed by patients with breast, gastrointestinal, and other primary sites of disease; gastrointestinal and other primaries had a similarly poorer expected median survival. There were 549 individuals with resected GU liver metastases reported in 16 cohorts with primary sites, including testicular (70 to 80 % of patients cured of their disease), ovarian, renal, and uterine cancer. The expected median survival was fairly high at 63.4 months with a high degree of variation (5–142 months). Patients with ovarian, renal, and testicular metastases appeared to have superior survival duration as compared to those with uterine cancer metastases (Table 5). The findings of this analysis are in agreement with those of the published
Langenbecks Arch Surg (2014) 399:989–1000
993
Table 2 Resected gastrointestinal primary metastases, manuscripts published from 1990 to 2013 Tumor type
Author
Year published
Time span
Patients (number)
Median survival
Esophagus
Adam [36] Bresadola [7] Adam [36] Ochiai [56] Taniguchi [77] Miyazaki [78] Elias [34] Ambiru [58] Imamura [79] Saiura [80] Zacherl [57] Sakamoto [62] Roh [61] Adam [36] Sakamoto [55] Bresadola [7] Schildberg [60]
2006 2011 2006 1994 1997 1997 1998 2001 2001 2002 2002 2003 2005 2006 2007 2011 2012
1983–2004 1989–2006 1983–2004 1963–1991 1998–1994 1980–1994 1984–1996 1975–1999 1990–1997 1981–1998 1980–1999 1985–2001 1988–1996 1983–2004 1990–2005 1989–2006 1972–2008
20 3 25 21 6 21 11 40 17 10 15 22 11 64 37 10 31
16 18 14 58 21.2 18.6 18 12 12.6 25 8.8 21 19 15 31 16 15.2
Takemura [54] Garancini [63] Miki [81]
2012 2012 2012
1993–2011 1998–2007 1995–2009
64 21
34 11
Wang [64] Adam [36] De Jong [15] Adam [36] Adam [36] De Jong [15] Hemming [82] Elias [34] Hemming [82] Bresasola [7] Adam [36]
2012 2006 2010 2006 2006 2010 2010 1998 2010 2011 2006
2003–2008 1983–2004 1993–2009 1983–2004 1983–2004 1993–2009 1996–2009 1984–1996 1996–2009 1989–2006 1983–2004
50 30 40 15 15 12 11 13 7 9 5 28
15 11 20 13 38 34 23 28 21 42 5 58
GE Junction Gastric
Pancreas Duodenum
Cholangiocarcinoma Gallbladder
Small bowel
literature. The largest series, a multi-institutional registry by Adam, reported similar median survival with GU metastases classified as a favorable subtype for hepatic resection. These outcomes appear to be superior to what one would expect with observation alone. Although survival is excellent with testicular cancer, hepatic resection should be reserved for only those poor prognostic patients with medically refractory disease [39]. It is assumed that such patients would have progressed without intervention. Expected survival for a patient with ovarian cancer with visceral metastasis is only 10.5 months. Prognosis is similarly poor for patients with metastatic uterine cancer (12.3–15 months) and renal cell cancer (10–20 months) with modern chemotherapy regimens [40, 41]. In this systematic review, liver resection was most frequently reported for breast carcinoma metastases. A sample
of 27 cohorts representing data from 1,281 patients was identified (Table 1). The expected median survival was 44.3 months (Table 5). There was a significant variation in survival among various studies (8–75 months). As mentioned above, patients with breast cancer liver metastases have an improved survival compared to those with gastrointestinal or other tumor types. These data are consistent with other studies [42, 43]. A thoughtful workup of such patients is imperative, as the incidence of isolated hepatic recurrence is unusual [44]. It is necessary to routinely work up such patients with positron emission tomography to ensure absence of disease at other sites. In addition, these patients also appear to have an increased survival if treated with hormonal or systemic chemotherapy in addition to resection [45–47]. Other factors associated with improved survival include lymph node negative disease, hormone receptor positivity,
994 Table 3 Resected genitourinary liver metastases, manuscripts published from 1990 to 2013
Langenbecks Arch Surg (2014) 399:989–1000
Tumor type
Author
Year published
Time span
Patients (#)
Median survival
Testicular
Goulet [83] Elias [34] Adam [36] You [13] Merideth [14]
1990 1998 2006 2009 2003
1976–1989 1984–1996 1983–2004 1975–2002 1976–1999
14 20 78 4 26
64.8 23 82 55.2 26.3
Adam [36] Abood [84] Roh [85] Stief [86] Alves [12] Weitz [37] Adam [36] Staehler [87] Ruys [88] Langan [89] Adam [36]
2006 2007 2011 1997 2003 2005 2006 2010 2011 2012 2006
1983–2004 1998–2006 1991–2008 1983–1993 1982–2001 1981–2002 1983–2004 1995–2006 1990–2008 1980–2010 1983–2004
65 10 18 13 14 11 85 68 33 10 43
98 33 38 16 26 115 36 142 33 24 32
Ovary
Renal
–
