The Role of Gastroesophageal Reflux and Other Factors during Progression to Esophageal Adenocarcinoma

This article is featured in Highlights of This Issue, p. 1001

Esophageal adenocarcinoma (EAC) incidence rates have increased over 500% in the United States since 1975, yet mechanistic drivers of this trend are not fully understood (1, 2). In particular, it is not clear how exposures may influence cellular processes during carcinogenesis to explain temporal trends for EAC, or the 5- to 6-fold increased EAC risk for U.S. men compared with women (1–5). Established risk factors include age, symptomatic gastroesophageal reflux disease (sGERD), central-obesity, smoking, White race, male sex, and an inverse association with Helicobacter pylori infection (6–10). Progression to EAC occurs through development of Barrett's esophagus (BE), dysplasia, and adenocarcinoma, with BE risk increasing with reflux severity (10–13). sGERD, defined as weekly or more frequent symptoms of reflux or heartburn, increases BE and EAC risk (10), although most individuals with sGERD do not have BE (14). Obesity and sGERD may be linked mechanistically (15–18), and with H. pylori, age, male gender, smoking, genetic factors, and gene–exposure interactions, influence BE development and EAC progression (19–26). These exposures and anthropometric factors likely account for most of the approximately 5-fold EAC incidence increase in the United States and other Western countries from 1975 to 2009 (2, 3, 27–29).

Long-term sGERD trends are unknown, but three U.S. population-based studies of residents of Olmsted County, Minnesota suggest that sGERD prevalence increased from about 13% to 20% in the early 1990s (30–32). Two UK studies of sGERD incidence show low incidence rates among children and higher rates among adults (33, 34). sGERD prevalence in Western countries ranges between 10% and 20% (35).

Long-term trends for other EAC risk factors are uncertain, especially for earlier birth cohorts. Cross-sectional data from the National Health and Nutrition Examination Surveys (NHANES) indicate U.S. prevalence of overweight and obese individuals increased markedly from 1980 to 1999 with a leveling by 2010 (36–38). BMI increased for males born in the 1920s, and increased for both sexes after both world wars but decreased during the Great Depression (38). H. pylori seroprevalence decreased among older adults (> age 50) between 1980 and 1999 (39) with decreases by birth cohort (40).

BE is considered a requisite step for EAC development, with BE risk increasing linearly with earlier sGERD onset-age (41). A recent meta-analysis found an OR of 4.92, [95% credibility interval (CI), 2.01–12.0] for at-least weekly sGERD in relation to long segment BE compared with no sGERD (42). Another meta-analysis also found an OR of 4.92 (95% CI, 3.92–6.22) associating sGERD with EAC (10).

It is instructive to do a “back-of-the-envelope” calculation to estimate the potential impact of changing rates of sGERD on EAC incidence. If 20% of the population has sGERD (35) and the OR is approximately 5 for EAC given sGERD compared with no sGERD (10), then sGERD should increase EAC incidence by at most 80% (assuming no sGERD occurred in the past). This simple calculation suggests that sGERD should account for at most approximately 16% of the approximately 5-fold increase in EAC incidence. EAC trends must be largely driven by other factors (OF)—obesity, eradication of H. pylori, smoking, less frequent or nonsymptomatic GERD, or perhaps unrecognized factors, for example, some studies suggest that proton pump inhibitors (PPI) may reduce neutralization of damaging bile salts, increase gastrin production, and (also with antibiotics) alter the esophageal microbiome to increase EAC risk (43–48).

Further insight into EAC incidence trends in the United States requires a better understanding of the mechanistic role of sGERD and its importance compared with OFs. In this study, multiscale models are calibrated to EAC incidence data from the Surveillance, Epidemiology, and End Results (SEER) registries between 1975 and 2009 (27). Although multiscale models were fit previously to EAC incidence trends in the United States (49, 50), this work represents the first systematic multiscale exploration of the mechanistic role of sGERD and OF as drivers of EAC incidence trends, while explicitly incorporating cohort and period trends for sGERD and OF that influence biologic processes. The multiscale approach includes a model of sGERD prevalence that depends on age, birth cohort, and period. A sGERD onset-stratified population model is used to combine risks for individuals without sGERD and for individuals acquiring sGERD during different decades of life.

