Journal of Surgical Oncology 2014;110:129–135
Is Microsatellite Instability a Prognostic Marker in Gastric Cancer?: A Systematic Review With Meta-Analysis YOON YOUNG CHOI, MD,1 JUNG MIN BAE, MD,2 JI YEONG AN, MD, PhD,1 IN GYU KWON, MD,1 IN CHO, MD,1 HYUN BEAK SHIN, MD,1 TANAKA EIJI, MD, PhD,1,3 MOHAMMAD ABURAHMAH, MD,1,4 HYUNG-IL KIM, MD, PhD,1 JAE-HO CHEONG, MD, PhD,1 WOO JIN HYUNG, MD, PhD,1 1,5 AND SUNG HOON NOH, MD, PhD * 1
Department of Surgery, Yonsei University Health System, Seoul, Korea Department of Dermatology, College of Medicine, The Catholic University of Korea, Seoul, Korea 3 Department of Surgery, Kyoto University, Kyoto, Japan 4 Department of Surgery, King Faisal Specialist Hospital and Research Centre, Alfaisal University College of Medicine, Riyadh, Saudi Arabia 5 Brain Korea 21 PLUS Project for Medical Science, Yonsei University College of Medicine, Seoul, Korea 2
Background and Objective: The relationship between survival in gastric cancer patients and the status of microsatellite instability (MSI) has not yet been established. The purpose of this meta‐analysis was to obtain integrated and more precise data for the value of MSI as a prognostic marker in gastric cancer. Methods: A comprehensive systematic review and meta‐analysis were conducted using major electronic databases (PubMed, EMBASE, and the Cochrane Central) with keywords related to “microsatellite instability,” “gastric cancer,” and “prognosis.” Results: Twenty‐four studies with 5,438 participants (712 cases were MSI gastric cancer) were included for pooling risk estimates of MSI in gastric cancer. Seventeen studies reported overall survival. The pooled hazard ratio (HR) for overall survival of MSI vs. non‐MSI was 0.72 (95%CI: 0.59– 0.88, P ¼ .001) in a random‐effects model. In the sensitivity analysis, the result from the most recent study showed the most heterogeneity. Conclusion: MSI gastric cancer was associated with good prognosis but there was heterogeneity in the recent studies. Changed epidemiology and effects of chemotherapy are potential causes of heterogeneity. Establishing a consensus for defining MSI in gastric cancer should be preferred for future studies.
J. Surg. Oncol. 2014;110:129–135. ß 2014 Wiley Periodicals, Inc.
KEY WORDS: MSI; gastric cancer; prognosis
INTRODUCTION
MATERIALS AND METHODS
Despite decreasing incidence in Western and Eastern countries, gastric cancer is still an important world health problem [1]. To identify a genetic marker of gastric cancer, innumerable genetic alterations have been identified; however, a definitive role for these alterations is still being debated. Microsatellite instability (MSI), a genetic alteration caused by genetic and epigenetic inactivation of DNA mismatch repair genes, has been thought to be related to carcinogenesis [2]. MSI‐high (MSI‐H) cancers are considered to have different clinic‐pathologic and molecular characteristics than MSI‐low (MSI‐L) or MS‐stable (MSS) cancers, regardless of their hereditary or sporadic origins, because the process of carcinogenesis would be different in each case. Because it is known that MSI is related to hereditary non‐polyposis colorectal cancer, the role of MSI in colorectal cancer is relatively well established. The characteristics of MSI tumors in colorectal cancers have been identified. These tumors tend to be more proximal, poorly differentiated, and mucinous, with less lymph node metastasis and a better prognosis [3–5]. In gastric cancer, many studies have suggested that MSI tumors are associated with intestinal type Lauren classification and less lymph node metastasis [6,7]. However, the relationship between survival in gastric cancer patients and MSI has yet to be determined, because the results of analysis of the prognostic value of MSI among studies have been discordant. To obtain integrated and more precise data for the value of MSI as a prognostic marker in gastric cancer, we performed a comprehensive review of the literature and a meta‐analysis of studies to date.
Search Strategy and Selection Criteria
ß 2014 Wiley Periodicals, Inc.
