Abstract
Gastric cancer lacks specific symptoms, resulting in diagnosis at later stages and high mortality. Serum pepsinogen is a biomarker for atrophic gastritis, a gastric cancer precursor, and may be useful to detect persons at increased risk of gastric cancer.
The Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial was conducted in the United States between 1993 and 2001. ELISA-based pepsinogen tests were conducted on prediagnostic serum samples of 105 PLCO participants who developed gastric cancer and 209 age, sex, and race-matched controls. Pepsinogen positive (PG+) was defined as pepsinogen I ≤ 70 μg/L and pepsinogen I/II ratio ≤3.0. Results of conditional logistic regression models, and sensitivity and specificity, of PG+ for gastric cancer are reported.
Gastric cancer cases were more likely to be PG+ (31.4% vs. 5.5%, P < 0.001) at baseline than controls. Compared to PG-, PG+ was associated with an 8.5-fold increased risk for gastric cancer [95% confidence interval (CI) = 3.8–19.4]. This risk remained significant after adjusting for Helicobacter pylori, family history of gastric cancer, education, smoking, and BMI (aOR, 10.6; 95% CI, 4.3–26.2). In subgroup analysis, PG+ individuals were 11-fold more like to develop non-cardia gastric cancer (OR, 11.1; 95% CI, 4.3–28.8); conversely, they were not significantly more likely to develop cardia gastric cancer (OR, 2.0; 95% CI = 0.3–14.2). PG+ status yielded low sensitivity but high specificity for both noncardia (44.3%; 93.6%) and cardia gastric cancer (5.7%; 97.2%).
Prediagnostic serum pepsinogen levels from a large, prospective cohort study were associated with risk of gastric cancer, particularly noncardia gastric cancer.
PG status may identify individuals at higher risk of noncardia gastric cancer for targeted screening or interventions.
Introduction
Gastric cancer mortality is unduly high in the United States; only 28% of gastric cancers are diagnosed at earlier, localized stages of cancer, resulting in an overall 5-year survival of 32% (1). When gastric cancer is diagnosed at an early stage, 5-year survival can exceed 90% (2, 3). Gastric cancer lacks specific symptoms, and, without screening, is generally discovered in later stages of cancer. Thus, gastric cancer mortality could be improved if biomarkers to detect individuals at higher risk of gastric cancer could be used to guide gastric cancer screening and prevention strategies.
Gastric carcinogenesis follows a multistep histopathologic pathway known as the Correa cascade, which involves the following steps: chronic active gastritis, atrophic gastritis, intestinal metaplasia, dysplasia, and, ultimately, cancer (4, 5). Serum pepsinogen (PG) is a plausible serum biomarker candidate. It is a pro-enzyme of the digestive enzyme pepsin, and is mainly produced by the chief cells of the fundic glands of the stomach (6). PG reflects the functional and morphological status of the gastric mucosa and serves as a marker of atrophic gastritis. Low serum PG I and PGI/II ratio have been associated with severe atrophic gastritis and gastric cancer due to loss of cells in the corpus and fundus (7–9). Given that PG identifies atrophic gastritis, a precursor of gastric cancer, PG has great potential to be used as a tool for identifying individuals at increased gastric cancer risk. PG has been examined extensively in Asia and Europe and found to be predictive of gastric cancer (10–16); however, very few studies are available of U.S. populations, and are mainly limited to select high-risk groups, including one study of Alaskans (17) and another limited to Hawaiian men of Japanese descent (18). There have been no studies that have examined PG as a predictor of gastric cancer in a broader U.S. population.
In this study, we conducted a nested case–control analysis of patients with gastric cancer and 2:1 matched controls from the Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (PLCO) cohort to examine the association between serum PG and subsequent gastric cancer development. Our aim was to examine PG as a biomarker of gastric cancer risk in the U.S. population.
Materials and Methods
Study population
PLCO was a large randomized trial that was conducted over 10 centers in the United States between 1993 and 2001. The study was designed to determine the effectiveness of prostate, lung, colorectal and ovarian cancer screening on cancer-related mortality (19). Subjects were enrolled at baseline and followed until the development of cancer or death. Approximately 155,000 men and women aged 55 to 74 years were enrolled in the study. Cancer diagnoses were ascertained though annual study update forms and subsequent medical record verification. Information on subjects’ demographic, lifestyle, and medical history were collected through a survey at the time of enrollment. Serum samples were collected at time of enrollment and during follow up. Only individuals with a complete baseline questionnaire and serum sample available from the enrollment visit were considered for matching.