Development of the multiscale model of EAC incidence and exploration of the mechanistic role of sGERD and OF proceeded in three phases. Phase I focused on identifying important biologic mechanisms that are likely driving EAC trends. Phase II focused on understanding the mechanistic role of sGERD and OF in acting through the biologic processes identified in phase I to drive EAC incidence. Both phases of model development were informed by EAC incidence data from SEER, sGERD incidence data from the UK, and U.S. sGERD prevalence data. Separate multiscale models of EAC incidence were built for all-race men and women. Phase III compared predicted trends for sGERD and OF with data on U.S. trends for obesity, smoking, and PPI use.

EAC incidence and population data for all-race men and women by single years for ages 20 to 84 years and calendar years 1975 to 2009 were downloaded from nine SEER incidence databases. EAC incidence was defined using International Classification of Diseases for Oncology, third edition (ICD-O-3) histology codes 8140–8141, 8143–8145, 8190–8231, 8260–8263, 8310, 8401, 8480–8490, 8550–8551, 8570–8574, and 8576. U.S. life tables for year 2000 were downloaded from the Centers for Disease Control website (51) to calculate age-adjusted rates for all-race males and females ages 20 to 84.

Data on sGERD incidence are from two cohort studies of the first diagnosis of weekly or more frequent GERD symptoms presenting in primary care in the UK. A study of children and adolescents (ages 1–17) identified 1,700 incident sGERD cases diagnosed between years 2000 and 2005, with a comparison group of 5,000 matched controls (34). A similar 1996 study identified 7,451 incident sGERD patients ages 2 to 79 (mostly adults) with 10,000 controls (33). Estimated sGERD prevalence during years 1990 and 2000 are based on two U.S. studies by Locke and colleagues (31) who found sGERD prevalence of approximately 20% (32). These data were used to construct a sGERD prevalence model, described below.

Maximum likelihood estimation (MLE) and Markov chain Monte Carlo (MCMC) methods were used to model sGERD prevalence (in the 1990–2000 time frame) as a function of age by assuming nonprevalent individuals become prevalent at rates based on the UK data for age-specific sGERD incidence rates among children (34) and adults (33). The model includes an sGERD remission rate by which prevalent cases become nonprevalent, with calibration to an age-adjusted target of 20% sGERD prevalence around year 2000 based on the two U.S. studies by Locke and colleagues (31, 32). Finally, the resulting sGERD prevalence was re-fit using a three-parameter change-point exponential model representing the net rate of becoming prevalent during childhood, adulthood, and a change-point time; see Supplementary Materials for details. This age-dependent sGERD prevalence model was used in all subsequent carcinogenesis models. During phase II modeling, linear or sigmoidal cohort, and period trends were estimated for sGERD prevalence while constraining the age-adjusted prevalence at approximately 20% in year 2000 in agreement with Locke and colleagues (31, 32).

MLE and MCMC methods were used to develop and compare models that represent the natural history of EAC while fitting to SEER EAC incidence data. All models include an age-dependent sGERD onset process, rates for transition to BE with or without sGERD, and a subsequent multistage carcinogenesis process that assumes any stem cells maintaining the BE tissue may acquire two initiating mutations to become premalignant, clonal expansion of premalignant cells, malignant transformations, and clonal expansions of malignant cells that may lead to cancer detection; see Supplementary Fig. S1.

The phase I model family was designed to identify biologic mechanisms that may potentially drive the observed EAC incidence trends. In these models, linear or sigmoidal trends for cohort and/or period were applied to one or more biologic processes. Thus, all individuals of a given age, period, birth cohort, and sex share the same set of biologic rates, but these rates may change with birth cohort and calendar year.

