Article Text

PDF

Original article
Cancer incidence and mortality risks in a large US Barrett's oesophagus cohort
  1. Michael B Cook1,
  2. Sally B Coburn1,
  3. Jameson R Lam2,
  4. Philip R Taylor1,
  5. Jennifer L Schneider2,
  6. Douglas A Corley2
  1. 1Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, Bethesda, Maryland, USA
  2. 2Division of Research, Oakland Medical Center, Kaiser Permanente, Oakland, California, USA
  1. Correspondence to Dr Michael Blaise Cook, Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH, DHHS, 9609 Medical Center Drive, Rm 7-E106, MSC 9774, Bethesda, MD 20892-9774, USA; cookmich{at}mail.nih.gov

Abstract

Objective Barrett's oesophagus (BE) increases the risk of oesophageal adenocarcinoma by 10–55 times that of the general population, but no community-based cancer-specific incidence and cause-specific mortality risk estimates exist for large cohorts in the USA.

Design Within Kaiser Permanente Northern California (KPNC), we identified patients with BE diagnosed during 1995–2012. KPNC cancer registry and mortality files were used to estimate standardised incidence ratios (SIR), standardised mortality ratios (SMR) and excess absolute risks.

Results There were 8929 patients with BE providing 50 147 person-years of follow-up. Compared with the greater KPNC population, patients with BE had increased risks of any cancer (SIR=1.40, 95% CI 1.31 to 1.49), which slightly decreased after excluding oesophageal cancer. Oesophageal adenocarcinoma risk was increased 24 times, which translated into an excess absolute risk of 24 cases per 10 000 person-years. Although oesophageal adenocarcinoma risk decreased with time since BE diagnosis, oesophageal cancer mortality did not, indicating that the true risk is stable and persistent with time. Relative risks of cardia and stomach cancers were increased, but excess absolute risks were modest. Risks of colorectal, lung and prostate cancers were unaltered. All-cause mortality was slightly increased after excluding oesophageal cancer (SMR=1.24, 95% CI 1.18 to 1.31), but time-stratified analyses indicated that this was likely attributable to diagnostic bias. Cause-specific SMRs were elevated for ischaemic heart disease (SMR=1.39, 95% CI 1.18 to 1.63), respiratory system diseases (SMR=1.51, 95% CI 1.29 to 1.75) and digestive system diseases (SMR=2.20 95% CI 1.75 to 2.75).

Conclusions Patients with BE had a persistent excess risk of oesophageal adenocarcinoma over time, although their absolute excess risks for this cancer, any cancer and overall mortality were modest.

  • BARRETT'S METAPLASIA
  • CANCER EPIDEMIOLOGY
View Full Text

Statistics from Altmetric.com

Video abstract

Significance of this study

What is already known on this subject?

  • Barrett's oesophagus increases the risk of oesophageal adenocarcinoma by 10–55 times that of the general population

  • There is conflicting evidence whether Barrett's oesophagus increases risks of colorectal cancer or of cardiovascular disease.

  • To date, there has not been a comprehensive study of cancer incidence and mortality risks in a US Barrett's oesophagus cohort, yet population-specific risks are essential for evidence-based medicine.

What are the new findings?

  • Barrett's oesophagus conferred an excess absolute risk for oesophageal adenocarcinoma of 24 cases per 10 000 person-years, and this risk persisted over time after Barrett's oesophagus diagnosis.

  • Patients with Barrett's oesophagus had increased risks of cardia and stomach cancers, but excess absolute risks were modest. Risks of colorectal, lung and prostate cancers were unaltered.

  • Patients with Barrett's oesophagus had a consistent 51% increased risk of respiratory system death; in contrast, apparently elevated mortality risks of all-causes, ischaemic heart disease and digestive system disease appeared to be the result of diagnostic bias, and all excess absolute risks were modest.

How might it impact on clinical practice in the foreseeable future?

  • Attaining population-specific risk estimates are essential for developing comprehensive disease-prevention interventions, scientific understanding, risk communication and to provide an evidence base to guide clinical decision-making. Excess absolute risks of cancer and mortality among patients with Barrett's oesophagus were modest. These results may inform strategies for surveillance endoscopy, additional screening colonoscopy and prophylactic interventions in Barrett's oesophagus populations within the USA.