Analyses reported here are based on 2:1 matched nested case–control sample. Cases were defined as those who were diagnosed with gastric cancer (ICD-O-2 codes C160-C169) after enrollment in the trial and controls were PLCO participants who had not developed gastric cancer by follow up date of September 26, 2008. Controls were 2:1 frequency matched to the cases by age (≤59, 60–64, 65–69, ≥70 years), sex, and race/ethnicity (White, Black, Hispanic, Asian). Serum samples of cases and matched controlled were used to conduct PG tests and Helicobacter pylori (H. pylori) IgG tests using commercial ELISA kits (Biohit Healthcare). The assay kit used for testing is based on sandwich enzyme immunoassay technique with a pepsinogen I/pepsinogen II/H. pylori antibody–specific capture antibody adsorbed on a microplate and a detection antibody labeled with horseradish peroxidase (HRP). Testing was conducted in 2013. The testing was performed by National Cancer Institute investigators under the study titled “Gastric Cancer Cohort Consortium Study,” which aimed to develop a large, nested case–control study by pooling prospectively collected specimens from multiple cohort studies. Data generated for the above study, including the test results of PG and H. pylori, were provided by PLCO by special request for the conduct of this research.
The original data provided by PLCO included 109 gastric cancer and 217 controls. Upon further examination we found 4 cases that did not have confirmed gastric cancer-3 subjects were found to be “erroneous reports of cancer” while the 4th case was an esophageal cancer that had been misclassified as gastric cancer. Removal of those cases as well as their matched controls resulted in 105 cases and 209 controls. Among the 105 cases and 209 controls there was one match with one case and a single corresponding control, and three matches which had 2 cases and 4 corresponding controls, due to identical matching characteristics. This explains why our sample size did not result in an exact 1:2 ratio of cases to controls. For subset analyses by site, subgroups of noncardia and cardia cancers were examined separately with their respective controls. One of the matches with 2 cases contained a noncardia and a cardia case, hence this match was included as a single case with 4 controls each for both the noncardia and cardia groups. This resulted in 70 cases and 141 matched controls for the noncardia gastric cancer subset and 35 cases and 72 matched controls for cardia gastric cancer subset.
Serum pepsinogen I levels ≤ 70 μg/L and pepsinogen I to II ratio ≤ 3.0 were classified as PG-seropositive (PG+) and all others were considered PG-seronegative (PG−). These cutoffs are widely accepted and are the most commonly used (20). H. pylori IgG ≥ 10 EIU/ml was considered H. pylori positive (21, 22).
Statistical analysis
Descriptive statistics were used to summarize baseline characteristics of the study sample. Conditional logistic regression of the matched sets was used to estimate the OR and associated 95% confidence intervals (95% CI) for gastric cancer comparing PG+ to PG− patients. Adjusted models included variables with univariate associations of P < 0.2 in addition to variables that were a priori deemed relevant such as family history of gastric cancer and education. The utility of the pepsinogen biomarker was based on its estimated sensitivity and specificity to identify gastric cancer cases versus controls. Confidence intervals for sensitivity and specificity were calculated on the basis of the normal approximation to the binomial distribution. Stability of PG as a biomarker of gastric cancer by time was examined by comparing the OR of PG in patients where the time interval between blood draw and cancer diagnosis was greater than the median follow-up time with those were less. Stability of PG irrespective of time from cancer diagnosis would indicate that PG is more a reliable biomarker for risk of development of gastric cancer, but may be less useful as a marker of the presence of gastric cancer. Finally, given that the addition of H. pylori antibody information has been shown to further stratify risk above using PG alone in the literature, we also examined the risk of gastric cancer of three groups; Group A (HP-/PG-), Group B (HP+/PG-) and Group C (PG+). We pooled Group C (HP+/PG+) and Group D (HP-/PG+) together as Group C (PG+) as there were only 5 participants in group D.
All analyses were conducted using SAS 9.4 (SAS Institute, Inc.).
Results
Participant characteristics at the time of enrollment are provided in Table 1. Median and range of time from the baseline pepsinogen measurement to incident gastric cancer diagnosis was 6.7 years and 0.5 to 12.8 years, respectively. For controls, the median and range of duration of follow-up was 13.1 years and 2.3 to 16.1 years, respectively, up to follow up date of September 26, 2008. Gastric cancer participants were more likely to be PG+ (31.4% vs. 5.5%, P < 0.001) and current smokers (20.0% vs. 7.7%, P = 0.009) at baseline than controls. No difference was observed for H. pylori IgG positivity between gastric cancer cases and controls (70% vs. 67%, P = 0.59).