This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) statement [8]. We searched the major electronic databases PubMed, EMBASE, and The Cochrane Central Register of Controlled Trials as of October 11, 2012, using search terms related to microsatellite instability, gastric cancer, and prognosis. Details of the PubMed search strategy are shown in the Appendix. Bibliographies of review articles and other related studies were searched manually to identify additional studies. The selection criteria followed the Population, Intervention, Comparator, Outcome, Setting, and Study Design (PICOTS‐SD) format, with gastric cancer patients who underwent gastrectomy as population, MSI status as intervention and comparator, and any kind of
Yoon Young Choi and Jung Min Bae contributed equally to this work. All authors have no conflict of interest and no financial disclosures to disclose. *Correspondence to: Sung Hoon Noh, MD, Department of Surgery, Yonsei University College of Medicine, 50 Yonsei‐Ro, Seodaemun‐gu, 120‐752 Seoul, Korea. Fax: þ82‐2‐313‐8289. E‐mail: [email protected] Received 16 January 2014; Accepted 17 March 2014 DOI 10.1002/jso.23618 Published online 15 April 2014 in Wiley Online Library (wileyonlinelibrary.com).
130
Choi et al.
survival as outcome. Because the major interest of this review was prognosis, that is, overall survival (OS), disease‐specific survival (DSS), or disease‐free survival (DFS), only studies reporting information on survival outcomes were included. Only studies published in peer‐ reviewed journals were included; review articles, abstracts, and editorial comments were not sought. There was no limitation on publication language or date, scale of the studies, or study design. When data was duplicated in multiple studies, the most reasonable study was included for a meta‐analysis. This comprehensive search and selection of studies was carried out by two authors (Y.Y.C. and J.M.B.) independently, and disagreements were resolved by discussion.
reviewing the titles and abstracts. Twenty‐four full text articles were excluded for the following reasons. One was a conference abstract [11], seven duplicated data [12–18], one was a review article [19], one was an editorial comment [20], three investigated only the response to chemotherapy without gastrectomy [21–23], three reported no survival results [24–26], one failed to extract survival outcome [27], and seven reported no information about MSI [28–34]. To obtain the HR for one study [27] that we were unable to extract from the article for meta‐analysis, we contacted the author by e‐mail but received no reply. Finally, 24 studies [6,7,35–56] with 5,438 participants (712 patients with MSI gastric cancer) from 14 countries were included for pooling risk estimates (Supplement Fig. S1).
Data Extraction and Assessment of Individual Outcomes Each included study was read by two authors (Y.Y.C. and J.M.B.) and the following data were extracted: publication year and country, enrollment period, total number of patients assessed in survival analysis, the range of gastric cancer stages, total number of markers and number of mononucleotides used to estimate MSI status, threshold to assign MSI, the proportion of MSI tumors, median or mean duration of follow‐up, control and intervention groups for estimation of hazard ratio (HR), and the type of survival outcomes. If the HR and 95% confidence interval (CI) were not available in a study, these were indirectly estimated from a Kaplan–Meier curve [9,10] using Plot Digitizer version 2.5.1 (http://plotdigitizer.sourceforge.net/). In addition, to analyze associations between the effect of MSI status and study characteristics, the proportions of lower and upper body gastric cancer, cases beyond stage III, female patients, intestinal type, cases beyond stage pN1, and mean age were extracted for each included study. When more detailed data were required for the analysis, we contacted the authors directly by e‐mail.
Statistical Analysis Analysis was conducted using Review Manager Ver. 5.1.7 (Nordic Cochrane Center, Copenhagen, Denmark) and STATA version 11.2 (Stata Corp, College Station, TX). For an appropriate analysis of time‐ to‐event outcomes, OS, DSS, or DFS, a generic inverse–variance method, was applied in this meta‐analysis. When the unadjusted or adjusted HR was available in a study, the log HR and its standard error (SE) were used directly with the generic inverse–variance method. Otherwise, log HR and SE were indirectly estimated using the method proposed by Parmar and coworkers [9,10]. HRs adjusted for other variables such as gastric cancer stage, age, and sex were assigned a higher priority for use in the meta‐analysis when provided. The heterogeneity of effect sizes across the studies was assessed by Cochran’s Q‐test and Higgin’s I2. A Cochran’s Q P‐value of less than 0.10 or an I2 value over 50% was considered substantial heterogeneity. When heterogeneity was suspected across the included studies, a random‐effects model was used to pool the data. Otherwise, a fixed‐ effects model was applied. The funnel plot summary approach was used to evaluate the presence of reporting bias. A sensitivity analysis was performed to examine whether the overall findings were robust to one or two outlying studies. For the sensitivity analysis, sequential meta‐ analyses were repeated, in which each study was excluded. Also, we performed meta‐regression to assess the associations between the effect of the MSI and study characteristics such as the proportion of lymph node positive population, stage III & IV gastric cancer, upper third gasric cancer, and so on in each study.
