There are no established criteria to classify patients into high or low risk of progressing to gastric cancer (GC). The aim of the study was to identify predictors of GC occurrence among patients with gastric preneoplastic lesions. A prospective and retrospective follow-up study was carried out in a province in Spain with one of the highest risk of GC. The study included 478 patients who underwent gastric biopsy in 1988–1994 with diagnoses of normal mucosa, nonatrophic gastritis (NAG), non-metaplastic multifocal atrophic gastritis (MAG) and complete or incomplete intestinal metaplasia (IM) and who accepted to undergo a new biopsy during 2005–2007 or had an event during follow up. Inter- and intra-observer variability of histological diagnosis was assessed. Analysis was done using Cox proportional hazards risk (HR) models. The mean age of the patients was 50 years, 47% were males and the mean follow-up time was 12.8 years. During follow-up, 23 GC (4.8%) were diagnosed (21 adenocarcinomas and 2 lymphomas) with an incidence of 3.77 per 1,000 person per year. The incidence rate of GC for those with incomplete IM was 16.5 per 1,000 person years. Out the 21 adenocarcinomas, 16 had an incomplete IM in the baseline diagnosis. Incomplete IM (HR 11.3; 95% CI 3.8–33.9) and a family history of GC (HR 6.1; 95% CI 1.7–22.4) were the strongest risk factors for gastric adenocarcinoma. Subtyping of IM and family history of GC may be useful for the identification of high-risk patients who need more intensive surveillance.
Experimental and epidemiological evidence indicates that gastric cancer (GC) is the result of a long multistep and multifactorial process.1 An inflammatory process in the antrum, usually associated with Helicobacter pylori (Hp) infection, is considered to be the cause of the initial lesion, which may progress toward a multifocal chronic atrophic gastritis in the corpus, intestinal metaplasia (IM), dysplasia and finally the invasive carcinoma.2 This seems to be the sequence of the intestinal type of gastric adenocarcinoma, which is the predominant type of GC in many countries, and these lesions are considered as precursors of this type of cancer. IM is usually subclassified as “complete” (small intestinal type or Type I) or “incomplete” (colonic type or Types II and III), which is thought to be the most advanced stage of IM.3 The diffuse type of GC, on the contrary, lacks well-defined precursors, although many studies have observed that it is also associated with intestinal metaplasia.3
Symptoms are often absent or nonspecific in patients suffering this long multistep process, and, therefore, GC is usually diagnosed at an advanced stage when curative options are very limited.4, 5 Five-year relative survival rates of GC in European patients are very low (less than 23%)6; therefore, better control of risk factors and identification of high-risk patients at an early stage of disease represent the most effective ways for reducing the burden of this tumor.
The need to reevaluate premalignant gastric lesions is controversial. Clinical guidelines for surveillance and treatment of patients with preneoplastic lesions are lacking4 and there are no clear criteria to classify patients into high or low risk of progressing to GC. Several prospective studies7, 8 have observed that the risk of progressing to more advanced preneoplastic lesions or GC depends on the severity of the lesions diagnosed at the start of follow-up and if these lesions accumulate and progress with advancing age.
A recent nationwide prospective study in the Netherlands9 showed that increasing severity of premalignant lesions at initial diagnosis (mainly severe dysplasia), male gender and greater age are relevant risk factors for GC development. Several years before, a higher risk of GC was found in subjects with IM Type III (incomplete type),10 and they concluded that subtyping of IM could be useful for identifying individuals at high risk of GC, although this is still controversial. Other authors11 support that subtyping IM has no clinical value and that the pattern, extent and severity of atrophy is a more important predictor than IM subtype.
