Background
Gastric cancer (GC) is the fifth most common malignancy and third leading cause of death from cancer worldwide [
1]. There is geographical variation in GC incidence, with Asia being the most common region, with lower rates in USA and Europe where, although incidence has been decreasing in recent decades, the 5-year survival remains poor (24% in Europe) [
2]. In Japan and Korea, high incidence and historically poor survival rates for gastric cancer have led to the introduction of surveillance programmes, which have increased 5-year survival to 60%, prompting suggestions to introduce such programmes in Western countries [
3]. However, the lower incidence of Helicobacter pylori (
H. pylori), the most common contributing factor to GC, [
4] in these lower risk areas, has impacted surveillance programme feasibility. In addition to
H. pylori, chronic inflammation of the stomach mucosa results in atrophic changes, including loss of structured glandular cells, which are replaced by intestinal-type epithelium, pyloric-type glands and fibrous tissue [
5]. The resultant gastric atrophy (GA) and intestinal metaplasia (IM) are known premalignant lesions for stomach cancer [
6,
7].
Studies have shown a significantly increased risk of GC in patients with either GA (5.8 times the risk of GC compared to patients without GA) [
8] or IM (10 times risk of GC compared to those individuals with no evidence of IM) [
9]. Several other studies have reported the association of these lesions with GC; however, the design and quality of these studies are varied, [
10‐
12] resulting in a wide range of observed cancer risk estimates.
To date, one systematic review has been published on the risk of GC in patients with GA, which used a serological method for diagnosis of these lesions [
13]. The aim of our systematic review was to determine risk of progression to GC in patients with histologically confirmed GA or IM and assess the quality of published studies.
Discussion
To our knowledge, this is the first systematic review to describe risk of patients with GA or IM developing GC, with these premalignant lesions diagnosed on histological examination. Our results show there is a wide variation in incidence rates in previous studies, including within continents. Also, due to poor study quality there is substantial heterogeneity.
There has been one previously published systematic review of the risk of GC in patients with GA, however diagnosis of GA was by serological methods [
13]. We only included GA and IM lesions diagnosed via histology in the current review as, although serological methods using pepsinogen ratios is a recognised technique for diagnosis, histological evaluation remains the most commonly used method. Laboratories use different pepsinogen ratio cut-offs to diagnose GA which, introduces inconsistency and potentially misdiagnosis of GA lesions when compared to histological analysis, so although serological diagnosis is a useful method, it is subject to limitations [
21].
Development of GA and IM commences a cascade of mucosal changes which can progress to GC, as described by Correa et al [
22]. In some Asian countries, patients who are diagnosed with GA or IM enter into a surveillance programme and prescribed
H. pylori eradication therapy, if positive on tissue biopsy [
23]. However, in most Western countries there are no surveillance programmes and patients are often not followed up. Recent European guidelines now recommend patients with GA or IM affecting both the antrum and the corpus should undergo endoscopic follow-up, but not in those with lesions limited to the antrum [
24]. As pepsinogen ratios cannot distinguish the extent of mucosal changes, histological diagnosis is required to determine patient qualification for entrance to this surveillance programme.
There was a large degree of heterogeneity between studies with GC, incidence rates ranging from 0.38 to 17.08 per 1000 person-years in individual studies, and high I
2 values when study estimates were pooled. Potential contributing factors include differences in study design, sample size and methods used to identify GC patients, which, when the Newcastle-Ottawa criteria are applied, result in a lower pooled GC incidence rate for studies of higher quality compared to analysis containing all studies, however high heterogeneity remains. Sensitivity analysis removing individual studies did not reduce this heterogeneity.
Causes for the high heterogeneity roots from study quality and consistency. Six of the 8 GA studies had incidence rates of between 1.0 and 5.2 per 1000 person-years, with the remaining two studies having more extreme results (0.53 and 15.24). There was a lack of detail regarding methods used to recruit patients for endoscopy, with only three studies reporting their techniques. Furthermore, there was a substantial variation in the amount of person-years follow up, ranging from 232 (which detected no gastric cancer cases) to 115,583. Of note, the studies producing the more extreme results had significantly less follow up than other studies (mean 33,076), which may have contributed to the wide range in incidence rates results. Takata et al. showed a much higher incidence rate than other studies (15.2 per 1000 person-years), a cause for which may include the outcome source [
25]. This article does not describe how they collected data for the study and thus their techniques may be inconsistent with the others included in this review.