RESULTS Included Studies Of 392 records identified through database searching and removal of duplicates, 48 articles were assessed in full text for eligibility after Journal of Surgical Oncology
Characteristics of Studies Characteristics of the included studies are presented in Supplement Table SI. Eight of 24 studies were from Eastern countries [Korea (3) [6,54,56], Japan (3) [39,47,53], China (1) [55], and Taiwan (1) [45]], and the remainder were from Western countries [Italy (5) [7,42,43,50,52], USA (3) [40,46,49], Slovenia (1) [51], Brazil (1) [48], Poland (1) [44], Spain (1) [41], Australia (1) [38], Germany (1) [37], Portugal (1) [36], and the UK (1) [35]]. Only three studies [54– 56] enrolled patients during the 21st century, and all studies analyzed patients with stage I–IV gastric cancer, with the exception of two studies (one study [35] included stage I–III, and one study [45] did not provide information on stage). The stage of gastric cancer was determined based on the 4th through 6th editions of the American Joint of Committee on Cancer, and in some cases on Japanese classification of gastric carcinoma [57]. The populations assessed for survival ranged in size from 24 [48] to 1990 [56]. The number of markers used to estimate MSI status ranged from one to eleven, and a diversity of thresholds for assigning MSI‐H or MSI was applied among the studies. Thirteen studies [7,36,37,39–41,43,47,49,51,54–56] used a combination of mono‐ and dinucleotide markers, seven [6,38,42,44,45,50,52] used only mononucleotide markers, and three [35,48,53] used only dinucleotide markers. In two studies [38,45], only one mononucleotides marker, BAT26, was used for estimating MSI status, and one study [46] considered hMLH1‐negative cases MSI‐H. Only four studies [49,54–56] followed the National Cancer Institute (NCI) guidelines for the determination of MSI in colorectal cancer [58] (i.e., two mononucleotide markers [BAT25 and BAT26] and three dinucleotide markers [D5S346, D2S123, and D17S250]). The reported proportion of tumors with MSI ranged from 8.5% [56] to 37.8% [41] (median and range, 10.8% and 8.5%–19.2% in Eastern studies, and 17.6% and 10%–37.8% in Western studies). Median follow‐ up period was 18–92.1 months; 10 studies [35,38–42,45,47,48,51] did not report follow‐up duration for survival. The OS was available for 17 studies [7,36–42,44,45,47–49,51–53,56]; DSS was reported in 5 studies [6,35,43,46,50]; and DFS was estimated in 4 studies [48,54–56]. The characteristics of gastric cancer in each study are shown in Supplement Table SII. The proportion of lower body gastric cancer ranged from 27.0% [41] to 63.0% [45], and that of upper body cancer ranged from 7.3% [46] to 37.0% [51]. The proportion of stage III–IV cancer was relatively low in recent studies from Korea [54,56], and the proportion of female patients was between 30% and 40% in most studies. The proportion of patients who had metastases to lymph nodes was over 40% in all studies that reported that information, with one exception [49]. It was strongly suspected that, in this exception, the presence of metastasis had been inadvertently reported as the absence of metastasis; however, it was impossible to confirm this because we failed to receive a reply from the author. Six studies [7,39,42,43,46,50] reported an adjusted HR (the HR in five studies was adjusted based on gastric cancer stages), and the log HR was extracted from raw data in two studies (one study [47] reported the raw data in the article, and it was possible to obtain the raw data from the author in one study [56]). The
MSI and Gastric Cancer HR was estimated from a Kaplan–Meier curve in 14 studies [6,35– 37,40,41,44,45,48,49,51–54] and from observed and expected events in two studies [38,55]. Ten studies reported that the prognosis of the MSI group was not significantly different from the non‐MSI group, and 14 studies reported that the prognosis of the MSI group was better than that of others (Supplement Table SII).