The aims of this study were to assess the progression of gastric preneoplastic lesions and to identify predictors of GC occurrence after a long-term follow-up in a region in Spain (province of Soria in Castilla Leon) with one of the highest incidence and mortality from GC (European adjusted mortality rates: 18.1 in males and 8.6 in females).12
Material and Methods
This is a prospective and retrospective follow-up study with 3 end points: GC, duodenal ulcer and Barrett's esophagus. In 2005, all patients who had undergone gastroscopy with gastric or esophageal biopsy (n = 2,229) at the Soria Hospital (the only National Health Service managed facility of a province of only 94,646 inhabitants in 2008) between January 1, 1988 and June 30, 1994, were identified from the Pathology Department Registry (Fig. 1). Preliminary study included all patients with preneoplastic gastric lesions, resident at the province of Soria, from both sexes. Of these, we excluded 1,427 patients with any of the following previous conditions: age outside the selected 25–69 year range (n = 610), prevalent peptic ulcer (n = 290), GC (n = 127) or prevalent Barrett's esophagus (n = 61), previous gastric resection (n = 52) and lack of information or other reasons (n = 287). The remaining 802 patients with a preliminary histological diagnosis of normal mucosa, chronic gastritis with or without IM or dysplasia in the first biopsy were selected for the study. All the patients' pathology reports were then reviewed to identify those patients who had already reached any of the pre-established end points and those were the candidates for a final gastroscopy. Ninety patients were considered to have already completed the selected follow-up outcomes. Among these, a gastric malignancy was diagnosed in 22 cases (20 adenocarcinomas, 2 lymphomas), a duodenal ulcer in 10 and Barrett's esophagus in 8 (Fig. 1). There were also 6 patients who had undergone a gastrectomy for a gastric ulcer and 44 who had died during the follow-up, for which the last previous gastroscopy and biopsy were taken into account. The death certificates were reviewed for those who had died to obtain the cause of death. Furthermore, we identified 49 patients that had undergone a gastroscopy with a gastric biopsy during the previous 2 years, and we took these biopsies as the end point, because it was considered unethical to perform a new one.
The remaining 663 patients were invited by letter and telephone call to undergo a new gastroscopy and gastric biopsy to provide a blood sample (30 ml), to answer a face-to-face questionnaire on history of tobacco consumption, use of non-steroidal anti-inflammatory drugs (NSAID), family history of GC (a case of GC in grandfathers, fathers, brothers or sons) and to assess prevalence of Hp infection. Two hundred and ninety declined and 16 were not suitable for a new gastroscopy. Information on these risk factors of interest in those patients with a predefined endpoint or event during follow-up who were not interviewed were obtained from medical records or questions to their relatives. All participating patients signed an informed consent giving permission for the procedures of the study, whose protocol was approved by the Ethical Committee of the Hospital.
According to the standard procedures of the institution at the time of the first gastric biopsy (between 1988 and 1994), 3–4 biopsy specimens were collected from the antrum, incisura or corpus for histological evaluation in all patients. For the final gastroscopy, the Sidney recommendations were followed and 5 specimens were obtained (1 from the angulus, 2 from the antrum and 2 from the corpus of the stomach). All biopsies (baseline and final), were formalin fixed and paraffin embedded. Sections of 4 μm were obtained and stained with hematoxylin–eosin, Alcian blue–periodic acid Shiff (pH 2.5) and Giemsa. Additionally, for selected baseline biopsies with incomplete IM, some sections were stained with high-iron diamine–Alcian blue to detect sulfated mucins. Two pathologists (MLP and JMSA) reviewed the initial and final biopsies of all the included cases.
Giemsa stained Hp was detected during the histopathological examination. Furthermore, Hp infection was also identified by reviewing the medical records (results from urease test on biopsy specimen and urea breath test) and interviewing subjects about previous prevalence of Hp, treatment and results of treatment. The presence of a positive Hp infection among any of these sources of information was taken as the overall diagnosis for each subject. Information of use of proton pump inhibitors in the treatment was not available and therefore not taken into account in the analysis.
Histological diagnosis classification
The most advanced lesion observed in any of the biopsy fragments collected at the beginning and at the end of the study was used to classify every patient. For the 139 patients who completed follow-up prior to clinical reevaluation (Fig. 1), the diagnosis from either the end point biopsy (GC, duodenal ulcer, Barrett's esophagus) or the last biopsy available prior to gastrectomy or death was taken into account. The classification of the diagnosis was the following: 1: normal mucosa/minimal change; 2: nonatrophic gastritis (NAG); 3: non-metaplastic multifocal atrophic gastritis (MAG); 4: multifocal atrophic gastritis with intestinal metaplasia (IM); 5: dysplasia; 6: carcinoma. On the basis of morphology and mucin staining, IM was subclassified10 as “complete” (small intestinal type or Type I, characterized by brush border cells and goblet cells, but lacking the hybrid columnar “intermediate” cells, which are typical of the incomplete IM or “incomplete” (colonic type or Types II and III) that also contain goblet cells, but the predominant cell type is the hybrid “intermediate” nongoblet mucous columnar cells with cytoplasmic drops of different size. This hybrid mucous cells are easily identified because they show, with the mucinohistochemical AB-PAS (pH 2.5), a mixed phenotype gastric and intestinal pattern: red for neutral gastric mucins and blue for a combination of intestinal mucins. Mixed, predominant complete types or incomplete types were also considered.