response to preoperative chemotherapy, solitary liver metastases, and prolonged disease-free interval [11, 42, 47–50]. Survival after liver resection for breast cancer appears to be better than one would expect with observation alone. Marinai et al. reported an 80.7 % 3-year survival period after surgical resection versus 50.89 % in a matched cohort not undergoing resection [51]. On a multivariate analysis, the risk for death was threefold higher in the non-surgical group. Similarly, Elias et al. found that patients undergoing resection of liver metastasis form breast cancer had a threefold improvement in survival compared to standard medical treatment [11]. Multiple series report outcomes for resection of noncolorectal gastrointestinal liver metastases. Unfortunately, in a vast majority of articles, gastrointestinal primaries are reported as a group making it impossible to define benefit for a given site [6, 8–10, 22, 24–34, 37]. Our data are consistent with most series that report poor long-term survival after liver resection [6]. Similarly, Slotta et al. found no benefit to resection for gastric, pancreatic, and esophageal liver metastasis, with an approximately 20 % 5-year survival period [38]. Although some subsets, such as small bowel and duodenal cancers, have surprisingly good survival, these data need to be interpreted with great caution because small sample size is likely to consist of highly selected group of patients. Expected survival periods for metastatic pancreatic and esophageal cancer are 4.1–11.1 and 9 months, respectively [52, 53]. The survival times reported after resections in this systematic review are only slightly better (18.9 and 20.6 months). There were more patients with gastric cancer than those with any other non-colorectal gastrointestinal primary. There were a total of 18 cohorts identified representing
481 patients with a median survival of only 20.6 months, lower than that of non-colorectal gastrointestinal metastases. However, three large Japanese series demonstrated an apparent survival advantage to liver resection [54–56]. Factors associated with improved survival included tumors of size less than 5 cm and no serosal involvement. Other authors suggest a benefit for patients with metachronous presentation, solitary tumors, and peritumoral fibrous pseudocapsule on pathology [56–64]. The final category, other, is a diverse group of tumors, including GISTs, sarcoma, melanoma, lung, head and neck, and adrenal primaries. The largest group of metastatic melanoma has 646 patients in 19 studies. These patients have relatively modest survival of 21.8 months, similar to those with non-colorectal gastrointestinal liver metastases. This modest survival may be an improvement compared to non-resection as prognosis for patients with non‐pulmonary visceral metastatic melanoma is poor, with a median survival time of 4– 6 months [65]. There are five studies reporting median survival in 106 patients with GISTs. Most of these data come from an era preceding the effective tyrosine-kinase inhibitor, Imatinib; therefore, the data must be interpreted with caution. Patients with sarcoma and adrenal metastases have a favorable prognosis compared to others in this category, at 53 and 40.9 months, respectively. Survival for patients is poor with unresected metastatic sarcoma with progression commonly occurring within 6 months after treatment and survival rarely exceeding 1 year [66]. Similarly poor outcomes can be expected with metastatic adrenal cancer, with a median survival between 6 and 20 months [17]. Although the
Langenbecks Arch Surg (2014) 399:989–1000 Table 4 Resection of other primary site liver metastases, manuscripts published from 1990 to 2013
Tumor type
Author
Year published
Time span
Patients (#)
Median survival
GIST
DeMatteo [90] Shima [19] Nunobe [91] Xia [92] Bresandola [7]
2001 2003 2005 2009 2011
1982–2000 1989–2001 1984–2003 2005 1989–2006
34 9 18 20 3
28.8 39 36 32.8 40
Sarcoma
Elias [34] DeMatteo [90] Groeschl [6] Zarcherl [93] Salmon [94] Elias [34] Rose [95] Hsueh [96] Rivoire [97] Kodjikian [98] Adam [36]* Adam [36]#
1998 2001 2012 2011 1998 1998 2001 2003 2004 2005 2006
1984–1996 1982–2000 1990–2009 1987–2006 1992–1997 1984–1996 1971–1999 1971–1999 1983–1996 1983–1996 1983–2004
20 22 98 15 19 10 24 24 28 14 44
23 38.4 72 33.6 22 21 28 38 15 25 27
Herman [99] Pawlik [100] Lendoire [33] Reddy [8] Frenkel [101] Woon [102]
2006 2006 2006 2007 2007 2008 2008
1983–2004 1999–2005 1988–2004 1989–2006 1995–2005 1988–2007 1990–2002
104 10 40 6 13 35 7
19 22 28.3 37 41 24 22
Mariani [18] Bresadola [7] Groeshl [6] Du [103] Weitz [37] Adam [36] Adam [36] Weitz [37] Adam [36] Ripley [104] Gaujoux [17]
2009 2011 2012 2012 2005 2006 2006 2005 2006 2011 2012
