Article Text

This article has a correction. Please see:

PDF

Demographic and phenotypic features of 70 families segregating Barrett’s oesophagus and oesophageal adenocarcinoma
  1. C M Drovdlic1,
  2. K A B Goddard4,
  3. A Chak3,
  4. W Brock3,
  5. L Chessler5,
  6. J F King5,
  7. J Richter6,
  8. G W Falk6,
  9. D K Johnston7,
  10. J L Fisher2,
  11. W M Grady7,
  12. S Lemeshow2,
  13. C Eng1,8
  1. 1Clinical Cancer Genetics Program, The Ohio State University, Columbus, OH, USA; James Cancer Hospital, Columbus, OH, USA; Solove Research Institute and Comprehensive Cancer Center; Division of Human Genetics, Department of Internal Medicine
  2. 2Center for Biostatistics, The Ohio State University, Columbus, OH, USA
  3. 3Division of Gastroenterology, University Hospitals of Cleveland, Cleveland, OH, USA
  4. 4Department of Epidemiology and Biostatistics, Case Western Reserve University Cleveland, OH, USA
  5. 5Department of Gastroenterology, Mercy Medical Center, Canton, OH, USA
  6. 6Department of Gastroenterology and Hepatology, Cleveland Clinic Foundation, Cleveland, OH, USA
  7. 7Division of Gastroenterology, Department of Internal Medicine, Vanderbilt University, Nashville, TN, USA
  8. 8Cancer Research UK Human Cancer Genetics Research Group, University of Cambridge, Cambridge, UK
  1. Correspondence to:
 Ms Carrie Melvin Drovdlic Clinical Cancer Genetics Program, The Ohio State University, 410 W 10th Avenue 303E Doan, Columbus, OH 43210, USA;
 drovdlic-1{at}medctr.osu.edu

Abstract

Background: Barrett’s oesophagus (BO) also termed metaplastic columnar lined oesophageal epithelium, is believed to result from a continual reparative response to chronic reflux of gastric contents. Although traditionally considered to be an acquired disorder, with several epidemiological risk factors involved, it is now recognised that there is a genetic component, and that this is likely to be autosomal dominant. Clustering of BO and of oesophageal adenocarcinoma (OAC) or oesophagogastric junctional adenocarcinoma (OGJAC) has been shown in a number of studies.

Methods: We investigated a large series of BO/OAC families to find evidence that a subset of BO/OAC is the result of a hereditary predisposition, and explored the potential of performing a linkage study with the families identified to date.

Results: Of 957 individuals in 70 families, 173 had a reported diagnosis of BO or OAC/OGJAC: 101 had BO only, 52 had OAC/OGJAC, and 20 had both BO and OAC/OGJAC. There were 133 affected males and 40 affected females, a male:female ratio of 3.3:1. In addition, 124 participants (12.9%) had a reported diagnosis of cancer other than OAC. Of these, 15 did (affected) and 109 did not (unaffected) have a diagnosis of BO or OAC. A cancer other than OAC was found in 13.9% of unaffected and 8.7% of affected individuals.

Conclusion: It is widely accepted that the majority of BO and OAC are sporadic, although familial clustering of BO and OAC has been recognised for at least three decades. Environment and lifestyle undoubtedly play a role in development of BO and OAC, and may affect the penetrance of the putative inherited factor. Although our 70 BO probands were not more likely to develop non-OAC/OGJAC cancers than normal, over a third had developed either OAC or OGJAC, and might have gone on to develop others. We intend to continue recruitment and initiate formal linkage studies to identify the causative gene(s).