The phase II model family extended the phase I models by stratifying the population according to sGERD duration, and then evaluating the mechanistic role of sGERD and OF acting on important biologic mechanisms identified in phase I. In phase II, linear or sigmoidal trends for cohort and/or period were applied to sGERD and OF, which influence biologic rates. Individuals of a given age, period, birth cohort, and sex were stratified by decade of sGERD onset, with individuals in each stratum modeled using baseline biologic rates before acquisition of sGERD and different rates after sGERD onset.

The phase I model family introduced linear (two parameter) or sigmoidal (three parameter) trends by cohort and/or period that were applied to one or more of five biologic processes represented by the model: (i) the transition from normal to BE tissue (⁠|$\nu_{BE}$|⁠), (ii) the (geometric mean) rate of two rate-limiting mutations transforming BE stem cells to premalignant cells (⁠|$\mu_{01}$|⁠), (iii) clonal expansion of premalignant cells (⁠|$\gamma_{p}$| and |$\alpha_{p}$|⁠), (iv) malignant transformation (⁠|$\mu_{2}$|⁠), and (v) clonal growth of malignant tissue (⁠|$\gamma_{m}$| and |$\alpha_{m}$|⁠). Beginning with linear trends, MLE and MCMC methods were used to systematically compare 10 models with a single linear trend on either cohort or period applied to each of the five biologic mechanisms. This was repeated for 10 models using sigmoidal cohort or period trends while adjusting for the number of parameters. Comparisons continued with models of increasing complexity combining cohort and period trends acting on different biologic processes, stopping at models with at most six trend parameters. These results were compared with a (non-nested) model with external (multiplicative) cohort and period adjustments similar to age-period-cohort (APC) models (49).

Models identified in phase I that provided the best fits to EAC incidence in SEER were extended to include sGERD prevalence and OF modulated independently by cohort and period trends. The population was stratified by decade of sGERD onset. Using weights derived from the sGERD prevalence model, EAC hazards for each year of age and calendar year were summed over strata representing individuals with different decades of sGERD onset-age, including individuals without sGERD. Both MLE and MCMC methods were used for model selection and to infer mechanistic roles of sGERD and OF as modifiers of BE initiation, premalignant-, and malignant-promotion while estimating cohort and period effects. All models were initially evaluated using MLE methods to make model comparisons. CIs for the best fitting models were estimated through extensive MCMC runs, including 36 to 40 independent chains with each chain running for approximately 75,000 cycles.

MLE's and MCMC samples were used to estimate contributions to the observed EAC incidence trends from sGERD and OF and their interactions by cohort and period, and according to biologic mechanisms mediating the actions of sGERD and OF. The respective contributions were calculated by switching on and off different model components using the stored (MLE and MCMC) parameter sets. Backgrounds were calculated by freezing period and cohort trends after year 1975 and after the 1900 birth cohort, respectively.

We compared our predicted model trends for sGERD and OF with U.S. trends for obesity, smoking, and prescription PPI use. Obesity trends data consist of median body mass index (BMI) of white males and females at age 50 by birth cohort using results from Komlos and colleagues (38).

Smoking trends by birth cohort 1890 to 2000 were calculated by simulating 222,000 smoking histories each for men and women using the Smoking History Generator (SHG, version 6.3.2) from the Cancer Intervention and Surveillance Modeling Network (CISNET; refs. 52, 53).

PPI trends data are from a PPI Drug Use Review and Duration Analysis by the FDA (54), based on prescription claims for approximately 60 million deidentified patients (approximately 9% of the commercially insured U.S. population) for years 2002 to 2009. PPIs were introduced in the United States in 1989 (55).

A family of nested multiscale models for EAC incidence was fit to SEER incidence data using MLE. Models were compared and selected on the basis of likelihood ratio tests.

The phase I analysis identified a highly significant sigmoidal birth cohort trend affecting premalignant cell promotion as the most important mechanistic driver of the observed EAC incidence trends in SEER between 1975 and 2009. Significant additional improvements were found in models combining this cohort trend on premalignant cell promotion with either a sigmoidal cohort or period trend on BE initiation. See Supplementary Table S1 for likelihood comparison of selected models.

