Causal analysis of gastroesophageal reflux disease and esophageal cancer

Abstract

Patients with gastroesophageal reflux disease (GERD) are more likely to develop esophageal cancer (EC). However, a causal relationship between the 2 has been difficult to determine. Therefore, this study aimed to evaluate the impact of GERD on EC using the Mendelian randomization (MR) method. The causal association between GERD and EC was analyzed based on 2 publicly available genetic summary datasets for the GERD cohort (129,080 cases vs 473,524 controls) and the EC cohort (740 cases vs 372,016 controls). The causal inference was mainly evaluated by the inverse variance weighted MR. The MR-Egger regression, MR Pleiotropy Residual Sum and Outlier test, and leave-one-out test were used to confirm the sensitivity of the MR results. Possible interfering factors were excluded by multivariate MR (MVMR) analysis. We used 73 single nucleotide polymorphisms as instrumental variables. GERD was associated with increasing EC risk (odds ratio [OR], 1.001; 95% confidence interval, 1.001–1.002; P < .001), which was identified using the inverse variance weighted method. The sensitivity analysis also demonstrated similar results with the causal explanation, and major bias in genetic pleiotropy was not identified (intercept, 0.001; standard error, 0.001; P = .418). The multivariate MR analysis demonstrated the effect of GERD on EC even after excluding possible mediating factors (OR, 1.003; 95% confidence interval, 1.001–1.005; P = .012). This study confirmed that GERD has a causal effect on EC. Therefore, interventional measures are recommended to prevent EC.

1. Introduction

Esophageal cancer (EC) is a common malignant tumor of the digestive tract originating in the epithelium of the esophageal mucosa; its pathological types include squamous cell carcinoma and adenocarcinoma, amongst others.^[1] In recent years, the worldwide EC incidence has been increasing annually, currently ranking eighth of all carcinomas worldwide.^[2] Moreover, the EC mortality rate ranks sixth among all carcinomas worldwide.^[2] The higher mortality rate may be because early symptoms are easy to miss. Therefore, most cases are diagnosed in the advanced or late stages, resulting in a poor prognosis.^[3] Observational studies have demonstrated that smoking, alcohol abuse, and obesity are risk factors for EC.^[1] The occurrence and development of EC are often accompanied by esophageal inflammation.^[4] Therefore, understanding the risk factors is crucial for the secondary prevention of EC.

Gastroesophageal reflux disease (GERD) is characterized by uncomfortable symptoms and complications caused by the reflux of gastric contents. GERD is a common clinical disease with an increasing incidence, seriously affecting human health.^[5] Clinical studies have reported that GERD can lead to EC.^[6–8] In addition, a rat duodenal reflux model confirmed that reflux caused EC.^[9] Although GERD and EC are strongly correlated, most studies have used traditional epidemiological methods, making it difficult to exclude the possibility of bias, such as confounding factors and reverse causal relationships, when determining GERD’s potential effects on EC.

The Mendelian randomization (MR) method assesses the potential causal relationships between risk factors and disease outcomes based on genetic variations.^[10] Because the allocation of genes from parent to offspring is random, MR is considered a “randomized controlled trial of nature.”^[11] Furthermore, since the genetic variation is randomly assigned at conception, it is not affected by common confounding factors, such as environmental, social, and behavioral factors; thus, it is not susceptible to confounding or reverse causation.^[12] Therefore, we used the MR method to investigate GERD’s effects on EC, to better guide clinical practice.