  • BO, Barrett’s oesophagus
  • FBO, familial BO
  • GORD, gastro-oesophageal reflux disease
  • OGJAC, oesophagogastric junctional adenocarcinoma
  • OAC, oesophageal adenocarcinoma
  • SIR, standardised incidence ratio
  • Barrett’s oesophagus
  • hereditary cancer syndrome

Statistics from Altmetric.com

Metaplastic columnar lined oesophageal epithelium, believed to result from a continual reparative response to chronic reflux of gastric contents, is termed Barrett’s oesophagus (BO). Traditionally, BO is believed to be an acquired disorder, and several epidemiological risk factors have been identified including gender (male), race (white), obesity, and smoking.1–3 Based on reports of familial occurrence, genetic predisposition has been implicated as another risk factor. The first report by Everhart et al in 19784 described a father and two sons with the condition. This report was followed by case reports of affected siblings and extended families in which BO and/or oesophageal adenocarcinoma (OAC) clustered.5–9 Fahmy and King were the first to describe a series of four families with familial clustering.10

Epidemiological studies looking at familiality and the occurrence of BO are few, but consistently show that family members of patients with BO are more likely to have reflux symptoms than family members of patients without BO.2,11 One recent study showed that family members of patients with BO, OAC or oesophagogastric junctional adenocarcinoma (OGJAC) are more likely to have BO themselves than family members of those without these conditions.3 Shared environmental factors could play a role in the development of BO among family members, but we propose that these alone do not account for the observed familial patterns of BO. Autosomal dominant transmission of a germline genetic predisposition may account for a subset of all BO, and a substantial portion, if not all, of familial BO (FBO). Identification of a causative gene would allow for predictive testing for predisposition to development of BO/OAC, and is likely to give important clues to the molecular pathogenesis of BO and OAC, and other cancers. More significantly, as this gene will probably have a role in either gastro-oesophageal reflux disease (GORD) or metaplasia, it may also be important in normal gastrointestinal tract development and in metaplasia–dysplasia–neoplasia progression. Therefore, in order to begin to localise and eventually identify the putative gene(s) that lends susceptibility to BO and/or OAC, we have begun to accrue a series of families with early onset or familial BO and/or OAC. The purpose of this report is to describe a large series of BO/OAC families, which adds evidence to the theory that a subset of BO/OAC is the result of a hereditary predisposition. In addition, we explore the potential of performing a linkage study with the families identified to date.

METHODS

Families were recruited from the following institutions: The Ohio State University Medical Center including the James Cancer Hospital; University Hospitals of Cleveland; Mercy Medical Center Northwest; Cleveland Clinic Foundation; and Vanderbilt University Medical Center and VA Tennessee Valley Health Care Authority (institutions forming a Familial Barrett Oesophagus Consortium). Eligible families were those with at least two members affected with BO and/or OAC/OGJAC, or a single affected individual diagnosed with BO under the age of 40 years, or a single affected individual diagnosed with OAC under the age of 50 years. The proband (the first patient to come to the attention of the researchers) of each family was approached through self or physician referral. Informed consent, blood samples, release of medical records and tissue samples (when available), and permission to contact relatives were obtained under the umbrella of protocols approved by the respective institutional review boards. We contacted probands who did not live locally, and interested family members, by telephone to obtain their informed consent, and we mailed release forms for medical records and tissue. Records and tissue samples of deceased individuals were released through the consent of the next of kin. Both affected and unaffected family members were recruited. Family history information was obtained through interview with the proband, and a family pedigree was created.

Pathology and/or endoscopy reports were reviewed by a study investigator and tissue samples were reviewed by a pathologist (Wendy Frankel, MD; Joseph E Willis, MD; Kay Washington, MD, PhD; Steve Schultenover, MD) when available. For probands, the diagnosis of BO was made by the presence of intestinal metaplasia with goblet cells on biopsies obtained from the oesophagus. Relatives were considered affected by proband report and confirmed by pathology or tissue review when possible. Clinical and pedigree data were entered into a Progeny Enterprise database (Genetic Data Systems, LLC, South Bend, IN, USA) and exported into Microsoft Excel for descriptive statistics. In addition, the PedInfo program (SAGE: Statistical Analysis for Genetic Epidemiology, Release 4.0, 2001, Department of Epidemiology and Biostatistics, Rammelkamp Center for Education and Research, Case Western Reserve University, Cleveland, OH, USA) was used to provide descriptive statistics on pedigree size and structures.