Phase II models compared different combinations of cohort and period trends on sGERD and OF that modify premalignant promotion and BE initiation, the two mechanisms found significant in phase I. Phase II models (stratified by sGERD onset-age decade) allowed rates for BE initiation and/or promotion to change at the sGERD onset-age and to follow the APC profile of the sGERD function. Consistent with phase I modeling, sGERD and OF were both found to significantly promote premalignant cells. Phase II models were extended to evaluate whether sGERD and OF also promote malignant cells. The final EAC incidence model was selected as the best phase II model using likelihood-based comparisons that account for the number of model parameters (see Supplementary Fig. S1).

The final model includes 18 estimated parameters, including five background biologic rates (BE initiation, initial mutation rates, premalignant promotion, malignant transformation, and malignant promotion), three response parameters for sGERD (increasing the rates for BE initiation, premalignant promotion, and malignant promotion), and 10 parameters for cohort and period effects for sGERD and OF. All cohort and period effects were modeled using sigmoidal curves representing cohort and period trends, as these provided the best likelihoods. This initially required 12 parameters (three for each sigmoid curve). However, likelihood-ratio testing indicated that the estimated sGERD and OF trends share statistically indistinguishable inflection points for both cohort and period trends, thus reducing the count of trend parameters to 10. MLEs, MCMC medians, and 95% MCMC CIs for final model parameters are shown in Supplementary Table S1.

Between years 1975 and 2009, age-adjusted incidence (age-standardized to year 2000) increased among men from 1.12 to 8.11 (per 100,000 person years) representing an increase of 626% (95% CI, 558%–696%). Among women, age-adjusted EAC incidence increased from 0.28 to 1.21, an increase of 329% (95% CI, 286%–389%; see Table 1, top).

Observed EAC incident cases in SEER were compared with model predictions for incident cases attributable to changing patterns of sGERD, OF, sGERD–OF interactions, and background representing no cohort or period trends for sGERD or OF. Observed versus expected annual EAC cases are shown by year of age and by 5-year age groups in Fig. 1 for men and Fig. 2 for women. Observed data are shown as black circles and model estimates by stacked bar graphs, including direct effects of sGERD (red), direct effects from OF (green), sGERD-OF interactions (violet), and background (black). Direct effects of OF make the largest contribution to the increase in EAC risk from 1975 to 2009, whereas sGERD and sGERD–OF interactions contribute less. sGERD-attributable EAC cases are generally delayed because of age-dependent sGERD prevalence patterns. EAC cases attributable to sGERD–OF interactions are due to an acceleration of premalignant promotion by OF followed by sGERD-associated promotion that decreases the time to EAC incidence. The decrease in number of EAC cases at older ages is due to normal population aging and death.

The parametric approach to modeling trends described here allows extrapolation of expected EAC cases into the future. An extrapolation to years 2025 to 2029 is shown in Fig. 1H for men and Fig. 2H for women. The expected rates for women remain below those of men.

Estimated contributions of sGERD and OF trends to age-adjusted increases in EAC incidence from 1975 to 2009 are shown for men and women in Table 1. Among men, direct effects of OF account for a 487% increase in age-adjusted incidence, direct effects of sGERD for an 84% increase, and sGERD–OF interactions for a 55% increase, combining for a total 626% increase in age-adjusted incidence among men. Estimates for women suggest that direct OF effects account for a 107% increase in incidence, direct effects of sGERD for a 46% increase, and sGERD–OF interactions for a 176% increase, combining for a total 329% increase in age-adjusted incidence among women. Proportionately, the impact of sGERD on EAC incidence trends is somewhat larger for women than men. The estimated total impact of sGERD, including sGERD–OF interactions, is represented by an increase in EAC incidence from 1975 to 2009 of 139% (out of 626%) for men, and 222% (out of a 329%) for women. (See Table 1 for additional details, MLEs, and MCMC medians and 95% CIs for all estimates).