Introduction

Barrett's oesophagus (BE) markedly increases the relative risk for oesophageal adenocarcinoma, with estimates ranging from 10-fold to 55-fold that of the general population,1–7 which translates to an absolute annual risk of approximately 0.5%, or 1/200 person-years.8 ,9 However, what is less well understood are the absolute and relative risks of extra-oesophageal malignancies as well as cause-specific death in individuals diagnosed with this precancerous lesion. Attaining population-specific estimates of such are essential for developing comprehensive disease-prevention interventions, scientific understanding, risk communication and to provide an evidence base to guide clinical decision-making.

Previous BE cohorts for which site-specific cancer risks have been assessed have been based in the UK7 ,10 and the Netherlands,11 while assessment of underlying causes of death have been conducted for cohorts in the UK,7 ,12–15 Northern Ireland16 and the Netherlands.11 These studies reported variable risks for incidences of site-specific cancers as well as underlying causes of death relative to the general population, some of which may be attributable to population (geographic) heterogeneity of disease and associated risk factors. For example, there have been disparate results regarding whether colorectal cancer risk or cardiovascular disease is increased in individuals with BE. If these relationships are substantiated by larger, population-specific studies, then this may provide an evidence base for additional clinical interventions. Herein, we present the first comprehensive study of cancer incidence and mortality risks in a large US BE cohort.

Methods

Study population

Kaiser Permanente Northern California (KPNC) is an integrated health services delivery organisation. KPNC includes, in any given year, >3.3 million members (approximately one-third of the insured population in the region). Research within this setting encompasses 20 medical centres in urban, suburban and semirural regions across a large geographic area, plus additional free-standing offices and endoscopy units. The membership demographics closely approximate the diverse underlying census population of Northern California, including members with Medicare, Medicaid (low-income) and commercial insurance; thus, studies within this setting provide results that can be generalised to a large region and general populations.17 ,18

Patients with BE diagnosed at KPNC at ages 18 years and older during 1995 through 2012 were eligible for selection into the cohort. Codes for a diagnosis of BE include the International Classification of Disease 9th revision (ICD-9) code 530.2 ‘Barrett's ulcer’, which on KPNC endoscopy reporting sheets was specified as ‘Barrett's esophagitis’; ICD-9 code 530.85 ‘Barrett's esophagus’, which was introduced in 2003 and replaced 530.2 and the Systematised Nomenclature of Medicine (SNOMED) code M73330 (BE). SNOMED codes are used by the pathology departments for assigning specific histological diagnoses to biopsy specimens. For this study, BE was defined as: at least one instance of the BE-specific ICD-9 code 530.85 or; at least one instance of ICD-9 code 530.2 and one instance of SNOMED code M73330, the latter combination of which we have shown to accurately classify patients with BE in a smaller cohort, which overlaps with the current study cohort.19 We used the first instance of any of these codes as the index diagnosis date of BE for all analyses. A sensitivity analysis was conducted in which we excluded patients with BE who were diagnosed with oesophageal adenocarcinoma within 6 months of their BE diagnosis date and patients with BE who did not have 6 months of continual follow-up immediately following their BE diagnosis date. We also conducted sensitivity analyses using a less specific BE definition of ICD-9 codes 530.85 or 530.2, without any requirement for SNOMED code M73330. Patients with BE were ineligible for analysis if they had any cancer diagnosis (excluding skin cancer) prior to their BE diagnosis date; had no diagnosis date associated with a cancer diagnosis; had no enrolment information or had unknown sex. The project and analyses were approved by the KPNC Institutional Review Board.

Variables for analysis

For each patient with BE, we extracted from electronic databases all instances and dates of ICD-9 codes 530.85 and 530.2, and SNOMED code M73330; date of birth; sex; membership gaps; cancer diagnoses and mortality data. Pathology data were from a comprehensive clinical source that includes all biopsies and their assigned SNOMED histological diagnoses. Categories of cancer and death outcomes were identified a priori based on prior literature and are shown in the results tables (Tables 14). Data tables of rates for the KPNC population were based on these outcome categories to provide the expected number of outcomes in the underlying population. Cancer diagnoses (date, ICD-O-3 codes for site and histology) were extracted from the KPNC cancer registry, which reports to the Surveillance, Epidemiology and End Results (SEER) registry. Causes of death (ICD-9/ICD-10)—including the underlying cause—and date of death were extracted from KPNC mortality files, which use standard probabilistic matching algorithms to state death certificate data and federal death files.