The Pooled Risk of MSI The pooled HR of MSI gastric cancer for OS from 17 studies [7,36– 42,44,45,47–49,51–53,56] was 0.76 (95%CI: 0.65–0.88, P ¼ .0003) in a fixed‐effects model (I2 ¼ 34%, Cochran’s Q P ¼ .08; Fig. 1), and 0.72 (95%CI: 0.59–0.88, P ¼ .001) in a random‐effects model. The pooled HR of MSI gastric cancer for DSS from five studies [6,35,43,46,50] was 0.55 (95%CI: 0.43–0.71, P < .0001) in a fixed‐effects model (I2 ¼ 0%, Cochran’s Q P ¼ .45; Fig. 2A), and the pooled HR of MSI for DFS from four studies [48,54–56] was 1.03 (95%CI: 0.71–1.48, P ¼ .90) in a fixed‐ effects model (I2 ¼ 39%, Cochran’s Q P ¼ .18, Fig. 2B). The OS adjusted by stage was available in five studies [7,39,42,47,56] and the pooled HR of MSI was 0.59 (95%CI: 0.33–1.06, P ¼ .08) in a random‐ effects model (I2 ¼ 75%, Cochran’s Q P ¼ .003, Fig. 2C). Sensitivity analyses showed that the influence of each study on the pooled HR of MSI gastric cancer was similar for all but one study [56] (Fig. 3A); when An et al. (KR) was removed from meta‐analysis, the HR of MSI for OS improved (HR: 0.68, 95%CI: 0.57–0.80). With the 17 studies that reported the OS of MSI, meta‐regression was conducted to identify the relationship between baseline characteristics of each study and the log HR for MSI. The prognosis MSI was better when a study had higher proportions of lymph node‐ positive cases (coefficient: 0.013, P ¼ .011), stage III–IV cases (coefficient: 0.013, P ¼ .008), and female patients (coefficient: 0.047, P ¼ .043; Table I). The effect of the proportion of lymph node‐positive cases became more prominent (coefficient: 0.020,
P < .001) when the analysis was done assuming that the proportion of patients beyond stage pN1 in one study (An et al., US) [49] was not 17.3%, as reported, but 82.7%. Other characteristics, such as the proportions of MSI, upper body cancer, lower body cancer, and intestinal type, and mean age and year of publication did not show any significant relationship with the prognosis of MSI. The funnel plot for reporting bias showed symmetrical effect sizes (Fig. 3B), which suggesting no evidence of publication bias. However, one study [56] was out of the lines and the study was the heterogeneous study in a sensitivity analysis.
DISCUSSION The results of the present meta‐analysis showed that gastric cancer with MSI is associated with a better prognosis than gastric cancer without MSI (HR: 0.76, 95%CI: 0.65–0.88). This result is similar to that seen with colorectal cancer [5], and MSI was a significant prognostic factor in gastric cancer as well. However, sensitivity analysis revealed that the most recent study, with the largest number of patients, from Korea (An et al., KR) [56] reported the most heterogeneous result, the direction of prognosis was favor to without MSI gastric cancer (Fig. 3A,B). Possible reasons for this heterogeneity include (1) changed epidemiology of gastric cancer or (2) different effects of chemotherapy in gastric cancer with and without MSI. The most heterogeneous study (An et al., KR) [56] had a low proportion of stage III–IV cancer, upper body lesions, and lymph node metastasis. Another recent study [54] from Korea, which had similar characteristics to those of An et al. (KR), also reported an unfavorable DFS for MSI gastric cancer. Because mass screening for gastric cancer has recently become popular [59], the incidence of early gastric cancer is relatively high in Korea. And these two studies enrolled the patients after 21st century. The trend between the baseline characteristics of a study and the log HR of MSI from the meta‐regression (Table I) showed that
Fig. 1. Data and forest plot of the effects of microsatellite instability on overall survival. Journal of Surgical Oncology
131
132
Choi et al.
Fig. 2. Data and forest plot of the effects of microsatellite instability. A: Disease‐specific survival. B: Disease‐free survival. C: Overall survival adjusted by stage.