Birth cohort and period trends contribute significantly to the effects of sGERD and OF in driving EAC incidence trends, with birth cohort trends most important. Among men the 626% increase in age-adjusted incidence from 1975 to 2009 includes 465% directly due to cohort, 53% directly due to period, and 108% due to cohort–period interactions. The 329% age-adjusted incidence increase among women includes 203% increase due to birth cohort, 45% due to period, and 81% due to cohort–period interactions. Table 1 includes additional details.

Model predictions for sGERD prevalence, BE prevalence, prevalence of BE given sGERD, and the relative risk of developing BE given sGERD are shown in Table 2, including comparisons between sexes by six age groups and by early period (1975–1984) versus late period (2000–2009). Depending on the age group, estimated sGERD prevalence increased by a factors of about 2 to 5 for men and 2 to 3 for women. These increases in sGERD prevalence do not translate into similar increases in BE prevalence. BE prevalence for both sexes approximately triples between the earliest (age 20–34) and the latest (age 75–84) groups, but increased generally less than 10% between early and late periods. Across age groups and periods, estimated BE prevalence among men is approximately twice that among women. Across groups, the estimated prevalence of BE given sGERD is approximately 50% higher than the prevalence of BE in the general population. Also, the estimated relative risk of developing BE given sGERD (at any age) is slightly over 3 for men, and somewhat over 4 for women.

In Table 3, BE prevalence is compared for different sGERD durations (no sGERD, >0–10 years, 10–20 years, 20–40 years, and >40 years.) during early (1975–1984) and late (2000–2009) periods for both sexes. Across age groups, BE prevalence increases approximately 2.5-fold among men and more than 3-fold for women for long-duration sGERD (over 40 years duration) compared with no sGERD. Controlling for sGERD duration, BE prevalence increases gradually with age.

EAC incidence is predicted to increase almost exponentially with sGERD duration as shown in Tables 4 and 5, in contrast with the gradual, essentially linear increase in risk for BE with sGERD duration, discussed above. These tables show relative and absolute EAC risk, respectively, by age group, sGERD duration, early versus late periods, and sex, using the same categories used in Table 3. In Table 4, the increase in estimated relative risk for EAC with sGERD duration is higher among women (who have lower baseline risk) than men. Similarly, relative risk increases with sGERD duration are higher in the earlier period (1975–1984) than later due to lower baseline risks in the earlier period. Table 5 shows absolute risks for EAC (per 100,000) with risk that increases nearly exponentially with sGERD duration. Absolute risks increase from early to late periods, and by age. Estimated risks for women with short or no sGERD duration are much lower than for men, but the relative difference decreases with increasing sGERD duration.

Supplementary Fig. S2 in Supplementary Materials shows age-adjusted EAC incidence rates for men and women for calendar years 1975 to 2009, including observed age-adjusted rates (black circles) compared with age-adjusted model predictions (stacked bar graphs) showing contributions from sGERD versus OF, cohort versus period for OF and sGERD, and contributions from sGERD-associated promotion versus sGERD-associated development of BE. OF cohort effects play a dominant role while sGERD plays a smaller part in driving EAC incidence trends. Mechanistically, both sGERD and OF act primarily through premalignant promotion as drivers of EAC trends.

In phase III modeling, the Pearson product–moment correlation coefficient (r) was used to compare the likelihood-based model predictions for cohort and period trends on sGERD and OF with data for US trends of obesity, smoking, and PPI utilization. There is a very high correlation of predicted OF cohort trends with median BMI at age 50 by birth cohort for (r = 0.988) for men and (r = 0.998) for women.

A sensitivity analysis of the maximal potential impact of obesity on EAC incidence trends was made by assuming that obesity is responsible for all of the OF cohort trends and their interactions. Under this assumption, BMI accounts for, at most, 84.7% (95% CI, 79.9%–89.6%) of the EAC incidence trend for men, and 86.1% (95% CI, 80.5%–88.4%) of the incidence trend for women.

The correlation of smoking with OF or sGERD trends was poor (r = −0.31) for men and (r = 0.23) for women. However, prescription PPI usage trends correlated highly with predicted sGERD period trends (r = 0.879) for men and (r = 0.992) for women.