Participant characteristics.
Variable . | Categories . | Cases, N=105 n (%) . | Controls, N=209 n (%) . | P . |
---|---|---|---|---|
Age | 55–59 | 16 (15) | 31 (15) | 0.99 |
60–64 | 34 (32) | 69 (33) | ||
65–69 | 30 (29) | 62 (30) | ||
70–74 | 25 (24) | 47 (22) | ||
Sex | Male | 85 (81) | 169 (81) | 0.98 |
Female | 20 (19) | 40 (19) | ||
Race | White | 84 (80) | 167 (80) | 1.00 |
Black | 7 (7) | 14 (7) | ||
Hispanic | 5 (5) | 10 (5) | ||
Asian | 9 (9) | 18 (9) | ||
Family History of gastric cancer | Yes | 8 (8) | 10 (5) | 0.31 |
No/Possibly | 96 (93) | 197 (95) | ||
Smoking | Never | 33 (31) | 81 (39) | 0.009 |
Former smoker (>6m) | 51 (49) | 111 (53) | ||
Current smoker | 21 (20) | 17 (8) | ||
Education | Less than HS | 16 (15) | 27 (13) | 0.31 |
HS | 23 (22) | 46 (22) | ||
Less than college | 42 (40) | 68 (33) | ||
College or more | 24 (23) | 68 (33) | ||
BMI ≥ 30 | Yes | 25 (24) | 33 (16) | 0.09 |
(missing =3) | No | 80 (76) | 174 (84) | |
HP IgG | Yes | 73 (70) | 139 (67) | 0.59 |
No | 32 (30) | 70 (33) | ||
Pepsinogen positivitya | PG+ | 33 (31) | 11 (5) | <0.0001 |
PG- | 72 (69) | 198 (95) | ||
Location, cases only | Cardia | 35 (33) | ||
Fundus, Body, Antrum | 42 (40) | |||
Stomach, NOS | 28 (27) | |||
Histology (ICD-O-3 codes), cases only | Carcinoma NOS (8010) | 5 (5) | ||
Adenocarcinoma NOS (8140) | 75 (71) | |||
Adenocarcinoma, Intestinal type (8144) | 4 (4) | |||
Mucinous Adenocarcinoma (8480) | 3 (3) | |||
Signet ring cell carcinoma (8490) | 18 (17) |
Variable . | Categories . | Cases, N=105 n (%) . | Controls, N=209 n (%) . | P . |
---|---|---|---|---|
Age | 55–59 | 16 (15) | 31 (15) | 0.99 |
60–64 | 34 (32) | 69 (33) | ||
65–69 | 30 (29) | 62 (30) | ||
70–74 | 25 (24) | 47 (22) | ||
Sex | Male | 85 (81) | 169 (81) | 0.98 |
Female | 20 (19) | 40 (19) | ||
Race | White | 84 (80) | 167 (80) | 1.00 |
Black | 7 (7) | 14 (7) | ||
Hispanic | 5 (5) | 10 (5) | ||
Asian | 9 (9) | 18 (9) | ||
Family History of gastric cancer | Yes | 8 (8) | 10 (5) | 0.31 |
No/Possibly | 96 (93) | 197 (95) | ||
Smoking | Never | 33 (31) | 81 (39) | 0.009 |
Former smoker (>6m) | 51 (49) | 111 (53) | ||
Current smoker | 21 (20) | 17 (8) | ||
Education | Less than HS | 16 (15) | 27 (13) | 0.31 |
HS | 23 (22) | 46 (22) | ||
Less than college | 42 (40) | 68 (33) | ||
College or more | 24 (23) | 68 (33) | ||
BMI ≥ 30 | Yes | 25 (24) | 33 (16) | 0.09 |
(missing =3) | No | 80 (76) | 174 (84) | |
HP IgG | Yes | 73 (70) | 139 (67) | 0.59 |
No | 32 (30) | 70 (33) | ||
Pepsinogen positivitya | PG+ | 33 (31) | 11 (5) | <0.0001 |
PG- | 72 (69) | 198 (95) | ||
Location, cases only | Cardia | 35 (33) | ||
Fundus, Body, Antrum | 42 (40) | |||
Stomach, NOS | 28 (27) | |||
Histology (ICD-O-3 codes), cases only | Carcinoma NOS (8010) | 5 (5) | ||
Adenocarcinoma NOS (8140) | 75 (71) | |||
Adenocarcinoma, Intestinal type (8144) | 4 (4) | |||
Mucinous Adenocarcinoma (8480) | 3 (3) | |||
Signet ring cell carcinoma (8490) | 18 (17) |
Abbreviations: BMI, body mass index; HS, high school; HP, H. pylori; PGI, pepsinogen-I; PGR, Pepsinogen-I to Pepsinogen–II ratio
aPepsinogen positivity defined as PGI ≤ 70 μg/L and PGR ≤ 3.