Statistical analysis

We compared the cancer incidence and mortality experience of the BE cohort with expected rates from the greater KPNC population using indirect standardisation (comparison with an external standard) with stratification by sex, age (0–4, 5–9, 10–14,…, 95–99, 100+ years) and calendar year (single years). For cancer incidence (standardised incidence ratio (SIR)) analyses, follow-up started at BE diagnosis and ceased at the earliest of time of event, death, last disenrollment or study end (31 December 2013). Follow-up time was restricted to enrolled periods (ie, membership gaps were excluded) because cancer incidences—for the BE cohort and the total KPNC population—were only captured during enrolment. Individuals were not right-censored at a cancer diagnosis when that cancer diagnosis was not the outcome of interest. Follow-up time for mortality (standardised mortality ratio (SMR)) analyses was similarly constructed with the only difference being that 1 year was added to the last disenrollment date, given that a large number of deaths coincided with this period; KPNC population death tables were similarly constructed. We estimated relative risks, in the form of SIR and SMR, and their respective 95% CIs. We also estimated risks stratified by time since BE diagnosis using the cut-points of 3 and 6 years, which were the approximate 33rd and 66th percentiles of follow-up time for the cohort. For the main analysis, we estimated excess absolute risks as the excess number of cancers per 10 000 BE person-years (observed–expected/person-years×10 000). Lastly, we used fixed effects meta-analytic methods to combine relative risks of prior studies with those from our KPNC study reported herein. All analyses were conducted using Stata software V.13.1 (StataCorp, College Station, Texas, USA).

Results

Primary analysis

There were 8929 eligible patients with BE for analysis, including 5854 (66%) males, resulting in a male-to-female sex ratio of 1.9:1. The mean age at BE diagnosis was 61.1 years (SD=13.2) and there were a total of 50 147 years of follow-up providing for an average of 5.6 years (SD=4.1) per patient with BE.

Cancer

Individuals diagnosed with BE had a 40% increased risk of any malignant neoplasm excluding skin cancer, compared with the total KPNC population (table 1 and figure 1, SIR=1.40, 95% CI 1.31 to 1.49). This increased risk was attenuated to 23%, but remained statistically significant, after exclusion of oesophageal cancers (SIR=1.23, 95% CI 1.14 to 1.31). The excess absolute risk of any malignant neoplasm excluding skin cancer was 57 cancers per 10 000 BE person-years, but reduced to 32 cancers when oesophageal cancers were excluded (figure 2). Importantly, time period-specific risks for any malignant neoplasm were elevated only for the first time period assessed (0 to <3 years) post-BE diagnosis and not thereafter (table 2).

Table 1

Standardised incidence ratios and excess absolute risks per 10 000 person-years for the Barrett’s oesophagus cohort*, overall and by sex

Table 2

Standardised incidence ratios for the Barrett’s oesophagus cohort*, stratified by follow-up time post-Barrett’s oesophagus diagnosis

Figure 1

Forest plot summarising the standardised incidence ratios of cancer risk in the Barrett's oesophagus cohort relative to the greater KPNC population. The filled circles represent the risk estimates on a linear scale, while the horizontal lines demarcate the extent of the 95% CIs. The vertical line demarcates the null hypothesis. KPNC, Kaiser Permanente Northern California; SIR, standardised incidence ratio.

Figure 2

Forest plot summarising the excess absolute cancer risks per 10 000 person-years for the Barrett's oesophagus cohort relative to the greater KPNC population. The filled circles represent the risk estimates on a linear scale, while the horizontal lines demarcate the extent of the 95% CIs. The vertical line demarcates the null hypothesis. EAR, excess absolute risk; KPNC, Kaiser Permanente Northern California.