A sensitivity analysis of the maximal potential impact of PPI use on EAC incidence was made assuming that PPIs are responsible for sGERD period trends and interactions. Under this assumption, PPIs could account for, at most, 18.2% (95% CI, 11.8%–26.4%) of the incidence trend for men, and 38.1% (95% CI, 27.7%–48.5%) for women.

Although sGERD contributes significantly to EAC risk, it accounts for a small fraction of the increase in SEER incidence observed between 1975 and 2009. OFs, modeled as OF, appear as the dominant driver of these increases. They interact with sGERD (at the tissue level) to increase risk beyond the additive contributions of sGERD and OF. As shown in Table 1, direct effects of sGERD are estimated to contribute approximately 13% to 14% of the observed increase in EAC incidence for men and women, which is consistent with the “back-of-the-envelope” estimates of the maximum expected effects of sGERD on EAC trends. Mechanistically, both OF and sGERD appear to act primarily through premalignant cell promotion in driving EAC trends.

Premalignant cell promotion may occur through esophageal inflammation and wounding from sGERD and OF that induces cytokine signaling and cell proliferation (56), leading to accelerated EAC development. Importantly, the effects of promotion are nonadditive—the risk from two (independent) promoters may exceed the sum of effects from each promoter acting alone. Another important consequence of promotion is that risk increases almost exponentially with duration of promotion, because a promoter increases the exponential growth rate of premalignant clones throughout the duration of exposure. The effects of increasing risk with duration of exposure are seen in Tables 4 and 5, which show nearly exponential increases in relative and absolute EAC risks, respectively, with sGERD duration. This contrasts with roughly linear increases in risk for BE with sGERD duration seen in Table 3.

sGERD promotion leads to a nearly exponential increase in EAC risk with sGERD duration that generally exceeds the nearly linear increase in risk from BE due to sGERD. The latter translates into an approximately linear increase in EAC risk. Risk for EAC is highest among individuals developing BE at an early age. Because sGERD prevalence is low at young ages, individuals who develop EAC may not have had sGERD at BE onset, although risk from early onset of sGERD (even that occurring after BE development) increases exponentially with duration due to its promoting effect.

Figures 1 and 2 show estimates of attributable EAC risks by age for effects of cohort and period acting on sGERD and OF, and their interactions, along with background rates representing expected EAC incidence in the absence of trends. These contributions are calculated as differences between the integrated EAC model hazards when different combinations of factors are turned on or off. Direct effects of sGERD and OF represent independent actions of these exposures acting on the background rate. The OF increase promotion above the background, and if sGERD occurs, it further shortens the time to cancer compared with OF alone. That is why the interaction risk rises earlier than the direct sGERD risk. The attributable interaction represents a separation of the biologic interaction into expected numbers of cancers due to different combinations of background, sGERD, and OF.

The predicted OF cohort trends correlate very highly with U.S. trends for median BMI at age 50 (r = 0.988) for men, and (r = 0.998) for women.

sGERD period trends are highly correlated with recent prescription PPI usage in the United States with predicted (r = 0.879) for men, and (r = 0.992) for women, possibly consistent with a study suggesting elevated risk for long-duration PPI usage (46). These results should be interpreted cautiously—correlation does not mean causation. Also, the estimated sGERD period trend begins to rise in the early 1980s for men, before the U.S. introduction of PPIs in 1989. The match is better for women (see Supplementary Fig. S1B and S1C).

The sensitivity of the model to estimate the effects of symptomatic reflux of frequency less than weekly or asymptotic reflux was tested by refitting the model under the assumption that reflux occurred with three times the frequency assumed for at-least weekly sGERD, and that this reflux acts on premalignant promotion, as found for sGERD. Under these assumptions, the reflux-associated attributable risk (direct plus interaction) increased by 25.8% (95% CI, 13.0%–53.6%) across males and females.