2. Materials and methods

European populations were selected as exposure sets from the largest publicly available genome-wide association study (GWAS) database (Table S1, Supplemental Digital Content, http://links.lww.com/MD/L862). This study comprised 602,604 participants (129,080 GERD cases and 473,524 controls).^[13] All patients with GERD were diagnosed based on the 10th edition of the International Classification of Diseases. Previous studies identified 2320,781 single nucleotide polymorphisms (SNPs) in this database, and 3.9% of the variation in GERD was explained by these SNPs (F statistic = 13.53, indicating a strong predictive effect of the instrument on GERD).^[14] The following criteria were set for SNP screening to ensure robustness and reliability of the dataset: GERD-related SNPs (P < 5.0 × 10⁻⁸), linkage disequilibrium (r² < 0.001), and genetic distance < 10,000 kb.^[15,16]

The GWAS database related to EC was obtained from the United Kingdom Biobank.^[17] Its sample size was 372,756, including 740 EC cases and 372,016 controls, with 8970,465 SNPs (Table S1, Supplemental Digital Content, http://links.lww.com/MD/L862). All cases of EC were consistent with the 10th edition of the International Classification of Diseases diagnostic codes, including esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC); the specific composition ratio was unclear. All the participants in the database were of European descent. Previous studies have demonstrated the reliability of the above database.^[18,19]

Exploring whether GERD’s effects on EC through other mediating factors was also necessary. Therefore, we evaluated the causal effect of GERD on risk factors for EC. Risk factors for EC were determined based on existing databases and common clinical risk factors.^[20] We identified 5 risk factors for EC: hypertension, type 2 diabetes, smoking, alcohol consumption, and body mass index (BMI). Table 2 summarizes the GWAS data for these risk factors. Five methods were used to determine the MR estimates for GERD and EC.

Table 2.

Causal effects from GERD to common risk factors of EC by the IVW method.

Risk factors	nSNPs	OR (95%CI)	P-value
BMI	73	1.393 (1.298–1.497)	<.001
Cigarettes per day	73	1.254 (1.168–1.347)	<.001
Alcoholic drinks per week	73	0.979 (0.949–0.998)	.101
Hypertension	76	1.002 (1.001–1.003)	.001
Type 2 diabetes	76	1.209 (1.045–1.399)	.011

BMI = body mass index, CI = confidence interval, EC = esophageal cancer, GERD = gastroesophageal reflux disease, IVW = inverse variance weighted, nSNPs = number of single-nucleotide polymorphisms, OR = odds ratio.

Inverse variance weighting (IVW) was the main MR method used to determine the potential causal relationships between GERD, EC, and mediating factors.^[21] The heterogeneity was detected, and the randomized-effects IVW model was applied; otherwise, the fixed-effects model was used.^[22] The weighted median method was used as a supplementary method for at least 50% of the analysis weights from inverse variance (IV).^[23] If all SNPs were ineffective tools, then the MR–Egger method was used.^[24] In addition, simple weighted model methods were applied for causal estimates such as odds ratios with their 95% confidence intervals.

Only IVs meeting the following criteria were included in the MR analysis: (i) associated with GERD; (ii) not related to factors affecting the relationship between GERD and EC; and (iii) affecting EC only through GERD. We used the MR-Egger regression, MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) test, and leave-one-out test for a sensitivity analysis to evaluate the reliability of the results. To test the pleiotropic level of IV, we used MR-Egger regression; a P-value of >.05 indicated no pleiotropy. MR-PRESSO test was conducted to evaluate the horizontal pleiotropy. Furthermore, a leave-one-out test was used to assess robust MR estimates. Finally, multivariate MR (MVMR) was used to exclude the interference of mediating factors and evaluate the direct effect of exposure on the outcomes.^[25] No additional ethical approval was required for this study as published GWAS summary statistics were used. All statistical analyses in this MR were performed using the “TwoSampleMR” software package in R (version 4.0.3; R Core Team, Vienna, Austria).