To assess the observer variability of the histological diagnosis, a random sample of 10% of the pathological samples were included in a blinded study, where they were examined twice by the same pathologist (intra-observer variability) and by another pathologist not involved in the study (inter-observer variability).
Cox proportional hazards models were used to estimate hazards ratios (HRs) with 95% confidence intervals (CI). The current analysis is based only on GC end points, because of the low observed number of cases of Duodenal Ulcer (n = 10) and Barrett's esophagus (n = 8). Entry time was defined as age at recruitment and final time as the age at diagnosis of GC, the age at censoring for at-risk subjects or the age at end of follow-up. The risk of GC was measured in a univariate and multivariate analysis, according to baseline histological diagnosis, age, sex, tobacco consumption and use of NSAID, family history of GC and prevalence of Hp infection.
The analysis was based on 478 subjects, after the exclusion of 10 patients diagnosed with duodenal ulcer and 8 diagnosed with Barrett's esophagus during the follow-up. The mean age of the patients was 50.0 years (SD = 12.19) and 47% were males (Table 1). The histological diagnosis at baseline examination was 1.7% (n = 8) normal, 11.5% (n = 55) with NAG, 46% (n = 223) with MAG and 40% (n = 192) with IM. No dysplasia cases were observed. Among the IM patients, 54.2% (n = 104) were diagnosed as complete or predominantly complete and 45.8% (n = 88) as incomplete or predominantly incomplete. Regarding intra-observer concordance, the weighted kappa statistic was 0.58 for diagnosis of subtype of metaplasia and 0.43 for the global histological diagnosis. For the inter-observer concordance, the ponderate kappa statistic was 0.55 for diagnosis of subtype of metaplasia and 0.41 for the global histopathological diagnosis.
Table 1. Histopathology diagnoses at baseline and risk factors
Table 1 also presents the histopathological diagnoses at baseline in relation to age, gender and other risk factors. A greater proportion of males and older patients had more advanced lesions. Up to 76.4% of NAG cases were Hp positive with a steady decrease in more advanced lesions. Prevalence of ever smokers was 38.7% and no clear differences were observed between histopathological groups. In contrast, use of NSAID varied according to the histological diagnosis. Prevalence of family history of GC was particularly high (23.9%) in subjects with incomplete IM.
After a mean follow-up of 12.8 years (SD = 4.07), 23 GC cases (4.8%) were diagnosed (Table 2). The incidence rate was 3.77 (CI 95% 2.51–5.67) per 1,000 person/years. Of the GC cases, 21 were adenocarcinomas (2 diffuse, 16 intestinal and 3 mixed) and 2 were malignant lymphomas. Of the 21 adenocarcinomas, 16 had an incomplete IM at baseline diagnosis (including the 2 diffuse types), 1 had a complete IM and 4 had a MAG. One gastric lymphoma had a MAG at baseline diagnosis and the other had a NAG. The proportion of subjects with incomplete IM that developed a GC during the follow-up was 18.2%, while it was only 2.2% for MAG. The incidence rate of GC was 16.46 (95% CI 10.08–26.86) for incomplete IM and 1.73 (95% CI 0.72–4.16) per 1,000 person/years for MAG. To estimate the extension of IM, the number of fragments with IM and the total number of fragments was recorded for each biopsy. The percentage of fragments with IM was higher among the cases that developed a GC in comparison with those that did not develop a GC, but the differences were not statistically significant (data not shown).
