Association of Common Use Pharmaceuticals in Reducing Risk of Esophageal Adenocarcinoma: A SEER–Medicare Analysis

Esophageal cancer is the ninth most common cancer worldwide and fifth leading cause of cancer-related deaths (1). Esophageal adenocarcinoma (EAC) incidence and mortality have increased markedly, with a rising trend ranging from 3.5% to 8.1% annually (2). EAC has become the dominant histotype of esophageal cancer in Westernized countries, reflecting an incidence of 0.7 cases per 100,000 persons (3). Furthermore, the 5-year survival for EAC has improved only marginally, from 9% in the 1970s to 22% by 2009 (4). These alarming trends highlight the public health impetus to understand the underlying mechanisms of EAC.

Multiple risk factors for EAC have been proposed. Epidemiologic studies have demonstrated associations between EAC and family history, smoking, older age, male gender, central obesity, gastroesophageal reflux disease (GERD), and Barrett's esophagus (BE; refs. 2, 5–8). BE is the predominant risk factor and only established precursor lesion associated with the development of EAC (9). BE develops when chronic gastroesophageal reflux damages the esophageal mucosa, replacing stratified squamous epithelial cells with intestinal-like columnar cells (10–12). Evidence suggests that when BE progresses to EAC, it occurs in a stepwise manner, from metaplasia to low-grade to high-grade dysplasia before invasion (13). The overall risk of EAC development among individuals with BE may be up to 5% with an average estimated rate of BE to EAC transition is approximately 0.5% per person-year (1 in 200; refs. 9, 11, 14). Additionally, the necessity of BE for the development of EAC remains controversial. EAC tumors are thought to originate from, and ultimately replace, esophageal intestinal metaplasia (IM), a theory supported by a case study of 74 patients who underwent esophagogastrectomy for EAC that found IM in 100% of tumors less than 1 cm in diameter (15). Though reasonable, to date this assumption has not been verified in any prospective clinical studies. Despite the high prevalence of BE in the general population (estimate 5.5%–5.7%), EAC incidence remains relatively rare, suggesting either a low rate of transformation to EAC or other unidentified requisite factors among progressors.

Current American Gastroenterological Association (AGA) guidelines recommend routine endoscopic surveillance for BE for individuals with multiple EAC risk factors (e.g., age, male, chronic GERD) at a frequency of 3 months to 5 years, depending on the degree of dysplasia (16). Several new screening techniques were recently developed, such as swallowable imaging capsules, transnasal endoscopy, and cell collection devices; however, these are not yet recommended by the AGA (17, 18). For individuals with high-grade dysplasia, endoscopic eradication therapy with radiofrequency ablation (RFA), photodynamic therapy, or endoscopic mucosal resection is suggested (16). Although RFA is a commonly utilized and highly successful strategy for the eradication of high-grade dysplasia and IM, with significantly fewer reported EAC cases posttreatment compared with individuals not receiving RFA, procedure-related complications, such as formation of strictures and chest discomfort, are not trivial (19, 20). These untoward effects, together with known variability in adherence to guidelines for BE and the high cost of procedures, have led to continued investigation of novel approached to BE care (21–24).

If effective, medical management of BE with proton pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs (NSAIDs), statins, or metformin may reduce the need for ablative therapy or resection. PPIs have been used to control reflux symptoms and may slow or inhibit the progression of BE by increasing the pH of gastric secretions (25). A meta-analysis examining the association between BE and EAC found a 71% reduction in the risk of BE progression to high-grade dysplasia or EAC with PPIs for 2 to 3 years of use, suggesting a cost-effective strategy for EAC prevention (26). Recently, the Aspirin and Esomeprazole Chemoprevention in Barret's Metaplasia Trial (AspECT) found a lower incidence of progression to high-grade dysplasia or EAC and all-cause mortality in participants randomized to high-dose PPI with aspirin compared with those on high-dose PPI, low-dose PPI, or low-dose PPI with aspirin (27). Additionally, several prospective and population-based studies showed significantly reduced risk of EAC among those with and without BE following NSAID or aspirin use (28, 29). Long-term statin use has also demonstrated a protective effect against the development of several gastrointestinal disorders, including esophagitis, BE and EAC, and reduced EAC-specific mortality (30–32). A recent systematic review indicated that chronic statin use had the most pronounced protective effect against EAC among individuals with preexisting BE (31). Of late, due to an observed association between type 2 diabetes and BE and promising in vitro and in vivo studies, a preventive effect of metformin has been proposed (33, 34).