In a univariable conditional logistic regression model, PG+ was associated with an increased risk for gastric cancer compared with PG− (OR, 8.5; 95% CI = 3.8–19.4). Risk of gastric cancer for PG+ remained significant after adjusting for H. pylori, family history of gastric cancer, education, smoking, and BMI (aOR, 10.6; 95% CI = 4.3–26.2). A complete summary of these analyses is presented in Table 2.
Univariable and multivariable conditional logistic regression analyses of gastric cancer risk.
. | . | Univariable model . | Multivariable modela . |
---|---|---|---|
Variables . | Reference . | OR (95% CI) . | aOR (95% CI) . |
PG+ | PG- | 8.5 (3.8–19.4) | 10.6 (4.3–26.2) |
HP+ | HP- | 0.8 (0.4–1.5) | |
Confirmed Family history of gastric cancer | none | 2.3 (0.8–7.0) | |
Less than HS | HS or more | 1.2 (0.5–3.0) | |
Former smoker | Never smoker | 1.5 (0.8–3.0) | |
Current smoker | Never smoker | 4.1 (1.7–9.9) | |
BMI>=30 | BMI<30 | 2.0 (0.9–4.5) |
. | . | Univariable model . | Multivariable modela . |
---|---|---|---|
Variables . | Reference . | OR (95% CI) . | aOR (95% CI) . |
PG+ | PG- | 8.5 (3.8–19.4) | 10.6 (4.3–26.2) |
HP+ | HP- | 0.8 (0.4–1.5) | |
Confirmed Family history of gastric cancer | none | 2.3 (0.8–7.0) | |
Less than HS | HS or more | 1.2 (0.5–3.0) | |
Former smoker | Never smoker | 1.5 (0.8–3.0) | |
Current smoker | Never smoker | 4.1 (1.7–9.9) | |
BMI>=30 | BMI<30 | 2.0 (0.9–4.5) |
aModel includes – HP, Family hx of gastric cancer, education, smoking, BMI.
Subgroup analyses were performed separately for noncardia gastric cancer (70 cases and 141 matched controls) and cardia gastric cancer (35 cases and 72 matched controls). This was done due to known heterogeneity in the etiology of these two gastric cancer subtypes. We were unable to perform separate analyses by diffuse- versus intestinal-type adenocarcinoma due to small numbers in these categories (Supplementary Table S1). We observed a substantially larger risk of gastric cancer for PG+ individuals for noncardia gastric cancer (OR, 11.1; 95% CI, 4.3–28.8) than for cardia gastric cancer (OR, 2.0; 95% CI, 0.3–14.2; Table 3). Risk of noncardia gastric cancer for PG+ remained significant after adjusting for H. pylori, family history of gastric cancer, education, smoking, and BMI (aOR, 14.3; 95% CI, 4.8–42.0). For these analyses, the definition of noncardia included all tumor locations that were not cardia, including stomach not otherwise specified (NOS). When stomach NOS was removed and magnitude of risk for non-cardia gastric cancer was limited to tumors documented as being in fundus, body, antrum, or pre-pylorus locations (42 noncardia gastric cancer, and 85 matched controls), associations of PG+ and noncardia gastric cancer were nonsignificantly greater (OR, 12.9; 95% CI, 3.8–43.6 and aOR, 17.4; 95% CI, 3.5–85.9; Supplementary Table S2).
Univariable and multivariable conditional logistic regression analysis of gastric cancer risk, by subtype.