Fig. 3. Sensitivity analysis and funnel plot. A: Effect of each study on the pooled HR of overall survival. B: Funnel plot of overall survival for estimation of publication bias. Journal of Surgical Oncology
MSI and Gastric Cancer TABLE I. The Results of Meta‐Regression for Log Hazard Ratio of MSI in Each Study
Proportion of LN‐positivea Proportion of LN‐positiveb Proportion of MSI tumors Proportion of LB Proportion of UB Proportion of stage III–IV Proportion of females Proportion of intestinal type Mean age Publication year
Coefficient
P‐Value
0.013 0.020 0.022 0.006 0.020 0.013 0.047 0.003 0.037 0.024
0.011
Is Microsatellite Instability a Prognostic Marker in Gastric Cancer?: A Systematic Review With Meta-Analysis YOON YOUNG CHOI, MD,1 JUNG MIN BAE, MD,2 JI YEONG AN, MD, PhD,1 IN GYU KWON, MD,1 IN CHO, MD,1 HYUN BEAK SHIN, MD,1 TANAKA EIJI, MD, PhD,1,3 MOHAMMAD ABURAHMAH, MD,1,4 HYUNG-IL KIM, MD, PhD,1 JAE-HO CHEONG, MD, PhD,1 WOO JIN HYUNG, MD, PhD,1 1,5 AND SUNG HOON NOH, MD, PhD * 1
Department of Surgery, Yonsei University Health System, Seoul, Korea Department of Dermatology, College of Medicine, The Catholic University of Korea, Seoul, Korea 3 Department of Surgery, Kyoto University, Kyoto, Japan 4 Department of Surgery, King Faisal Specialist Hospital and Research Centre, Alfaisal University College of Medicine, Riyadh, Saudi Arabia 5 Brain Korea 21 PLUS Project for Medical Science, Yonsei University College of Medicine, Seoul, Korea 2
Background and Objective: The relationship between survival in gastric cancer patients and the status of microsatellite instability (MSI) has not yet been established. The purpose of this meta‐analysis was to obtain integrated and more precise data for the value of MSI as a prognostic marker in gastric cancer. Methods: A comprehensive systematic review and meta‐analysis were conducted using major electronic databases (PubMed, EMBASE, and the Cochrane Central) with keywords related to “microsatellite instability,” “gastric cancer,” and “prognosis.” Results: Twenty‐four studies with 5,438 participants (712 cases were MSI gastric cancer) were included for pooling risk estimates of MSI in gastric cancer. Seventeen studies reported overall survival. The pooled hazard ratio (HR) for overall survival of MSI vs. non‐MSI was 0.72 (95%CI: 0.59– 0.88, P ¼ .001) in a random‐effects model. In the sensitivity analysis, the result from the most recent study showed the most heterogeneity. Conclusion: MSI gastric cancer was associated with good prognosis but there was heterogeneity in the recent studies. Changed epidemiology and effects of chemotherapy are potential causes of heterogeneity. Establishing a consensus for defining MSI in gastric cancer should be preferred for future studies.
J. Surg. Oncol. 2014;110:129–135. ß 2014 Wiley Periodicals, Inc.
KEY WORDS: MSI; gastric cancer; prognosis
INTRODUCTION
MATERIALS AND METHODS
Despite decreasing incidence in Western and Eastern countries, gastric cancer is still an important world health problem [1]. To identify a genetic marker of gastric cancer, innumerable genetic alterations have been identified; however, a definitive role for these alterations is still being debated. Microsatellite instability (MSI), a genetic alteration caused by genetic and epigenetic inactivation of DNA mismatch repair genes, has been thought to be related to carcinogenesis [2]. MSI‐high (MSI‐H) cancers are considered to have different clinic‐pathologic and molecular characteristics than MSI‐low (MSI‐L) or MS‐stable (MSS) cancers, regardless of their hereditary or sporadic origins, because the process of carcinogenesis would be different in each case. Because it is known that MSI is related to hereditary non‐polyposis colorectal cancer, the role of MSI in colorectal cancer is relatively well established. The characteristics of MSI tumors in colorectal cancers have been identified. These tumors tend to be more proximal, poorly differentiated, and mucinous, with less lymph node metastasis and a better prognosis [3–5]. In gastric cancer, many studies have suggested that MSI tumors are associated with intestinal type Lauren classification and less lymph node metastasis [6,7]. However, the relationship between survival in gastric cancer patients and MSI has yet to be determined, because the results of analysis of the prognostic value of MSI among studies have been discordant. To obtain integrated and more precise data for the value of MSI as a prognostic marker in gastric cancer, we performed a comprehensive review of the literature and a meta‐analysis of studies to date.
Search Strategy and Selection Criteria
ß 2014 Wiley Periodicals, Inc.