Using Surveillance, Epidemiology, and End Results (SEER) Program data linked to Medicare records, and based on the classic framework of EAC risk (e.g., BE as a precursor lesion to EAC), we first investigated the association between BE and EAC risk, hypothesizing that a diagnosis of BE would be more frequent among individuals with EAC. Second, to investigate the effects of medications regularly prescribed for common chronic conditions on EAC risk in individuals diagnosed with BE, we analyzed use of PPIs, NSAIDs, statins, and metformin in our total study population, as well as stratified by BE status. We hypothesized that, among individuals with BE, claims for prescribed PPIs, NSAIDs, statins, or metformin would be associated with a lower risk of EAC and that, likewise, claims for any of those drugs by individuals with BE, would be associated with a lower EAC risk, compared with individuals without BE with those prescription claims or those without claims.

The SEER 18 Program database of the NCI includes 18 registries (Alaska Native Tumor Registry, Connecticut, Detroit, Atlanta, Rural Georgia, Greater Bay Area Cancer Registry (San Francisco–Oakland, San Jose–Monterey), Hawaii, Iowa, Los Angeles, New Mexico, Seattle-Puget Sound, Utah), covering approximately 34.6% of the U.S. population (https://seer.cancer.gov/about/overview.html). Medicare is a federally administered health insurance program for (i) individuals who are 65 or older, (ii) younger individuals with disabilities, and (iii) people with end-stage renal disease. For the purposes of this analysis, we included individuals ages 65 to 99. Medicare covers hospital and medical insurance (Parts A and B), with optional prescription drug coverage (Part D). Part A coverage includes hospital stays, care in a skilled nursing facility, and home health care. Part B covers outpatient services, including outpatient care, medical supplies, and preventive services. In 2019, approximately 52.2 million individuals age 65 and older were enrolled onto Medicare Part A, and 45.5 million individuals with accompanying Part D coverage.

The SEER–Medicare linkage provides detailed information (clinical, demographic, and cause of death) of Medicare beneficiaries with claims for covered health care services and can be used for a variety of epidemiologic research purposes. For each linkage, updated biennially, 95% of persons age 65 and older in SEER are matched to their Medicare enrollment file (https://www.medicare.gov/what-medicare-covers/your-medicare-coverage-choices/whats-medicare). Demographic and clinical information were obtained for the Patient Entitlement and Diagnosis Summary File (PEDSF). Medicare claims data were extracted from the Medicare Provider Analysis and Review (MEDPAR) file for inpatient stays and the Outpatient Claims file for outpatient stays. This study was performed in accordance with the ethical standards of the 1964 Declaration of Helsinki, and the NIH Office of Human Subjects Research has determined that SEER–Medicare data are exempt from Institutional Review Board review (CFR 46.104; ref. 4), and informed consent was waived.

Medicare beneficiaries age 65 and older diagnosed with EAC between 2007 and 2013 in the SEER 18 registries (excluding the Alaskan Native Tumor Registry) were eligible for the study. Between 2007 and 2013, 1,943 eligible EAC cases were identified in the SEER–Medicare database. Inclusion criteria for cases included: (i) enrollment in Medicare for age, (ii) participant age 66 to 99 at time of diagnosis or end of the study year, (iii) enrolled in Medicare Part A and Part B non-health maintenance organization (HMO) coverage in the prior year, (iv) Part D enrollment for at least 1 month in the prior year, and (v) ICD-O-3 histology codes 8140–8575 and ICD-9 morphology codes 150.0–150.9 for EAC cases.