Although EAC incidence increased for both sexes between 1975 and 2009, risk for men increased more than 6-fold whereas risk increased slightly over 3-fold for women. Premalignant cell promotion emerges as the dominant biologic mechanism driving the incidence increase for both sexes, with estimated trends almost entirely due to birth cohort effects. OF appears as the primary driver of the increasing trends, especially among men, whereas both sGERD and OF contribute more evenly to the historical increase in incidence for women. In absolute terms, the increased risk in women associated with long-duration sGERD and its interactions is about one third to one half that of men, consistent with a similar biologic effect for both sexes given that rates for BE in women are about half those of men while sGERD rates are similar between sexes.

The analysis used SEER incidence data to simultaneously estimate cohort and period effects for sGERD and OF along with biologic parameters in the EAC incidence model, and thus may be less precise than if detailed historical exposure data were available. The available data did not allow calculation of separate contributions of different risk factors comprising OF. Estimates of OF reflect period and birth cohort trends for the composite effects of obesity, H. pylori, nonsymptomatic or less intense esophageal reflux, smoking, and other exposures. Thus, the estimated effects of sGERD do not capture the full impact of esophageal reflux in general—instead less-intense and nonsymptomatic reflux are modeled as contributions to OF. As discussed above, we attempted to address this issue by fitting models to SEER data while assuming rates of reflux three times that of sGERD. Extrapolations of future EAC risk are based on continuation of estimated sGERD and OF trends, and do not account for changes in population screening or gradual improvements in surveillance and treatment of BE patients over time. Finally, correlations do not imply causation, and thus in phase III modeling, the high correlations between OF cohort trends and U.S. median BMI trends, and between PPI usage and sGERD period trends should be interpreted with caution. Further studies are needed to establish a causal link between these exposures and the trends we have identified in this study

This analysis suggests that premalignant promotion is the most important biologic mechanism driving EAC incidence trends, accounting for 95.0% (95% CI, 88.4%–100.0%) of the increase among men from 1975 to 2009, and 90.1% (95% CI, 84.5%–97.3%) among women. Individuals with early onset of both BE and sGERD are at highest risk. For extended duration of sGERD (greater than 40 years), the absolute sGERD-associated EAC risk for women approaches one third to one half that of men, depending on age and calendar year, whereas the risk is 10- to 20-fold lower for women than men for individuals who never acquire sGERD.

Importantly, the dominant driver of promotion is OF. The high correlation of OF cohort trends with U.S. median BMI trends for age 50 males and females is striking. If OF trends are driven by BMI, a sensitivity analysis suggests that BMI trends may account for a large proportion of the increase in U.S. EAC incidence since 1975 (57).

Premalignant cell promotion is an important driver of carcinogenesis that causes incidence to increase exponentially with sGERD and OF exposure duration. Thus, prevention and screening should focus on long-duration exposures, including early-onset sGERD (58–60).

J.M. Inadomi reports receiving a commercial research grant from Ninepoint [provided equipment for an NIH grant (U01)] and is a consultant/advisory board member for ChemImage (Clinical Advisory Committee). J.H. Rubenstein is a consultant/advisory board member of ORC, International and Analogy Growth Partners. No potential conflicts of interest were disclosed by the other authors.

Conception and design: W.D. Hazelton, C. Hur, E.G. Luebeck

Development of methodology: W.D. Hazelton, K. Curtius, R. Meza

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Inadomi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W.D. Hazelton, K. Curtius, J.M. Inadomi, R. Meza, J.H. Rubenstein, C. Hur, E.G. Luebeck

Writing, review, and/or revision of the manuscript: W.D. Hazelton, K. Curtius, J.M. Inadomi, T.L. Vaughan, R. Meza, J.H. Rubenstein, C. Hur, E.G. Luebeck

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): W.D. Hazelton

Study supervision: W.D. Hazelton, E.G. Luebeck

The authors thank Drs. Komlos and Brabec for providing access to data on the U.S. obesity trends (38).

This research was supported by the National Cancer Institute (NCI) under grant nos. UO1 CA152926 and UO1 CA182940 (to W.D. Hazelton, J.M. Inadomi, C. Hur, and E.G. Luebeck), and by a Graduate Research Fellowship from the National Science Foundation, DGE-0718124 (to K. Curtius).

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.