Table 2. Number and proportion (as percentage between brackets) of patients according to diagnoses at baseline and at the end of follow-up
Incomplete IM was the baseline diagnosis with the highest risk of developing a gastric adenocarcinoma (HR 11.3; 95% CI 3.8–33.9) (Table 3). We also found a strong and significant risk of developing a gastric adenocarcinoma for those subjects with a family history of GC (HR 6.11; 95% CI 1.67–22.4). While age and gender were significant risk factors in the univariate analysis (HR 4.18; 95% CI 1.12–15.61 for patients older than 60 years and HR 5.51; 95% CI 1.85–16.4 for males), they became weaker and nonsignificant after adjusting for the other risk factors. We observed a nonsignificant negative association between regular use of NSAID for more than 5 years and risk of GC (HR 0.43; 95% CI 0.12–1.49). Prevalence of Hp infection at baseline and smoking were not associated with the risk of GC (data not shown).
Table 3. Hazard ration1 of gastric cancer of preneoplasic lesion after a mean of 12.8 years of follow-up, according to different factors (based on 21 adenocarcinoma cases and 478 subjects with prenoplastic lesions)
Multivarite analysis adjusted by each of the other factors.
In a province located in a region with one of the highest incidence of and mortality from GC in Spain and Western Europe, we carried out a relatively large observational follow-up study with a long follow-up. Gastroscopies at the beginning and at the end of follow-up were performed allowing us to evaluate the occurrence of GC among patients with preneoplastic lesions and the influence of Hp infection and other factors in this process. We found that the probability of developing a GC is strongly and significantly associated with incomplete IM and a family history of GC.
During 12.8 years of follow-up, 23 patients (4.8%) developed GC, this translates to an average of 0.37% per year. This is similar to the results obtained in a prospective study in Japan8 where 2.9% of patients (36 incident cases) developed a GC after 7.8 years of follow-up, which is an average of 0.37% per year. A slightly lower proportion was obtained in another prospective study in a high risk area of China7 where 1% (34 incident cases) developed a GC after a mean of 4.5 years of follow-up, which means an average of 0.22% per year. It is well known that the age-standardized incidence rate of GC is two-fold higher in men than women13 and other prospective studies7, 8 have also found a higher GC risk among males and older subjects. Moreover, it has been suggested that sex hormones may protect women against GC14 and results from recent epidemiological studies support this hypothesis.15, 16 While in our population a higher risk of developing GC was observed for individuals older than 60 years and males in the univariate analyses, it disappeared after adjusting for other variables.
Incomplete IM was the only baseline diagnosis associated with a high and significant risk of developing GC. This supports that subtyping of IM is useful to identify subjects at high risk for GC. In addition, it has previously been shown that close endoscopic surveillance of patients with type III IM (columnar cells with sulfated mucines) was associated with an increased frequency of early GC diagnosis.17 Another prospective study10 also observed a higher risk of developing GC for Types II and III IM, compared with Type I. A short follow-up study of patients with atrophic chronic gastritis or IM18 has shown that patients with Types II or III IM were at highest risk for developing a dysplasia and only patients with Type III IM progress to high grade dysplasia. A study in Italy found that type and extension of IM, and immature hyperproliferative lesions, labeled as “indefinite for dysplasia” according to the Padova classification, were the most sensitive screening parameters for identification of subjects with increased GC risk. However, evidence is not consistent and in other studies20–22 IM subtyping was not found to play a major role in predicting GC. These studies20, 21 were cross-sectional observations with no follow-up or limited to a small number of patients.22 Whether IM is reversible or not is still a matter of controversy.23 A follow-up study of different subtypes of IM24 concluded that Type I IM is a short-term reactive process that generally regresses while Type III IM is related to prolonged injury and chronicity and may regress or progress to dysplasia. In our study, regression to less advanced lesions was found in 29 (28%) of the 104 cases with complete or predominant complete IM and only 12 (14%) of the 88 cases with incomplete or predominant incomplete IM.
The pattern, extent and severity of gastric atrophy, with or without IM, have been postulated as the best prognostic markers of GC with pre-malignant lesions.11 However, the predictive value of NAG and MAG in our population was very low since only one of the 23 incident GC arose from a NAG and 5 from a MAG. It is noteworthy that the classification of atrophic gastritis suffers from considerable inter-observer variation and that recognition and grading of gastric atrophy remains inadequate.4 On the other hand, subtyping of IM, as described by Jass and Filipe,25 showed a very good predictive value: 16 out of 23 incident GC developing from an incomplete IM. Contrary to gastric atrophy, inter-observer agreement in the classification of these subtypes of IM is generally considered satisfactory.4 In line with these observations, inter- and intra-observer concordance in our study was also good for the diagnosis of type of metaplasia.