There were 121 oesophageal adenocarcinomas diagnosed in this BE cohort resulting in a 24-times increased risk relative to the greater KPNC population and an excess absolute risk of 24 cancers per 10 000 person-years. The crude incidence rate of oesophageal adenocarcinoma in this BE cohort was 2.5 per 1000 person-years (95% CI 2.1 to 3.0), which translates to a crude absolute annual risk of 0.25% (95% CI 0.21% to 0.30%). Oesophageal cancer overall (including squamous cell carcinoma and other oesophageal malignancies) had a slightly lower relative risk (SIR) of 16 compared with the total KPNC population, which decreased further when assessed as a joint outcome of either all oesophageal plus all stomach cancers (including cardia) (SIR=8.94) or all oesophageal cancers plus cardia cancers (SIR=14.34). Cardia cancer alone had an SIR of 10.48, and any stomach cancer (including cardia) an SIR of 4.36. Non-cardia gastric cancer risk was not altered in this BE cohort when we assessed this outcome as malignant neoplasms of the stomach (C160–C169) excluding cardia (C160) and overlapping sites (C168) (SIR=1.49, 95% CI 0.77 to 2.60) or when we additionally excluded unspecified (C169) malignancies (SIR=1.10, 95% CI 0.44 to 2.27) (results not tabulated). Importantly, risks for all categories of upper GI malignancies decreased with time since BE diagnosis (table 2); for example, the SIRs for oesophageal adenocarcinoma were 31.66 for 0 to <3 years, 21.94 for 3 to <6 years and 16.62 for ≥6 years.

Risk of colon, rectosigmoid and anal cancers; trachea, bronchus and lung cancers and prostate cancers were not altered in this BE cohort compared with the greater KPNC population (table 1 and figure 1), although SIRs for each of first two grouping were elevated only in the first period (0 to <3 years) of time following BE diagnosis (table 2).

Cancer risks were comparable between males and females with BE, with the exception that the relative risk for oesophageal adenocarcinoma was much higher for female patients with BE (SIR=59.61) compared with male patients with BE (SIR=21.46), a difference mainly driven by the much lower expected numbers in females relative to males (see online supplementary figure S1). The excess absolute number of oesophageal adenocarcinomas per 10 000 person-years attributable to BE was much lower in females relative to males (estimated absolute risk 12 vs 30, respectively) (table 1 and see online supplementary figure S2).

supplementary tables and figures

Mortality

Using the SEER Cause of Death Recode classification system (http://seer.cancer.gov/codrecode/1969+_d04162012/index.html), the leading causes of death for patients with BE were diseases of the heart (n=348; 24%), other (includes numerous non-malignant conditions: n=205; 14%), digestive system cancer (n=186; 13%), chronic obstructive pulmonary disease and allied conditions (n=98; 7%), cerebrovascular disease (n=71; 5%) and respiratory system cancer (n=68; 5%).

Patients with BE had an increased overall risk of death of 31% compared with the total KPNC population (SMR=1.31; table 3 and figure 3), which translated to 69 excess deaths per 10 000 person-years. A majority of this excess absolute risk remained for any cause of death after excluding oesophageal cancer deaths (table 3 and figure 4). However, these excess risks decreased with time since BE diagnosis and were no longer evident after 6 years when oesophageal malignancy was excluded (table 4).

Table 3

Standardised mortality ratios and excess absolute risks per 10 000 person-years for the Barrett’s oesophagus cohort*, overall and by sex

Table 4

Standardised mortality ratios for the Barrett’s oesophagus cohort*, stratified by follow-up time post-Barrett’s oesophagus diagnosis

Figure 3

Forest plot summarising the standardised mortality ratios in the Barrett's oesophagus cohort relative to the greater KPNC population. The filled circles represent the risk estimates on a linear scale, while the horizontal lines demarcate the extent of the 95% CIs. The vertical line demarcates the null hypothesis. ; KPNC, Kaiser Permanente Northern California; SMR, standardised mortality ratio.

Figure 4

Forest plot summarising the excess absolute mortality risks per 10 000 person-years for the Barrett's oesophagus cohort relative to the greater KPNC population. The filled circles represent the risk estimates on a linear scale, while the horizontal lines demarcate the extent of the 95% CIs. The vertical line demarcates the null hypothesis. EAR, excess absolute risk; KPNC, Kaiser Permanente Northern California.

Oesophageal cancer had the highest relative mortality risk with an SMR of over 10 for this BE cohort and an excess absolute risk of 15 deaths per 10 000 person-years. Risk of oesophageal cancer death did not vary by time since diagnosis of BE. Mortality risks of trachea, bronchus and lung and colon, rectosigmoid junction and anus were not altered in this BE cohort overall or by time since BE diagnosis.