A 5% random sample of Medicare enrollees in the SEER 18 geographic regions (excluding Alaskan Native Tumor Registry) were eligible to be selected as controls (N = 931,298). Controls were matched to cases on gender, race, age, and calendar year of Medicare enrollment, and chosen at a ratio of 10 to 1 (Table 1). Inclusion criteria for controls were identical to those of cases; however, these controls lacked the need for histology and morphology codes and must have been enrolled in Medicare with Part D coverage between 2007 and 2013. From the eligible population, 19,430 controls (2.1%) were selected for this study.

History of BE for cases and controls was identified using the ICD-9 codes 530.85. A diagnosis of BE was considered sufficient if the code appeared at least once on the MEDPAR file or at least twice in National Claims History files or outpatient records 30 or more days apart, with the diagnosis date being the date of the first claim.

Analyses of pharmaceutical use were performed on BE and/or EAC cases for Medicare Part D enrollees only. We utilized prescription drug events identified in Medicare Part D for selected pharmaceuticals (Supplementary Table S1). Pharmaceuticals were selected based on clinical recommendations, publications, and available data within Medicare. To be considered a ‘user’ in a drug category, an individual needed to have at least two prescriptions in the same drug category on different days and drug use must have occurred prior to study selection. The use of two different drugs in the same category was considered a user of that category.

ICD-9 codes for selected comorbidities were collected (Supplementary Table S2). Charlson comorbidities were scored according to Quan and colleagues (35). Comorbidities were required to occur prior to EAC diagnosis, but not before BE. Cancer, one of the included comorbidities, was not considered an exclusion if it was from a participant age 65 or older, occurring at or after their first month of Part A or Part B non-HMO coverage, and was prior to their selection date for the current cohort. Those noted with cancer in the comorbidity scoring were cancers other than EAC. Additional comorbid conditions, due to their association with EAC, were also evaluated. These comorbidities included overweight/obesity, GERD/esophagitis, and tobacco use, among others. Charlson and additional comorbidities were scored as 0, 1, and 2+, corresponding to no comorbidities, one comorbidity, and two conditions or one severe condition, respectively.

Participant demographics and comorbidities were compared using χ² or Fisher exact test. Conditional logistic regression was used to generate odds ratios (OR) and 95% confidence intervals (95% CI) to evaluate risk of EAC with use of PPIs, NSAIDs, statins, or metformin. Models were adjusted for age (65–69, 70–79, 80–89, 90–99), race (non-Hispanic white, non-Hispanic black, Hispanic, Other), gender, Charlson comorbidity score (0, 1, 2+), and non-Charlson comorbidity score (0, 1, 2+). Wald χ² tests were used to assess the significance of logistic regression model covariates. All data analyses were performed in SAS version 9.4 (SAS Institute), and P values are two-sided.

The demographics of the unmatched variables of study population are shown in Table 2. Matched variables are included in Supplementary Table S3. Compared with controls, EAC cases had significantly higher overall mortality (78.6% vs. 25.9%), Charlson comorbidities (2+: 68.5% vs. 57.5%), and non-Charlson comorbidities (2+: 57.0% vs. 44.9%; Table 2). Participant comorbidities are shown in Table 3. The most prevalent comorbidities among both cases and controls were chronic obstructive pulmonary disease (COPD), type 2 diabetes, and GERD/esophagitis. Of the Charlson comorbidities, EAC cases had significantly higher reports of congestive heart failure (30.5% vs. 25.8%), COPD (47.1% vs. 39.1%), mild liver disease (12.8% vs. 9.9%), moderate or severe liver disease (1.0% vs. 0.6%), malignancies other than malignant neoplasms of the skin or EAC (40.8% vs. 26.7%), metastatic disease (8.1% vs. 3.8%) and peripheral vascular disease (38.3% vs. 33.6%). However, EAC cases had significantly lower reports of dementia (5.8% vs. 8.6%), rheumatic disease (5.5% vs. 7.2%), and hemiplegia or paraplegia (2.5% vs. 3.4%). Evaluating the non-Charlson comorbidities, EAC cases had statistically increased prevalence of GERD/esophagitis (55.5% vs. 36.4%), and tobacco use (34.5% vs. 22.3%), compared with non-EAC controls, and lower prevalence of non-infective inflammatory bowel disease (IBD), IBD, or Crohn's disease (3.3% vs. 5.4%). No differences were noted for obesity (14.4% vs. 13.7%) or type 2 diabetes (45.3% vs. 47.6%) frequencies (Table 3). Further details on Charlson and non-Charlson comorbidities can be found in Supplementary Table S4.