This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) statement [8]. We searched the major electronic databases PubMed, EMBASE, and The Cochrane Central Register of Controlled Trials as of October 11, 2012, using search terms related to microsatellite instability, gastric cancer, and prognosis. Details of the PubMed search strategy are shown in the Appendix. Bibliographies of review articles and other related studies were searched manually to identify additional studies. The selection criteria followed the Population, Intervention, Comparator, Outcome, Setting, and Study Design (PICOTS‐SD) format, with gastric cancer patients who underwent gastrectomy as population, MSI status as intervention and comparator, and any kind of
Yoon Young Choi and Jung Min Bae contributed equally to this work. All authors have no conflict of interest and no financial disclosures to disclose. *Correspondence to: Sung Hoon Noh, MD, Department of Surgery, Yonsei University College of Medicine, 50 Yonsei‐Ro, Seodaemun‐gu, 120‐752 Seoul, Korea. Fax: þ82‐2‐313‐8289. E‐mail: [email protected] Received 16 January 2014; Accepted 17 March 2014 DOI 10.1002/jso.23618 Published online 15 April 2014 in Wiley Online Library (wileyonlinelibrary.com).
130
Choi et al.
survival as outcome. Because the major interest of this review was prognosis, that is, overall survival (OS), disease‐specific survival (DSS), or disease‐free survival (DFS), only studies reporting information on survival outcomes were included. Only studies published in peer‐ reviewed journals were included; review articles, abstracts, and editorial comments were not sought. There was no limitation on publication language or date, scale of the studies, or study design. When data was duplicated in multiple studies, the most reasonable study was included for a meta‐analysis. This comprehensive search and selection of studies was carried out by two authors (Y.Y.C. and J.M.B.) independently, and disagreements were resolved by discussion.
reviewing the titles and abstracts. Twenty‐four full text articles were excluded for the following reasons. One was a conference abstract [11], seven duplicated data [12–18], one was a review article [19], one was an editorial comment [20], three investigated only the response to chemotherapy without gastrectomy [21–23], three reported no survival results [24–26], one failed to extract survival outcome [27], and seven reported no information about MSI [28–34]. To obtain the HR for one study [27] that we were unable to extract from the article for meta‐analysis, we contacted the author by e‐mail but received no reply. Finally, 24 studies [6,7,35–56] with 5,438 participants (712 patients with MSI gastric cancer) from 14 countries were included for pooling risk estimates (Supplement Fig. S1).
Data Extraction and Assessment of Individual Outcomes Each included study was read by two authors (Y.Y.C. and J.M.B.) and the following data were extracted: publication year and country, enrollment period, total number of patients assessed in survival analysis, the range of gastric cancer stages, total number of markers and number of mononucleotides used to estimate MSI status, threshold to assign MSI, the proportion of MSI tumors, median or mean duration of follow‐up, control and intervention groups for estimation of hazard ratio (HR), and the type of survival outcomes. If the HR and 95% confidence interval (CI) were not available in a study, these were indirectly estimated from a Kaplan–Meier curve [9,10] using Plot Digitizer version 2.5.1 (http://plotdigitizer.sourceforge.net/). In addition, to analyze associations between the effect of MSI status and study characteristics, the proportions of lower and upper body gastric cancer, cases beyond stage III, female patients, intestinal type, cases beyond stage pN1, and mean age were extracted for each included study. When more detailed data were required for the analysis, we contacted the authors directly by e‐mail.
Statistical Analysis Analysis was conducted using Review Manager Ver. 5.1.7 (Nordic Cochrane Center, Copenhagen, Denmark) and STATA version 11.2 (Stata Corp, College Station, TX). For an appropriate analysis of time‐ to‐event outcomes, OS, DSS, or DFS, a generic inverse–variance method, was applied in this meta‐analysis. When the unadjusted or adjusted HR was available in a study, the log HR and its standard error (SE) were used directly with the generic inverse–variance method. Otherwise, log HR and SE were indirectly estimated using the method proposed by Parmar and coworkers [9,10]. HRs adjusted for other variables such as gastric cancer stage, age, and sex were assigned a higher priority for use in the meta‐analysis when provided. The heterogeneity of effect sizes across the studies was assessed by Cochran’s Q‐test and Higgin’s I2. A Cochran’s Q P‐value of less than 0.10 or an I2 value over 50% was considered substantial heterogeneity. When heterogeneity was suspected across the included studies, a random‐effects model was used to pool the data. Otherwise, a fixed‐ effects model was applied. The funnel plot summary approach was used to evaluate the presence of reporting bias. A sensitivity analysis was performed to examine whether the overall findings were robust to one or two outlying studies. For the sensitivity analysis, sequential meta‐ analyses were repeated, in which each study was excluded. Also, we performed meta‐regression to assess the associations between the effect of the MSI and study characteristics such as the proportion of lymph node positive population, stage III & IV gastric cancer, upper third gasric cancer, and so on in each study.