For non-neoplastic mortality, risk of death from diseases of the circulatory system was slightly increased (SMR=1.24, excess absolute risk=17 per 10 000 person-years), a result driven mainly by an increased risk of death from ischaemic heart disease (SMR=1.39, excess absolute risk=9). Risks of death from diseases of the respiratory system (SMR=1.51, excess absolute risk=12) and diseases of the digestive system (SMR=2.20, excess absolute risk=9) were also increased, relative to the greater KPNC population. When stratified by time since BE diagnosis, all of these risks decreased with time, and only death from diseases of the respiratory system remained significantly elevated in the last time period (≥6 years) assessed.

Risks by sex were largely comparable with a few exceptions. Similar to the analysis of oesophageal adenocarcinoma incidence, the relative risk of any oesophageal cancer death in females (SMR=19.90) was more than twice that of males (SMR=9.09), yet excess absolute risks were much higher in males (excess absolute risk=18 vs 10 per 10 000 person-years; see online supplementary figures S1 and S2). Lastly, both relative (SMRs) and excess absolute (excess absolute risks) risks of death from diseases of the respiratory and digestive systems were slightly higher in females compared with males.

Sensitivity analyses

The cohort size was reduced by 308–8621 individuals when we excluded patients with BE who were diagnosed with oesophageal adenocarcinoma within 6 months of their BE diagnosis date (as well as subjects who did not have at least 6 months of continual follow-up). SIRs and SMRs from these sensitivity analyses were largely unaltered, with only slight attenuation in risk of oesophageal adenocarcinoma incidence (SIR=18.44, excess absolute risk=18 per 10 000 person-years) and oesophageal cancer mortality (SMR=7.98, excess absolute risk=12 per 10 000 person-years), as well as categories to which these end points also contributed (see online supplementary tables S1 and S2).

Use of a more liberal definition of BE that required an ICD-9 code of either 530.85 or 530.2 expanded the cohort for analysis to 11 987 patients with BE and total follow-up time to 71 122 years. SIRs of this sensitivity analysis did not markedly differ from the main analysis, with slight attenuation in risk for outcomes that included oesophageal cancer (see online supplementary table S3), which were further reduced when the 6-month ‘prevalent’ oesophageal adenocarcinoma exclusion was applied (see online supplementary table S4). Conversely, SMRs for all causes were slightly increased when the more liberal definition of BE was used (SMR=1.70, excess absolute risk=170 per 10 000 person-years; see online supplementary table S5). This increase appeared to be driven by an increase in SMRs for non-neoplastic mortality. These mortality risk increases were slightly attenuated when the 6-month exclusion was applied (see online supplementary table S6).

Discussion

This study of a large community-based US cohort found that patients with BE have modest excess absolute increased risks for malignancies of the oesophagus, cardia and stomach and similar risks for malignancies of the colon, rectosigmoid junction and anus compared with the KPNC population. Overall mortality risk and, specifically, risks of death from oesophageal cancer, diseases of the circulatory system—particularly ischaemic heart disease—diseases of the respiratory system and diseases of the digestive system were all increased in this BE cohort. The relative and absolute risks from this study provide important evidence that will help inform ongoing debates about the value of endoscopic surveillance and preventative strategies (including ablation) in patients with BE.

The crude absolute annual risk of oesophageal adenocarcinoma in BE has been estimated from meta-analyses to be 0.61% overall, 0.53% when excluding early incident cancers, 0.41% when additionally excluding baseline high-grade dysplasia8 and 0.33% when additionally excluding baseline low-grade dysplasia.20 Estimates from our large US BE cohort were slightly lower (0.25%) and further reduced when we excluded early (likely prevalent) cancer diagnoses (0.20%), despite our inability to exclude baseline dysplasia. Our lower estimated risk may be partly due to the community-based cohort we studied here as opposed to surveillance cohorts, which may have upwardly biased the previously cited meta-analytic estimates.21 In addition, crude estimates of the risk of oesophageal adenocarcinoma in BE will vary due to study differences in age, sex and possibly calendar year, which emphasises the importance of our presentation of relative risks (SIR=23.9) and excess absolute risks (24 per 10 000 person-years), which account for these factors. It is noteworthy that our SIR is in the range of 10–55, which is derived from studies that have adjusted for age, sex and calendar period in estimating the relative risk of oesophageal adenocarcinoma in BE.1–7 In addition to age, sex and calendar year, cancer risk may also vary due to length and distribution of follow-up time,21 the definition of BE employed (eg, whether to include patients without intestinal metaplasia), dissimilar referral practices for endoscopy, institutional preferences for certain biopsy protocols as well as variations in population genetics and environmental exposures, which places a premium on population-specific risk estimates. In sum, the evidence from our study indicates that the risk of progression to oesophageal adenocarcinoma in the US patients with BE is modest, supporting the supposition that current endoscopic surveillance programmes that include all patients with BE, including those without initial dysplasia, may not be cost-effective.22