We found an almost 5 times higher GC risk in patients with a family history of GC compared to patients without family history. This is higher than the risk observed in another prospective study in Japan.26 It is well known that individuals reporting a family history of GC experience an increased risk of GC.27 Our result could be an indicator of the role of genetic susceptibility on the risk of GC,28 which could explain, in part, the high risk in this population, but it could also be due to the effect of sharing environmental factors between family members.29
Hp infection is a well-recognized risk factor for precancerous lesions of the stomach.30 In our study, we found that it is not a predictor of GC. It is well known that severe chronic gastritis and extended IM in stomach mucosa create conditions in which Hp is unable to survive and favor clearance of the infection. In agreement with this, we observed that the prevalence of the infection decreased as the lesion advanced and therefore the HR was lower in the more advanced lesions. This fact, together with the low number of normal mucusa analyzed in this study, could explain our results. Similarly, several epidemiological studies31–34 have shown that smoking is a strong risk factor of precancerous stomach lesions and is associated with their progression toward GC. We observed a relatively high prevalence of smokers in subjects with all types of lesions, but the group of subjects with normal mucosa was too small to enable comparison to be made. We also observed a nonsignificant negative association between consumption of NSAID and the incidence of GC. A meta-analysis35 has shown that users of aspirin and non-aspirin NSAID experienced a lower risk of GC than non-users, which is in agreement with the role of NSAID as inhibitors of COX-2, a stimulator of cell proliferation and angiogenesis.36
Strengths of this study are the long follow-up and the relatively large size. Although the sample is not based in a random sample from the general population, the cases were selected from the single hospital facility in the small province of Soria. This meant that the study included all the cases occurring in this population, who probably visited the Hospital due to gastric related symptoms, and who had a previous gastric biopsy between January 1, 1988 and June 30, 1994. In addition, screening and surveillance of the general population is not cost-effective except in countries with a very high risk of GC like Japan. The main aim of the study was to identify markers of high risk patients among those who have had an endoscopic examination. The same pathologists classified the patients at baseline and at the end of follow-up, and the level of agreement in the study regarding intra- and inter-observer variability was quite good. There are also some limitations: the endoscopic techniques were more standardized at the end of follow-up than at baseline, although the gastroenterologist that took the biopsies at the end of the study was one of those who performed gastric biopsies at baseline. The number of biopsies was higher at the end, when the Sidney recommendations were followed. As the accuracy of the diagnosis depends on the number of biopsy sites37 and the chance of detecting advanced lesions is higher if the number of biopsies increases, some level of underestimation of advance lesion at baseline is possible in our study. The participation rate of those patients selected for a second biopsy was relatively low, although a 56% of participation for a gastroscopy and biopsy for research purposes is good, especially for a Western country. We checked for potential bias associated with the participation rate. We compared the basal diagnosis, age and sex of those who did not accept to participate with those who accepted, and we observed no differences regarding sex and basal diagnosis and only minor differences regarding age, as those who did not accept were slightly older. Finally, although serum pepsinogen status has been proposed as an effective measure to predict the development of GC,38 in this study we could not assess its value because blood samples from patients at recruitment were not available.
To conclude, our study indicates that subtyping of IM and family history of GC may be useful parameters to identify high-risk patients. A study on patients with gastric lesions requiring follow-up,39 that compared survival of those who accepted annual surveillance with those who declined, has shown an impressive improvement in those who accepted (50% at 5 years vs. 10%). To identify high-risk patients who should be included in a program of more intensive surveillance, further research is needed to confirm whether subtyping of IM is useful marker as well as about the role of other prognostic markers.
The authors thank the participating patients and their families and to Ms. Pilar Lopez and Ms. Amalia Solana for their valuable collaboration in the field work, as well as Dr. J.M. Arrinda Yeregui (Pathological Department, Hospital de Bidasoa, Guipuzcoa, Spain) for his collaboration in the histophatological concordance.