RESULTS Included Studies Of 392 records identified through database searching and removal of duplicates, 48 articles were assessed in full text for eligibility after Journal of Surgical Oncology
Characteristics of Studies Characteristics of the included studies are presented in Supplement Table SI. Eight of 24 studies were from Eastern countries [Korea (3) [6,54,56], Japan (3) [39,47,53], China (1) [55], and Taiwan (1) [45]], and the remainder were from Western countries [Italy (5) [7,42,43,50,52], USA (3) [40,46,49], Slovenia (1) [51], Brazil (1) [48], Poland (1) [44], Spain (1) [41], Australia (1) [38], Germany (1) [37], Portugal (1) [36], and the UK (1) [35]]. Only three studies [54– 56] enrolled patients during the 21st century, and all studies analyzed patients with stage I–IV gastric cancer, with the exception of two studies (one study [35] included stage I–III, and one study [45] did not provide information on stage). The stage of gastric cancer was determined based on the 4th through 6th editions of the American Joint of Committee on Cancer, and in some cases on Japanese classification of gastric carcinoma [57]. The populations assessed for survival ranged in size from 24 [48] to 1990 [56]. The number of markers used to estimate MSI status ranged from one to eleven, and a diversity of thresholds for assigning MSI‐H or MSI was applied among the studies. Thirteen studies [7,36,37,39–41,43,47,49,51,54–56] used a combination of mono‐ and dinucleotide markers, seven [6,38,42,44,45,50,52] used only mononucleotide markers, and three [35,48,53] used only dinucleotide markers. In two studies [38,45], only one mononucleotides marker, BAT26, was used for estimating MSI status, and one study [46] considered hMLH1‐negative cases MSI‐H. Only four studies [49,54–56] followed the National Cancer Institute (NCI) guidelines for the determination of MSI in colorectal cancer [58] (i.e., two mononucleotide markers [BAT25 and BAT26] and three dinucleotide markers [D5S346, D2S123, and D17S250]). The reported proportion of tumors with MSI ranged from 8.5% [56] to 37.8% [41] (median and range, 10.8% and 8.5%–19.2% in Eastern studies, and 17.6% and 10%–37.8% in Western studies). Median follow‐ up period was 18–92.1 months; 10 studies [35,38–42,45,47,48,51] did not report follow‐up duration for survival. The OS was available for 17 studies [7,36–42,44,45,47–49,51–53,56]; DSS was reported in 5 studies [6,35,43,46,50]; and DFS was estimated in 4 studies [48,54–56]. The characteristics of gastric cancer in each study are shown in Supplement Table SII. The proportion of lower body gastric cancer ranged from 27.0% [41] to 63.0% [45], and that of upper body cancer ranged from 7.3% [46] to 37.0% [51]. The proportion of stage III–IV cancer was relatively low in recent studies from Korea [54,56], and the proportion of female patients was between 30% and 40% in most studies. The proportion of patients who had metastases to lymph nodes was over 40% in all studies that reported that information, with one exception [49]. It was strongly suspected that, in this exception, the presence of metastasis had been inadvertently reported as the absence of metastasis; however, it was impossible to confirm this because we failed to receive a reply from the author. Six studies [7,39,42,43,46,50] reported an adjusted HR (the HR in five studies was adjusted based on gastric cancer stages), and the log HR was extracted from raw data in two studies (one study [47] reported the raw data in the article, and it was possible to obtain the raw data from the author in one study [56]). The
MSI and Gastric Cancer HR was estimated from a Kaplan–Meier curve in 14 studies [6,35– 37,40,41,44,45,48,49,51–54] and from observed and expected events in two studies [38,55]. Ten studies reported that the prognosis of the MSI group was not significantly different from the non‐MSI group, and 14 studies reported that the prognosis of the MSI group was better than that of others (Supplement Table SII).