It is noteworthy that risk of oesophageal adenocarcinoma decreased with time since BE diagnosis, but the risk of oesophageal cancer death did not vary over time. Taken together, these findings suggest that the variable risk of oesophageal adenocarcinoma incidence observed over time is likely an artefact attributable to prevalent disease (as supported by our sensitivity analyses), increased lead time from ‘front-loading’ of events and overdiagnosis, in combination with progressively widening surveillance intervals. Thus, contrary to studies which accrued oesophageal cancers over a short period following diagnosis of BE, our oesophageal cancer mortality results suggest that oesophageal cancer risk persists for >6 years, with little decay in risk over time. This finding is consistent with a recent rigorous modelling study that suggested that BE cohorts with limited follow-up may underestimate long-term risk of oesophageal adenocarcinoma.21

Although BE appeared to increase the risk of oesophageal adenocarcinoma to a greater extent in females relative to males, the excess absolute risk and crude incidence rate were approximately 2.5 times higher in males than females. This sex difference is in line with estimates from prior studies23 ,24 and underscores the value of sex in predictive models of oesophageal adenocarcinoma in the setting of BE.25 Non-mutually exclusive hypotheses that may explain why male patients with BE have a higher absolute cancer risk than female patients with BE include differences in age at onset of disease, android obesity, production and concentration of gastric acid, hiatal hernia, defective lower oesophageal sphincter, intragastric pressure and sex steroid hormones.23 ,26

Debate as to whether individuals with BE have an increased risk for colorectal cancer has been complicated by giving equal, or even greater, attention to cross-sectional (prevalence) studies as compared with the more methodologically rigorous and appropriate prospective studies of BE cohorts. The latter should be the focus when considering an intervention (eg, screening colonoscopy) in individuals diagnosed with BE. Four such studies have been published previously7 ,10 ,27 ,28 which, when combined via fixed effects meta-analysis with this study herein, provide some evidence of an increased risk of colorectal/anal cancers in BE (relative risk=1.56, 95% CI 1.46 to 1.67, I2=80%). The large heterogeneity accompanying these summary risk estimates necessitates a cautious interpretation and may further endorse the importance of population-specific risks. Also of importance is evidence indicating that these increased risks may be attributable to diagnostic bias.27 ,28 Even if the higher upper CI (1.67) of the meta-analytic relative risk estimate represents the true association in the USA, when applied to the KPNC BE cohort this would only result in 11 additional colorectal cancers being diagnosed per 10 000 person-years. Since the magnitude of this effect is comparable to, or smaller than, that of other risk factors that do not modify screening guidelines (eg, increasing age, higher body mass index (BMI)), even if present, this small magnitude of excess absolute risk is unlikely to support additional interventions for colorectal cancer in individuals diagnosed with BE.

The overall risk of death, even after excluding oesophageal cancer, was slightly higher in this BE cohort relative to the greater KPNC population, an observation supported by some studies3 ,5 ,12 ,15 but not by others.11 ,13 ,16 ,29 ,30 The decrease in all-cause mortality with increased time since BE diagnosis suggests that prior studies, which showed increased risk, may have had diagnostic bias whereby a condition increased the risk of both diagnosing BE and other causes of mortality. This could occur, for example, if BE was coincidentally diagnosed during endoscopic evaluations for GI bleeding around the time of treatment for cardiac ischaemia or for iron deficiency anaemia related to unrelated extra-oesophageal malignancies. Primary contributors to the increased mortality risk in the KPNC BE cohort were oesophageal adenocarcinoma and non-neoplastic mortality. Of the latter, diseases of the circulatory system were increased (SMR=1.24, excess absolute risk=17 per 10 000 person-years), primarily due to ischaemic heart disease (SMR=1.39, excess absolute risk=9), a finding supported by a prior UK-based study.14 Some prior BE cohort studies have assessed ischaemic heart disease combined with acute myocardial infarction, thus to meta-analyse these relative risk estimates we assessed this combined, singular end point. The four prior studies that reported such relative risks7 ,11 ,15 ,16 combined with the KPNC data herein, suggest a modest but statistically significant increased risk (relative risk=1.14, 95% CI 1.04 to 1.24, I2=32%). Even if the higher KPNC risk estimate of ischaemic heart disease alone (1.39) is real and accurate for the US BE populations, the modest excess absolute risk of 9 per 10 000 person-years is unlikely to warrant additional interventions beyond what would be offered to the general population.