. | . | Noncardia gastric cancer (70 noncardia, 141 matched controls) . | Cardia gastric cancer (35 cardia, 72 matched controls) . | ||
---|---|---|---|---|---|
. | . | Univariable . | Multivariablea . | Univariable . | Multivariablea . |
Variables . | Reference . | OR (95% CI) . | aOR (95% CI) . | OR (95% CI) . | aOR (95% CI) . |
PG+ | PG- | 11.1 (4.3–28.8) | 14.3 (4.8–42.0) | 2.0 (0.3-14.2) | 5.3 (0.6–48.7) |
HP+ | HP- | 1.1 (0.4–2.8) | 0.5 (0.1–1.3) | ||
Family history of gastric cancer | none | 3.6 (1.0–12.6) | 0.6 (0.1–7.7) | ||
Less than HS | HS or more | 1.5 (0.4–5.1) | 0.99 (0.2–4.2) | ||
Former smoker | Never smoker | 1.9 (0.8–4.9) | 1.2 (0.4–3.3) | ||
Current smoker | Never smoker | 4.5 (1.4–14.8) | 3.0 (0.7–13.7) | ||
BMI ≥ 30 | BMI<30 | 1.23 (0.4–3.8) | 3.95 (1.1–14.3) |
. | . | Noncardia gastric cancer (70 noncardia, 141 matched controls) . | Cardia gastric cancer (35 cardia, 72 matched controls) . | ||
---|---|---|---|---|---|
. | . | Univariable . | Multivariablea . | Univariable . | Multivariablea . |
Variables . | Reference . | OR (95% CI) . | aOR (95% CI) . | OR (95% CI) . | aOR (95% CI) . |
PG+ | PG- | 11.1 (4.3–28.8) | 14.3 (4.8–42.0) | 2.0 (0.3-14.2) | 5.3 (0.6–48.7) |
HP+ | HP- | 1.1 (0.4–2.8) | 0.5 (0.1–1.3) | ||
Family history of gastric cancer | none | 3.6 (1.0–12.6) | 0.6 (0.1–7.7) | ||
Less than HS | HS or more | 1.5 (0.4–5.1) | 0.99 (0.2–4.2) | ||
Former smoker | Never smoker | 1.9 (0.8–4.9) | 1.2 (0.4–3.3) | ||
Current smoker | Never smoker | 4.5 (1.4–14.8) | 3.0 (0.7–13.7) | ||
BMI ≥ 30 | BMI<30 | 1.23 (0.4–3.8) | 3.95 (1.1–14.3) |
aModel includes – HP, Family hx of gastric cancer, education, smoking, BMI.
Among the 209 controls enrolled in this nested case–control study, 198 were determined to be PG seronegative, yielding a specificity of 94.7% (95% CI, 90.8–97.3). Among the 105 cases, 33 were determined to be seropositive yielding an estimated sensitivity of 31.4% (95% CI, 22.7–41.2). In a subgroup analysis, PG yielded low sensitivity and high specificity for both noncardia [44.3%; 95% CI, 32.4–56.7) and 93.6% (95% CI, 88.2–97.0), respectively] and cardia gastric cancer [5.7%; 95% CI, 0.7–19.2) and 97.2% (95% CI, 90.3–99.7) respectively; Table 4]. Single variable and multivariable models, as well as sensitivity and specificity using different definitions of PG sensitivity were also examined and found to not differ significantly (Supplementary Table S3). In addition, no clear difference was observed by signet ring cell type as compared with nonsignet cell ring type (Supplementary Table S4).
Examination by groups according to ABC method.