3. Results

We selected 73 SNPs of GERD from the GWAS database as instrumental variables after removing 4 SNPs that lacked relevant information and 3 SNPS (rs2145318, rs2358016, rs957345) that were palindromic (Table S2, Supplemental Digital Content, http://links.lww.com/MD/L863) and found a significant association between GERD and EC (Table 1). Specifically, using the IVW method, we found that the EC risk increased by 0.1% for every 1 standard deviation increase in GERD. Although not statistically significant, similar results were obtained using the other 4 methods (Table 1). A forest plot illustrating GERD’s effects on EC also indicated that GERD increased the risk of EC (Fig. 1). No significant intercept was found in the MR–Egger regression (intercept, 0.001; standard error, 0.001; P = .418), indicating no pleiotropy. We conducted MR-PRESSO analysis, and the result showed no horizontal pleiotropy (RSSobs = 76.92; P = .379). The fixed-effects IVW model was applied because no heterogeneity was detected. Based on the omission analysis results, the overall MR results were robust and unaffected by the SNPs (Fig. 2). The funnel plot and scatter plot showed the consistency of the results (Figs. 3 and 4).

Table 1.

Mendelian randomization estimates of the causality between GERD and EC.

Exposure	Outcome	nSNPs	Method	OR (95% CI)	P-value
GERD	EC	73	MR-Egger	1.003 (0.999–1.008)	.170
			Weighted median	1.001 (1.000–1.002)	.070
			Inverse variance weighted	1.001 (1.00–11.002)	<.001
			Simple mode	1.000 (0.997–1.003)	.847
			Weighted mode	1.000 (0.997–1.002)	.850

CI = confidence interval, EC = esophageal cancer, GERD = gastroesophageal reflux disease, MR = Mendelian randomization, nSNPs = number of single-nucleotide polymorphisms, OR = odds ratio.

Figure 1.

Forest plot of single nucleotide polymorphisms associated with gastroesophageal reflux disease and the risk of esophageal cancer. MR = Mendelian randomization.

Figure 2.

Leave-one-out analysis of single nucleotide polymorphisms associated with gastroesophageal reflux disease and the risk of esophageal cancer. MR = Mendelian randomization.

Figure 3.

The funnel plot of single nucleotide polymorphisms associated with gastroesophageal reflux disease and the risk of esophageal cancer. MR, Mendelian randomization; 1/SE_IV, instrument precision; β_IV, in (OR) estimates.

Figure 4.

The scatter plot of single nucleotide polymorphisms associated with gastroesophageal reflux disease and the risk of esophageal cancer. MR = Mendelian randomization.

We also investigated possible mediating factors in the causal relationship between GERD and EC. The previously reported EC risk factors were significantly associated with GERD (Table 2). Specifically, GERD was significantly associated with hypertension, smoking, BMI, and type 2 diabetes, suggesting that these factors are important components of the GERD-EC pathway. To control for multiple and potential exposure pathways, the MVMR analysis included GERD and the above intermediate factors to estimate the direct effect of GERD on EC. None of these factors had a causal effect on EC in the MVMR analysis (all P > .05). However, GERD remained significantly positively associated with EC after adjusting for these factors (Table 3).

Table 3.

MVMR results for EC.

Outcome	Exposure	nSNPs	OR (95% CI)	P-value
EC	GERD	24	1.003 (1.001–1.005)	0.012
	Hypertension	0	NA	NA
	Type 2 diabetes	3	1.000 (1.000–1.001)	0.114
	Cigarettes per day	5	1.001 (1.000–1.002)	0.739
	BMI	458	1.002 (1.000–1.003)	0.834

BMI = body mass index, CI = confidence interval, EC = esophageal cancer, GERD = gastroesophageal reflux disease, MVMR = multivariable Mendelian randomization, NA = not available, OR = odds ratio, SNP = single-nucleotide polymorphism.

4. Discussion

This study used large-scale GWAS summary data to conduct two-sample MR analyses, demonstrating that GERD slightly promoted EC. To our knowledge, this is the first large-scale MR analysis of the causal effect of GERD on EC. These analytical methods avoid confounding factors from a genetic perspective, which helps clarify the causal link and effect between GERD and EC.