1991–2007 1989–2006 1990–2009 1984–2010 1981–2002 1983–2004 1983–2004 1981–2002 1983–2004 1979–2009 1978–2009
225 3 31 9 4 32 15 15 28 19 28
14 10 39 30.7 17 16 18 40 63 22.8 31.5
Melanoma
Lung Head and neck Adrenal
Asterisk is cutaneous and pound sign is ocular
995
sample size is small, similar to other investigators, we also observed that patients with lung and head and neck primaries have poor survival at 16 and 18 months, respectively [36]. Although this large systematic review allows us to better understand the treatment of NCRNNE liver metastases, conclusions need to be interpreted in light of the study limitations. This is a systematic review of prospective and retrospective cohort studies of patients who were likely to be highly selected. Survival for less well-selected patients is likely to be shorter. In addition, there also remains a possibility of biased reporting favoring superior outcomes. In addition, there is a lack of granularity inherent to this review. Details regarding tumor histology,
patient comorbidities, demographics, and other clinically germane factors could not be obtained. The authors realize that the approach of comparing groups in terms of differences in expected median survival using a permutation test is not standard; however, in this case, this has been allowed for new insights. The authors recognize the limitation in this approach compared to a standard metaanalysis. A standard meta-analysis would allow for superior comparisons across groups. However, a standard meta-analysis requires similar information across studies, such as proportions, odds ratios, means (with variances), or (in the case of survival data) hazard ratios. Had that information been consistently available, then a standard meta-analysis would provide superior results. Limitations
996
Langenbecks Arch Surg (2014) 399:989–1000
Table 5 Expected median survival after resection of non-colorectal, non-neuroendocrine liver metastasis Category
Tumor type
Data points
Number of patients
Median
Minimum
Maximum
Breast Gastrointestinal
Breast
GIST
27 31 2 1 18 2 3 1 3 1 16 4 4 7 1 35 5
1281 684 23 25 481 55 38 13 21 28 549 153 119 234 43 1082 106
44.3 22.3 16.26 14 20.6 18.09 32.39 28 26.19 58 63.4 61.65 67.8 67.87 32 23.7 31.96
8.0 5 16 14 8.8 13 23 28 5 58 5.4 5.4 26.3 16 32 10.0 28.8
75.0 58 18 14 58 20 38 28 42 58 142 82 98 142 32 72.0 40
Sarcoma Melanoma Lung Head and neck Adrenal
4 19 2 1 4
189 646 36 15 90
53.81 21.77 16.11 18 40.88
23 10 16 18 22.8
72 41 17 18 63
Esophagus GE Junction Gastric Pancreas Duodenum Cholangiocarcinoma Gallbladder Small bowel Genitourinary Testicular Ovary Renal Uterine Other
of this approach include that no funnel plot could be generated, and there was no way to verify that heterogeneous groups were not combined. The authors make some assumptions here, and the largest of which is that studies with greater sample sizes make better estimates of median survival. Note that the only way to definitively define benefit would be via a prospective randomized clinical Fig. 2 Expected median survival for patients undergoing resection of non-colorectal, nonneuroendocrine liver metastasis, studies reported from 1990–2013. Line represents overall expected median survival
trial. The authors contend that this analysis is unique and deemed exploratory in nature, but faced with limited consistent data between studies, it is an important first step in exploring the results across the myriad of studies. Finally, the study period is quite long, and it must be recognized that treatment and imaging techniques have changed over that timeframe.
Langenbecks Arch Surg (2014) 399:989–1000
Conclusions Resection appears to be a benefit for patients with NCRNNE liver metastases. The degree of survival advantage is predicated by the primary site. The greatest survival advantage is for those with GU primaries, particularly those with testicular, ovarian, and renal cell cancer. In addition, resection also has a significant survival advantage for isolated breast metastases to the liver. This benefit is positively impacted by treatment with chemotherapy, response to chemotherapy, and hormone receptor status. Although not clear from the literature, it is likely that similar multimodal modal management may also be beneficial for patients of cancer of other primaries. Patients with sarcoma and adrenal metastases also appear to derive clinical benefit from resection, although the sample size is relatively small. There is a minimal benefit of resection for non-colonic gastrointestinal, lung, and head and neck cancers metastatic to the liver. This has been substantiated in previously published literature. An exception may be well-selected patients with gastric cancer (no serosal involvement,