. | . | Cases, n . | Controls, n . | Single variable model HR (95% CI) . | Multivariable modela aHR (95% CI) . |
---|---|---|---|---|---|
All gastric cancer (105 gastric cancer, 209 controls) | Group A (HP-/PG-) | 27 (26%) | 70 (33%) | Ref | Ref |
Group B (HP+/PG-) | 45 (43%) | 128 (61%) | 1.04 (0.6–1.9) | 0.9 (0.5–1.6) | |
Group C (PG+) | 33 (31%) | 11 (5%) | 8.8 (3.5–22.1) | 9.3 (3.4–25.2) | |
Noncardia gastric cancer (70 non-cardia, 141 matched controls) | Group A (HP-/PG-) | 8 (11%) | 40 (28%) | Ref | Ref |
Group B (HP+/PG-) | 31 (44%) | 92 (65%) | 1.7 (0.7–4.2) | 1.44 (0.5–3.8) | |
Group C (PG+) | 31 (44%) | 9 (6%) | 17.0 (5.2–55.6) | 18.8 (5.1–68.7) | |
Cardia gastric cancer (35 cardia, 72 matched controls) | Group A (HP-/PG-) | 19 (54%) | 32 (44%) | Ref | Ref |
Group B (HP+/PG-) | 14 (40%) | 38 (53%) | 0.7 (0.3–1.5) | 0.5 (0.2–1.3) | |
Group C (PG+) | 2 (6%) | 2 (3%) | 1.5 (0.2–11.4) | 2.8 (0.3–25.7) |
. | . | Cases, n . | Controls, n . | Single variable model HR (95% CI) . | Multivariable modela aHR (95% CI) . |
---|---|---|---|---|---|
All gastric cancer (105 gastric cancer, 209 controls) | Group A (HP-/PG-) | 27 (26%) | 70 (33%) | Ref | Ref |
Group B (HP+/PG-) | 45 (43%) | 128 (61%) | 1.04 (0.6–1.9) | 0.9 (0.5–1.6) | |
Group C (PG+) | 33 (31%) | 11 (5%) | 8.8 (3.5–22.1) | 9.3 (3.4–25.2) | |
Noncardia gastric cancer (70 non-cardia, 141 matched controls) | Group A (HP-/PG-) | 8 (11%) | 40 (28%) | Ref | Ref |
Group B (HP+/PG-) | 31 (44%) | 92 (65%) | 1.7 (0.7–4.2) | 1.44 (0.5–3.8) | |
Group C (PG+) | 31 (44%) | 9 (6%) | 17.0 (5.2–55.6) | 18.8 (5.1–68.7) | |
Cardia gastric cancer (35 cardia, 72 matched controls) | Group A (HP-/PG-) | 19 (54%) | 32 (44%) | Ref | Ref |
Group B (HP+/PG-) | 14 (40%) | 38 (53%) | 0.7 (0.3–1.5) | 0.5 (0.2–1.3) | |
Group C (PG+) | 2 (6%) | 2 (3%) | 1.5 (0.2–11.4) | 2.8 (0.3–25.7) |
aModel includes – HP, Family hx of gastric cancer, education, smoking, BMI.
The association of PG with gastric cancer was stratified by the median interval of time (6.7 years) between baseline to diagnosis; 53 cases (with the corresponding 106 controls) with the shorter interval and 52 gastric cancer (with the corresponding 103 controls) with the longer interval. The association of PG+ with gastric cancer at a shorter interval (OR, 8.02; 95% CI, 2.69–23.95) and longer interval (OR, 9.23; 95% CI, 2.66–32.03) was similar, suggesting that the association of PG with gastric cancer was relatively stable over an extended period of time.
Finally, risk of gastric cancer was examined as Group A (HP-/PG-), Group B (HP+/PG-), and Group C (PG+). We found no difference for Group B compared with Group A (reference group; OR, 0.9; 95% CI, 0.5–1.6). Risk of gastric cancer in Group C, incorporating PG status, was significantly different compared with Group A both in the univariable model (Group C: OR, 8.8; 95% CI, 3.5–22.1), as well as the adjusted model (Group C: aOR, 9.3; 95% CI, 3.4–25.2). Among the non-cardia gastric cancer cohort alone, again no difference was found for Group B compared with Group A; however, Group C showed a substantial increase in risk (aOR, 18.8; 95% CI, 5.1–68.7). For cardia gastric cancer, no difference in risk was found by Group categorization. (Table 5).
Test characteristics of PG.
. | All gastric cancer (105 cases, 209 matched controls) . | Noncardia gastric cancer (70 non-cardia, 141 matched controls) . | Cardia gastric cancer (35 cardia, 72 matched controls) . |
---|---|---|---|
Sensitivity (95% CI) | 31.4% (22.7–41.2) | 44.3% (32.4–56.7) | 5.7% (0.7–19.2) |
Specificity (95% CI) | 94.7% (90.8–97.3) | 93.6% (88.2–97.0) | 97.2% (90.3–99.7) |
. | All gastric cancer (105 cases, 209 matched controls) . | Noncardia gastric cancer (70 non-cardia, 141 matched controls) . | Cardia gastric cancer (35 cardia, 72 matched controls) . |
---|---|---|---|
Sensitivity (95% CI) | 31.4% (22.7–41.2) | 44.3% (32.4–56.7) | 5.7% (0.7–19.2) |
Specificity (95% CI) | 94.7% (90.8–97.3) | 93.6% (88.2–97.0) | 97.2% (90.3–99.7) |
Discussion
Our nested case–control study from a large prospective study was the first to examine the utility of PG as a predictor of gastric cancer in a broad, generalizable U.S. population. Despite the low incidence of gastric cancer in the United States, the findings in our study are similar to prior studies conducted in Asia and Europe (7, 10, 14–16, 20, 23–29), and corroborate the predictive role of PG for gastric cancer. In our examination of the PLCO cohort, PG+ consistently conferred higher risk for gastric cancer. Stratified analyses for cardia and noncardia gastric cancer demonstrated an association of increased risk for PG+ individuals for noncardia gastric cancer but not for cardia gastric cancer, similar to what other studies have found (7, 17, 30). In addition, we found high specificity of PG for both noncardia and cardia gastric cancer, suggesting that PG may offer a method to noninvasively identify individuals at high risk for gastric cancer.