As shown in Table 2, EAC cases had a higher prevalence of BE (26.2%), compared with controls (3.2%). Most BE diagnoses occurred more than 12 months before EAC diagnosis (36.1%). Notably, 22.4% of the BE diagnoses occurred within 1 month of the participants’ diagnosis of EAC, suggesting simultaneous diagnosis of BE and EAC (Table 2). When we examined EAC case demographics, we found that a higher percentage of individuals with BE were between the ages of 80 to 89 (37.9% vs. 27.7%), compared with those without a BE diagnosis (Table 4). EAC cases with BE were diagnosed in earlier stages (stage I: 33.1% vs. 18.4%), had well-differentiated tumors (8.8% vs. 5.1%), underwent surgery (32.2% vs. 25.7%) but less radiation (36.4% vs. 49.3%), and lower mortality (70.1% vs. 80.6%). No differences were noted between gender and race.

Part D enrollment claims by year is shown in Supplementary Table S5. Statins had the most claims in both EAC cases and controls. Among individuals without BE, EAC cases had significantly higher reported use of PPIs (38.2% vs. 4.3%; P < 0.001), NSAIDs (16.4% vs. 3.9%; P < 0.001), statins (52.1% vs. 12.0%; P < 0.001), and metformin (14.3% vs. 3.5%; P < 0.001) compared with controls (Table 5). Similar trends were observed for individuals with BE (PPI: 62.2% vs. 11.3%, P < 0.001; NSAIDs: 12.7% vs. 5.2%, P < 0.001; statins: 52.8% vs. 9.1%, P < 0.001; metformin: 12.2% vs. 3.2%, P < 0.001). EAC cases also had significantly longer duration of PPI, NSAID, statin, and metformin use, compared with controls (Supplementary Table S6). In the total study population, those who used PPIs had a 90% reduced risk of EAC (OR, 0.10; 95% CI, 0.09–0.12), 38% reduced risk with use of NSAIDs (OR, 0.62; 95% CI, 0.51–0.74), 85% reduced risk with use of statins (OR, 0.15; 95% CI, 0.13–0.17), and 24% reduced risk with use of metformin (OR, 0.76; 95% CI, 0.62–0.93; Fig. 1). When stratified by BE status, we found similar magnitudes of EAC risk reduction for individuals without BE as in the total study population (PPI: OR, 0.13; 95% CI, 0.11–0.15; NSAIDs: OR, 0.57; 95% CI, 0.47–0.69; statins: OR, 0.14; 95% CI, 0.12–0.17; metformin: OR, 0.73; 95% CI, 0.59–0.89). However, among individuals with BE, there was no observed association between NSAID or metformin use and EAC risk (NSAIDs: OR, 1.21; 95% CI, 0.61–2.40; metformin: OR, 0.85; 95% CI, 0.39–1.84). Additionally, the association between PPI or statin use and reduction of EAC risk was slightly stronger in the BE population (PPIs: OR 0.08; 95% CI, 0.05–0.13; statins: OR, 0.13; 95% CI, 0.08–0.21).