Multiple observational studies have shown that GERD increases the risk of EC,^[26,27] and GERD has been associated with the development of EC. Specifically, the duration of GERD is positively associated with the risk of EAC. For example, patients with GERD symptoms persisting for over 20 years have a 6 times higher risk of EAC.^[28] Additionally, one study suggested that Barrett esophagus, a pathological phenomenon in which the normal esophageal squamous epithelium is replaced by columnar epithelium, results from chronic GERD injury. Patients with Barrett esophagus have a 30 to 40-fold increased risk of EC.^[29] Furthermore, a study of healthy young men in Western developed countries showed that the EAC subtype became more common as the prevalence of GERD and Barrett esophagus increased.^[30] However, the incidence of ESCC is also increasing worldwide, and a duodenal content reflux rat model suggested that EAC and ESCC were related to GERD.^[31]Currently, the relationship between GERD and ESCC remains controversial, with some population-based studies supporting an association between the 2,^[32,33] whereas others do not.^[34,35] These conflicting results may be the result of potential confounding factors and reverse causal relationships identified in observational studies. In this study, we reduced such effects by using an MR analysis. Our robust and reliable results support GERD’s effects on EC, and provide additional evidence to the existing research.

The mechanism by which GERD’s effects on EC is still being investigated; however, existing studies suggest that chronic exposure of the esophagus to gastric and bile acids can lead to chronic inflammation and promote cancer development.^[36] With the development of genomic technology, researchers have gained a new understanding of EC. The presence of gastric and bile acids may influence the expression of tumor necrosis factor, that can in turn mediate a shift from cell apoptosis to cell proliferation.^[37] Tumor necrosis factor also promotes the progression of EC. Additionally, p53 expression gradually increases in normal mucosa, Barrett esophagus, and the mucosa of EAC, indicating that p53 is a risk factor for EAC.^[38] Moreover, studies have reported that NOTCH mutations are risk factors for ESCC, and they also play a role in the occurrence of ESCC.^[39] In addition, microbial changes in patients with GERD may increase immune activity, potentially leading to chronic inflammation and eventually to EC development.^[40,41]

We also investigated possible mediating factors in the causal relationship between GERD and EC, focusing on intermediate phenotypes identified in the databases. We found that hypertension, type 2 diabetes, smoking, and BMI may be important components of the GERD-EC pathway. Furthermore, we applied an MVMR model to correct for the above intermediate phenotypes, finding that the above factors no longer had causal roles in EC. However, GERD maintained a significant association with EC even after excluding possible mediating factors.

Our MR analysis has some limitations. First, the study’s generalizability to other populations may be limited because all participants were from Europe. Second, the EC incidence rate varies by region. For example, ESCC is highly prevalent in East Asia, Eastern and Southern Africa, and Southern Europe, whereas EAC is more prevalent in North America and other European countries. However, owing to the lack of relevant data we failed to analyze EAC and ESCC hierarchically, and more extensive studies of EC subgroups are needed. Third, because of limited resources for the GWAS database, we did not perform stratified analyses based on important covariates such as age, sex, and tumor stages. At last, the biological mechanisms between GERD and EC are still being explored. Therefore, the results of this study only provide partial evidence for GERD’s effects on EC.

5. Conclusion

This MR study support GERD’s effects on EC. The information provided by this study calls for the implementation of interventional measures for preventing EC. Future studies should examine the biological mechanisms underlying this association.

Acknowledgments

We express our gratitude to the participants and research teams who made the GWAS results publicly accessible. Thank you for the language assistance provided by Editage.

Author contributions

Conceptualization: Min Kang, Shuangyue Wang.

Data curation: Shuangyue Wang, Zhiru Li, Ziyan Zhou.

Formal analysis: Shuangyue Wang.

Writing – original draft: Min Kang, Shuangyue Wang, Zhiru Li, Ziyan Zhou.

Writing – review & editing: Min Kang, Shuangyue Wang, Zhiru Li, Ziyan Zhou.