Similar to the studies of GA, two of the nine studies in the IM cohorts had extreme incidence rates (0.38 and 17.08 per 1000 person-years), [
26,
27] whereas the remaining incidence rates were between 1.26 and 4.10 per 1000 person-years. The study by Kim et al. had a markedly different demographic than others whereby the average age of patients was 45 and there was a male predominance of 88%, in contrast to other studies where there was a more even distribution of age and gender [
27]. Horsley-Silva’s study had a shorter follow up than seven of the eight other IM studies (820 person-years), contrasting with the mean (16,774 person years) of these remaining IM publications, potentially contributing to the high incidence rate [
26].
Despite the significant increase in 5-year survival from GC in Japan, attributed to introduction of GC surveillance methods, a 2009 report indicated that introduction of such a programme was not feasible in the UK, [
28]. This is at variance with recently published European guidelines which advise follow-up endoscopy, for extensive lesions [
24]. Our study has shown that, compared with background risk of GC, patients with GA or IM in European countries are at greater risk of this tumour. These incidence rates are comparable with recent studies of oesophageal adenocarcinoma risk in patients with the premalignant lesion Barrett’s oesophagus, for whom there is currently an endoscopic surveillance programme [
29,
30].
Consistent with published literature, we found study location impacted the rate of transformation to GC [
31]. Patients with GA had a higher risk of GC in Asia, when compared to Europe, whereas, conversely, the IM group showed an increased incidence rate in European countries compared to the Asian continent. A lower rate of GC in patients with IM in Asia may reflect the introduction of surveillance programmes, however this was not found in patients with GA. As there was only one Asian study in the IM group, [
27] further studies are required to confirm this lower rate. Significant variation in demographics and methodology in this study may contribute to the findings, such as the lower mean age of patients (45 years old).
H. pylori is the most common risk factor for GC development, often resulting in atrophy of the gastric mucosa, commencing a cascade toward carcinoma [
32]. We were unable to perform a sensitivity analysis limited to studies that adjusted for
H. pylori infection as those that included this risk factor in analysis described its prevalence but not adjustment in GC incidence calculation. Therefore, due to the prominent role
H. pylori bacteria has in carcinogenesis, [
33] further studies should include consideration of adjustment for
H. pylori infection when determining GC incidence in patients with GA or IM.
A recognised limitation of systematic reviews includes the impact of demographic and within/between study methodology variation [
34]. Furthermore, bias can arise from non-publication of smaller studies with non-statistically significant results. However, the funnel plot demonstrates that there is little evidence of such publication bias in this review. In addition, comparing the incidence rates of patients with GA or IM to the general population can be affected by study timing, since worldwide incidence rates of GC are decreasing; studies from 1974 (Siurala et al.) [
35] may be less applicable compared to more recent publications. In addition, as only one publication reflecting two studies was population-based, [
36] incidence rates from individual tertiary referral centres may not be representative of the entire population. A further weakness occurred in five studies where the total number of person-years was not described, but was estimated from the median number of person-years. Furthermore, there were several studies that determined GC in a GA/IM cohort but did not report sufficient information for the incidence rate to be calculated. Therefore, we recommend that future studies should report the total number of person years and GC incidence to enable comprehensive meta-analyses to be conducted. In addition, there were multiple studies that did not exclude GCs detected in the first 6 months post-initial endoscopy. This is an important factor in cancer diagnosis as a GC may have been present in the first endoscopy but not detected. Thus excluding GC diagnosed within the first 6 months post-endoscopy resulted in the number of GC cases being lower than described. It is known the level of heterogeneity is associated with the predictive value of meta-analyses, [
37] and, thus due to the high heterogeneity in this study (>90%), when estimates are pooled an overall incidence rate is not presented. Significant heterogeneity affects reliability of combining results and thus we recommend future studies of this important subject is required in order to produce a reliable pooled incidence rate of cancer risk.