Extensive research on gastric cancer screening has been conducted in East Asia where there is a higher prevalence of gastric cancer (31). Given the importance of the Correa pathway in the development of gastric cancer, changes in pepsinogen levels, signaling atrophy of the stomach mucosa, are potentially well suited to be biomarkers for the development gastric cancer precursors (e.g., intestinal metaplasia and dysplasia) as well as gastric cancer. Numerous studies have used the PG I and PGR combination to determine the utility of pepsinogen as a biomarker for gastric cancer. On the basis of cutoffs used, ranging between 10 and 70 ng/mL for PG I and 2–4.5 for PGR, studies have shown differing sensitivity and specificity (7, 10, 14–16, 23–28). The most widely accepted cutoff for PG+, and the one used in this study – serum PG I levels ≤ 70 μg/L and PG I/II ratio ≤ 3.0 – was most predictive in a meta-analysis of PG+ and gastric cancer in population-based studies (20). Using these cutoffs, a meta-analysis of 27 studies, including 18 cohort studies, reported a pooled sensitivity of 59%, and specificity of 73% for gastric cancer (27), and a meta-analysis of 8 studies, including 1 cohort study, reported pooled sensitivity of 59% and specificity of 89% for the gastric cancer precursor chronic atrophic gastritis (32).
The examination of the utility of PG+ for non-cardia gastric cancer compared with cardia gastric cancer has produced mixed results. Some studies have found PG levels to be inversely associated with the detection of cardia gastric cancer (30), while others were unable to find any significant association (7, 17). This difference is thought to be attributed to the fact that most noncardia gastric cancers are related to H. pylori infection, and follow the Correa pathway (33–35), while cardia cancers have at least two separate etiologies. The first is associated with H. pylori and resembles noncardia cancers (36) while the second is not associated with H. pylori and resembles esophageal adenocarcinoma (36). As cardia gastric cancer does not always correlate to H. pylori exposure and the Correa pathway, PG+ would not be expected to be associated with these types of gastric cancer. This would especially be true for Western populations in which the incidence of cardia gastric cancer has been thought to be increasing (37–39), with GERD and obesity playing an important role (40–42).
One unexpected finding of this study was the lack of an association of H. pylori antibodies at baseline and gastric cancer incidence. The role of H. pylori in increasing the risk for gastric cancer has been well established in the literature (35, 43). It was officially declared a human carcinogen in 1994 (44, 45). Over 4 billion people, more than half of the world's population, are affected by H. pylori infections (46). In the United States, H. pylori infection rates vary greatly by race; for example, in a national dataset, while the prevalence was found to be 26.2% in non-Hispanic Whites, it was 61.6% for non-Hispanic Blacks and 61.6% for Mexican Americans (47). Recent literature suggests that 90% of gastric cancer is associated with H. pylori (48). Using a combination of H. pylori and pepsinogen to determine risk for gastric cancer began in 1993 (49), as the “ABC” model, where gastric cancer risk was stratified into three levels from lowest to highest risk (Group A: HP-/PG-, Group B: HP+/PG-, Group C: PG+). Later, the “ABCD” model was created whereby Group C was further divided by H. pylori status into Group C (HP+/PG+) and Group D (HP-/PG+). HP- in the setting of PG+ reflects highest risk, in which the extreme loss of gastric mucosa no longer makes it a viable environment for H. pylori to survive. Since serum PG levels and H. pylori antibody titers are relatively stable, the study investigators suggested this to be an effective means of stratifying persons into risk categories (21). Multiple studies have evaluated this method with various degrees of success in risk stratification based on a person's biomarker results (22, 50–53).