PPI, NSAID, statin, and metformin use were positively correlated with one another, regardless of BE status, indicating the high likelihood of pharmaceutical co-use within the study population. Therefore, we analyzed co-pharmaceutical use in the study population. Due to limited study numbers, we focused on dual pharmaceutical use, specifically involving PPIs given its frequent use in individuals with BE. Among all participants, the exclusive use of PPIs (e.g., participants without the combined use of NSAIDs, statins, or metformin) had a 98% reduced risk of EAC (OR, 0.02; 95% CI, 0.02–0.03), whereas combined use of PPIs with NSAIDs, statins, or metformin had a 65%, 97%, and 28% reduced risk of EAC, respectively (PPI + NSAIDs: OR, 0.35; 95% CI, 0.26–0.47; PPI + statins: OR, 0.03; 95% CI, 0.03–0.04; PPI + metformin: OR, 0.72; 95% CI, 0.53–0.98; Fig. 2A). The strength of association was reduced when participants used NSAIDs or statins without PPIs; however, the association with metformin use strengthened (OR, 0.69; 95% CI, 0.35–0.88). When stratified by BE status, the associations were similar, with a few notable exceptions. First, the association between metformin use alone and EAC risk was slightly strengthened in individuals without BE prior to selection, with a risk reduction of 35% (OR, 0.65; 95% CI, 0.50–0.82), and no association for individuals with BE (OR, 0.99; 95% CI, 0.28–3.46; Fig. 2B and C). However, when individuals with BE had dual use of PPIs and metformin, risk of EAC reduced by 69% (OR, 0.31; 95% CI, 0.11–0.85). Second, the association between combined use of PPIs and NSAIDs and EAC risk was strengthened among individuals with BE, with a risk reduction of 73% (OR, 0.27; 95% CI, 0.11–0.67), compared with all participants and those without BE (OR, 0.36; 95% CI, 0.26–0.49; Fig. 2B and C).

We first sought to substantiate the BE precursor paradigm in the SEER–Medicare population by demonstrating that the prevalence of known BE in individuals with EAC was indeed significantly higher than in controls. We then investigated the risk of EAC in individuals with and without prior BE who had exposure to one or more of several well-known pharmaceuticals, finding a lower risk of EAC with use of PPIs, NSAIDs, statins, or metformin, compared with those without reported use. Interestingly, we observed differential associations by BE status. For example, the use of prescribed NSAIDs is associated with lowered risk of EAC incidence among individuals without BE, but not for those with a prior diagnosis of BE (Fig. 1). Conversely, the dual use of PPIs and metformin had more significant impact on EAC risk reduction in those with BE, compared with those without a history of BE (Fig. 2). These results suggest not only variability in the impact of the evaluated pharmaceuticals on risk of EAC incidence, but also the combined effect of these agents on EAC risk.

The results of this study strengthen previous evidence suggesting the protective effect of PPI, NSAID, statin, or metformin use against EAC development in the context of BE and provide new insight into the potential benefits of these pharmaceuticals. PPIs are commonly utilized for the alleviation of heartburn (persistent acid reflux) and prescribed for the management of BE, and the link between GERD and EAC has been well documented (36). Evidence of the association between PPI use and gastrointestinal cancer risk has been variable. Consistent with the current findings, a meta-analysis of seven case–control and cohort studies showed a 71% reduced risk of high-grade dysplasia or EAC among individuals with BE and PPI use (26). However, a series of population-based cohort studies in Sweden demonstrated up to a 10-fold higher risk of esophageal or gastric cancer in the first year of PPI use, and four times increased risk of EAC for individuals on PPI maintenance therapy, compared with nonusers, significantly contrasting the results of our study (37).

Aspirin and NSAIDs have well-demonstrated roles in cancer prevention, particularly for gastrointestinal cancers (38–40). An analysis of studies by the BEACON investigators indicated that, compared with nonusers, current NSAID users had a significantly reduced risk of EAC and esophagogastric junction adenocarcinoma (28). However, the role of NSAIDs may have differential effects across cancer types (41–44). For example, randomized clinical trial conducted in China did not observe differences in esophageal squamous cell carcinoma incidence among participants receiving the NSAID celecoxib (45). The impact of statins on cancer risk has been evaluated in multiple gastrointestinal cancers, though the effects have been best characterized in colorectal and gastric cancers (46). Further, the evaluation of long-term statin use among individuals with other comorbid conditions, such as hypercholesterolemia, and those with EAC risk warrant investigation. Several studies indicated reduced risk of EAC with statin use, alone and in combination with NSAIDs or aspirin, as well as a reduction in mortality with use post-diagnosis (29, 47, 48).