The Pooled Risk of MSI The pooled HR of MSI gastric cancer for OS from 17 studies [7,36– 42,44,45,47–49,51–53,56] was 0.76 (95%CI: 0.65–0.88, P ¼ .0003) in a fixed‐effects model (I2 ¼ 34%, Cochran’s Q P ¼ .08; Fig. 1), and 0.72 (95%CI: 0.59–0.88, P ¼ .001) in a random‐effects model. The pooled HR of MSI gastric cancer for DSS from five studies [6,35,43,46,50] was 0.55 (95%CI: 0.43–0.71, P < .0001) in a fixed‐effects model (I2 ¼ 0%, Cochran’s Q P ¼ .45; Fig. 2A), and the pooled HR of MSI for DFS from four studies [48,54–56] was 1.03 (95%CI: 0.71–1.48, P ¼ .90) in a fixed‐ effects model (I2 ¼ 39%, Cochran’s Q P ¼ .18, Fig. 2B). The OS adjusted by stage was available in five studies [7,39,42,47,56] and the pooled HR of MSI was 0.59 (95%CI: 0.33–1.06, P ¼ .08) in a random‐ effects model (I2 ¼ 75%, Cochran’s Q P ¼ .003, Fig. 2C). Sensitivity analyses showed that the influence of each study on the pooled HR of MSI gastric cancer was similar for all but one study [56] (Fig. 3A); when An et al. (KR) was removed from meta‐analysis, the HR of MSI for OS improved (HR: 0.68, 95%CI: 0.57–0.80). With the 17 studies that reported the OS of MSI, meta‐regression was conducted to identify the relationship between baseline characteristics of each study and the log HR for MSI. The prognosis MSI was better when a study had higher proportions of lymph node‐ positive cases (coefficient: 0.013, P ¼ .011), stage III–IV cases (coefficient: 0.013, P ¼ .008), and female patients (coefficient: 0.047, P ¼ .043; Table I). The effect of the proportion of lymph node‐positive cases became more prominent (coefficient: 0.020,
P < .001) when the analysis was done assuming that the proportion of patients beyond stage pN1 in one study (An et al., US) [49] was not 17.3%, as reported, but 82.7%. Other characteristics, such as the proportions of MSI, upper body cancer, lower body cancer, and intestinal type, and mean age and year of publication did not show any significant relationship with the prognosis of MSI. The funnel plot for reporting bias showed symmetrical effect sizes (Fig. 3B), which suggesting no evidence of publication bias. However, one study [56] was out of the lines and the study was the heterogeneous study in a sensitivity analysis.
DISCUSSION The results of the present meta‐analysis showed that gastric cancer with MSI is associated with a better prognosis than gastric cancer without MSI (HR: 0.76, 95%CI: 0.65–0.88). This result is similar to that seen with colorectal cancer [5], and MSI was a significant prognostic factor in gastric cancer as well. However, sensitivity analysis revealed that the most recent study, with the largest number of patients, from Korea (An et al., KR) [56] reported the most heterogeneous result, the direction of prognosis was favor to without MSI gastric cancer (Fig. 3A,B). Possible reasons for this heterogeneity include (1) changed epidemiology of gastric cancer or (2) different effects of chemotherapy in gastric cancer with and without MSI. The most heterogeneous study (An et al., KR) [56] had a low proportion of stage III–IV cancer, upper body lesions, and lymph node metastasis. Another recent study [54] from Korea, which had similar characteristics to those of An et al. (KR), also reported an unfavorable DFS for MSI gastric cancer. Because mass screening for gastric cancer has recently become popular [59], the incidence of early gastric cancer is relatively high in Korea. And these two studies enrolled the patients after 21st century. The trend between the baseline characteristics of a study and the log HR of MSI from the meta‐regression (Table I) showed that
Fig. 1. Data and forest plot of the effects of microsatellite instability on overall survival. Journal of Surgical Oncology
131
132
Choi et al.
Fig. 2. Data and forest plot of the effects of microsatellite instability. A: Disease‐specific survival. B: Disease‐free survival. C: Overall survival adjusted by stage.
Fig. 3. Sensitivity analysis and funnel plot. A: Effect of each study on the pooled HR of overall survival. B: Funnel plot of overall survival for estimation of publication bias. Journal of Surgical Oncology
MSI and Gastric Cancer TABLE I. The Results of Meta‐Regression for Log Hazard Ratio of MSI in Each Study
Proportion of LN‐positivea Proportion of LN‐positiveb Proportion of MSI tumors Proportion of LB Proportion of UB Proportion of stage III–IV Proportion of females Proportion of intestinal type Mean age Publication year
Coefficient
P‐Value
0.013 0.020 0.022 0.006 0.020 0.013 0.047 0.003 0.037 0.024
0.011