It is unclear why H. pylori was not predictive in our study. One possibility may be the virulence factors expressed by the bacterial stains represented in this cohort of patients (54). Cytotoxin-associated gene pathogenicity island-encoded protein (CagA) and vacuolating cytotoxin (VacA) contribute to differences in disease severity and have been used as individual markers for determining the risk of H. pylori infection progressing to cancer (55). Distinct variation of H. pylori CagA and VacA subtypes are found globally (56). For example, more than 90% to 95% of H. pylori-positive individuals in East Asian countries such as Japan, Korea, and China are CagA-positive, while approximately 40% to 60% of strains in Western nations are CagA-positive (57, 58). Within the United States, CagA-positive strains have been found more commonly among people of non-white race (59), and the PLCO population is predominantly white. These differences along with the dramatically decreased prevalence of H. pylori infections over time are possible reasons gastric cancer rates are so much lower in Western nations (57, 58). Thus, the lower rates of virulent strains found in the West may be part of the explanation for the lack of association between H. pylori and gastric cancer in this U.S. cohort of patients.
High-burden countries like Korea and Japan have experienced a 30–60% decrease in mortality rates following the implementation of population-based endoscopic screening and treatment (60–67), with a concomitant stage shift from late- to early-stage gastric cancer and better overall 5-year survival. However, gastric cancer is not as common in the United States, making the use of endoscopy as a population-wide screening tool unacceptable in terms of population benefits and harms. When screening for less common cancers, the screening test to identify high-risk individuals must be noninvasive and inexpensive for widespread applicability, and have high specificity (hence low false-positive rates) to reduce the number of people subjected unnecessarily to costly diagnostic procedures and psychologic stress (68). Pepsinogen is a well-developed ELISA-based blood test that is simple, cheap, and non-invasive and thus, we propose, has great potential to be a cost-effective method to identify individuals at increased risk for noncardia gastric cancer who should undergo further testing. Despite the modest sensitivity that was found in this sample of U.S. patients, we believe PG deserves to continue to be explored in prospective U.S. studies, in combination with other biomarkers, as a potential screening tool for gastric cancer in the United States.
Strengths of this study include the use of prediagnostic samples from a large prospective cohort study representative of the U.S. population. It is particularly encouraging that the use of PG+ was found to be statistically significant for the detection of noncardia gastric cancer, which is in line with more extensive studies of higher incidence populations.
A limitation of using serum PG is that it does not detect the existence of cancer per se, but is a measure of the mucosal changes that lead to gastric cancer. However, as it is believed that most non-cardia gastric cancer is associated with H. pylori and follows the Correa pathway, PG offers both a chance to detect gastric cancer as well as an opportunity to find precursors of gastric cancer whereby local treatment (such as endoscopic excision of dysplasia) can prevent the development of gastric cancer. Another limitation of the present study is that the analyses were unable to examine risk associated with age, sex, and race, since the controls were matched on these factors to the cases. No data regarding patient history of H. pylori was recorded, and we were unable to include information about previous or current infections or treatment for H. pylori. Finally, the PLCO was a randomized control trial, despite it being a relatively large study, it reflects only people in the study, not the entirety of the U.S. population, including less representation of minorities with only about 14% of the sample being minorities (6% non-Hispanic blacks; 2% Hispanics; and 6% Asian or Pacific Islanders; ref. 69).
Prediagnostic serum pepsinogen levels from a large prospective cohort study were strongly associated with development of noncardia gastric cancer but not cardia gastric cancer in a low-risk U.S. population. PG shows promise as a potential risk biomarker to identify individuals at higher risk of noncardia gastric cancer for targeted screening or interventions in the United States.
Authors' Disclosures
No disclosures were reported.
Disclaimer
The opinions expressed by the authors are their own and this material should not be interpreted as representing the official viewpoint of the U.S. Department of Health and Human Services, the National Institutes of Health, or the National Cancer Institute.
Authors' Contributions
H. In: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. S. Sarkar: Data curation, writing–original draft, project administration, writing–review and editing. J. Ward: Investigation, project administration, writing–review and editing. P. Friedmann: Formal analysis, validation, methodology, writing–review and editing. M. Parides: Formal analysis, methodology. J. Yang: Investigation, methodology, writing–review and editing. M. Epplein: Conceptualization, methodology, writing–review and editing.
Acknowledgments
H. In was supported by The Society for Surgery of the Alimentary Tract (SSAT) Health Care Disparities Research Award and NIH/NCATS grant 5UL1TR002556–05 (Clinical and Translational Science Award). The authors thank the National Cancer Institute for access to NCI's data collected by the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.