Increased risk of death from respiratory system diseases in patients with BE has support from one UK study15 but not from three other European studies.7 ,11 ,16 The combined estimate of these four studies with KPNC data indicates a modest increased risk (relative risk=1.23, 95% CI 1.10 to 1.35, I2=80%), although the large heterogeneity suggests that population-specific risk may be more appropriate than estimates across multiple populations. Point estimates from the KPNC data indicate that patients with BE have a 51% increase in their relative risk of respiratory disease death and an excess absolute risk of 12 deaths per 10 000 person-years.

Lastly, there was an increased risk of digestive disease death in the KPNC BE cohort, an observation which we7 and others15 ,16 have made previously in BE cohorts based in the UK and Northern Ireland, but which was not observed in a BE cohort in the Netherlands.11 When all estimates were combined with KPNC, the summary estimate was indicative of an association (relative risk=1.97, 95% CI 1.66 to 2.28, I2=58%) but, as we have previously suggested,7 this increased risk may be a function of selection bias with the diagnosis of BE being incidental after an endoscopic referral pertaining to symptoms of their eventual cause of death. Indeed, 41 of these 79 deaths were ascribed to the SEER grouping of ‘chronic liver disease and cirrhosis’, 5 to ‘stomach and duodenal ulcers’ and the remainder to ‘other’, a miscellaneous group of digestive diseases. Also in support of our hypothesis is the fact that risk of digestive disease death decreased to the null with increased time since BE diagnosis.

Strengths of this study include the fact that this is the first prospective analysis of extra-oesophageal cancer risks and mortality risks in a large community-based US BE cohort. The study used a specific definition of BE which incorporated validation work from a large prior study that included endoscopy and pathology review. Comparison rates employed here came from the total KPNC population and thus are representative of the underlying population from which our BE cases arose. Indirect standardisation was based on age, sex and calendar year, thus accounting for these variables which are strongly associated with risk of the outcomes assessed. This study was limited by its lack of information on dysplasia, segment length and other risk covariates such as BMI, tobacco smoking and alcohol consumption. The lack of dysplasia data is unlikely to bias these results for two reasons. First, relatively few patients have high-grade dysplasia at diagnosis and their effect, if excluded, would be relatively small. Second, low-grade dysplasia, although more common than high-grade dysplasia, is still uncommon and is a poor predictor of cancer risk. In large US multicentre studies, a diagnosis of low-grade dysplasia conferred a similar cancer risk to non-dysplastic BE.31 The study was also limited by its inability to generate race-specific risk estimates because race is not routinely collected in the electronic KPNC data files, which precluded generation of race-specific KPNC population rates for comparison with the BE population.

Understanding the experience of a population with a precancerous lesion is an essential component of understanding the scientific importance of the ailment, communicating with affected individuals on risks and their magnitude and being able to make informed clinical decisions. Excess absolute risks of cancer and mortality conferred by BE are modest, including oesophageal cancer—for which the absolute risk is low and persistent—and all-cause mortality, where the modest elevation we observed may be attributable to diagnostic bias. These results may inform the potential impact of surveillance endoscopy, additional screening colonoscopy and prophylactic interventions in BE populations within the USA.

Acknowledgments

We would like to thank Natalia Udlatsova and Wei Zhao (Kaiser Permanente, Division of Research) for their assistance with data extraction.

References

View Abstract

Footnotes

  • Contributors Study concept and design (MBC and DAC); acquisition of data (DAC, JL and JS); statistical analysis (MBC and SBC); interpretation of results (all); drafting of the manuscript (all); critical revision of the manuscript (all); obtained funding (MBC and PRT); study supervision (MBC and DAC).

  • Funding This study was supported entirely by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA. No funding or other financial support was received.

  • Competing interests None declared.

  • Ethics approval National Cancer Institute Institutional Review Board.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.