The relationship between metformin use and EAC has not been well established. Approximately 30 million people in the United States have diabetes, and metformin remains one of the most commonly used medications for this condition (49). A meta-analysis of epidemiologic studies investigating metformin use and cancer risk among type 2 diabetics found a consistent reduction in risk of cancer incidence at any site, with the most substantial risk reduction observed for colorectal cancer (50, 51). Given its low cost, high tolerability with chronic use, and mechanistic potential for cancer preventive effects, prospective clinical testing of metformin for cancer prevention with validated clinical endpoints is worth proper consideration (52). Although the data suggesting that metformin offers protection against EAC are minimal, the results presented here may begin those discussions.

Although the associations that we observed between selected pharmaceuticals and EAC risk were significant, several limitations of the study must be considered. By definition, the SEER–Medicare linkage includes only Medicare enrollees, who may not be representative of individuals covered by other insurance plans or without insurance coverage. Medicare records also do not include claims for the use of over-the-counter formulations of PPIs and NSAIDs, such that we were unable to assure matched cases and controls for that potential confounder. Despite the widely accepted use of this database for BE studies, another inherent concern is the ICD coding for BE. Prior to 2003, the ICD code 530.2 identified BE diagnoses but included patients with a diagnosis of “ulcers of the esophagus” (53). Because the records used for the current study began in 2007, this issue likely has been corrected with the ICD code 530.85, though the possibility of esophageal ulcers having been included remains. However, the ICD code 530.85 itself is also a complicating factor because before 2013, it applied to all forms of BE, regardless of the presence or absence of dysplastic changes, known prognostic factors. Finally, the fact that billing codes are provided by endoscopists before pathologic confirmation may also account for some misidentification of BE cases. Nevertheless, although there may be some limitations related to BE case identification and generalizability, use of this resource provided highly reliable prescription and EAC data. To our knowledge, this is the first analysis investigating the relationship between the use of common pharmaceuticals, BE and EAC risk using the SEER–Medicare database. Using this rich data source, we were able to evaluate a large population of EAC cases with linked patient medical history data that include prescription drug use. Further, it is possible that the agents of interest may mechanistically affect EAC carcinogenesis differently. For example, the protective effects of PPIs and statins may be necessary in the early stages of carcinogenesis (e.g., initiation), whereas the impact of NSAIDs and metformin may be most beneficial at a later stage (e.g., metastatic progression) of carcinogenesis.

This study provides evidence that the use of PPIs, NSAIDs, statins, and metformin may help protect against the development of EAC in individuals with and without diagnosed BE. Repurposing agents that are approved for common chronic conditions, and either individually or in combination have demonstrated an association with cancer risk reduction. This line of evidence from SEER–Medicare linkage data is an attractive approach to cancer preventive agent identification. Given their previously established low toxicity profiles with long-term administration in similar populations, the typically long, costly safety-testing phase of the drug development process would be dramatically reduced. The high safety profile of these agents makes them particularly appropriate for individuals with a low risk for cancer, such as those with BE in whom a more toxic approach is not justifiable. Further, the agents investigated in this study are highly accessible at a low cost, augmenting their likelihood for clinical dissemination. However, because there may be harms associated with utilizing some of these agents, a cost–benefit analysis of long-term use should be considered. In conclusion, the results of the current study together with the known safety profile, availability, and cost of these agents support further evaluation for preventive effects against the development of EAC in prospective clinical trials.

No disclosures were reported.

H.A. Loomans-Kropp: Conceptualization, formal analysis, methodology, writing–original draft, writing–review, and editing. M. Chaloux: Data curation, software, and formal analysis. E. Richmond: Writing–original draft, writing–review, and editing. A. Umar: Conceptualization, methodology, writing–review, and editing.

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.