Mean ages of diagnosis of reported and confirmed diagnoses of BO and OAC were calculated for males and females for whom age of diagnosis data were available. Twelve individuals with a young age of diagnosis were not included in the mean age of diagnosis calculations.

A standardised incidence ratio (SIR) was calculated to compare observed non-OAC or OGJAC invasive cancer incidence to expected non-OAC/OGJAC invasive cancer incidence among the 70 FBE probands. For each of the 70 probands, expected lifetime invasive cancer risk was calculated by adding the annual sex and age specific (18 age groups) US invasive cancer incidence rates corresponding to each year of each proband’s life, until the end of 2001. US invasive cancer incidence rates were derived from the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute.12 For years of life between 1935 and 1973, for which there are no corresponding age specific US SEER cancer incidence rates, average annual age-specific rates (based on 5 year periods) from the Connecticut Tumor Registry13 were used, and for the years 2000 and 2001, 1999 SEER cancer incidence rates were used. For years of life prior to 1935, average annual Connecticut cancer incidence rates (based on the period from 1935 to 1939) were used.

Simulations (power calculations for gene mapping) were performed using Simlink14 based on three “hypothetical” sets of individuals. Group 1 includes all pedigree members whose DNA has already been obtained to date, and non-sampled connecting individuals, corresponding to 33 pedigrees with 104 sampled subjects. Group 2 includes individuals in Group 1 plus individuals whose DNA has not yet been collected but could readily be collected because they have personally contacted one of the study sites, corresponding to 43 pedigrees with 202 sampled subjects. Group 3 includes individuals in Group 2 plus all remaining pedigree members who were alive at the last contact with the family, and assumes all these individuals can be sampled. This group includes 52 pedigrees with 314 sampled subjects.

Individuals were considered affected if they had a reported diagnosis of BO or OAC, regardless of whether the diagnosis has been confirmed. The expected lod score was computed for each pedigree, under four realistic recessive and four realistic dominant inheritance model assumptions, in order to estimate the probability of achieving a maximum lod score greater than 2.0 or 3.0 in a parametric linkage analysis. The genetic models assume a two allele locus with susceptibility allele D and normal allele d, and are described by the susceptibility allele frequency, p, and the disease penetrances fDD, fDd, and fdd. The dominant models were as follows. DOM1: p=0.01, fDD=fDd=0.2, fdd=0.008 corresponding to λS=3.6 and a prevalence of 1.2%; DOM2: p=0.001, fDD=fDd=0.45, fdd=0.0043 corresponding to λS=8.4 and a prevalence of 0.5%; DOM3: same as DOM1 for males, while for females fDD=fDd=0.1; and DOM4: same as DOM2 for males, while for females fDD=fDd=0.225. The recessive models were REC1: p=0.07, fDD=0.4, fDd=fdd=0.008, corresponding to λS=3.15 and a prevalence of 1.0%; REC2: p=0.05, fDD=0.55, fDd=fdd=0.004 corresponding to λS=8.07 and a prevalence of 0.5%; REC3: same as REC1 for males, while for females fDD=0.2; and REC4: same as REC2 for males, while for females fDD=0.275. For all models, the expected lod score was computed assuming a single marker locus with four equally frequent alleles (average PIC of 0.7). This represents typical marker informativity for microsatellite markers, although the information may actually be higher in a multipoint analysis. A recombination fraction (ϑ) of 0, 0.05, or 0.1 between the marker and the trait locus was used. Simulations were performed under the assumption that the probability that a particular pedigree is segregating the major locus is α, where α has values of 1.0 and 0.5 corresponding to the entire sample, and that approximately half the sample is linked to the major locus specified in the genetic model, respectively. Model parameters for the disease locus, including the allele frequency and penetrance parameters, were chosen to span the range of λs and the population prevalence consistent with the values computed in the Chak et al3 study, including values of λS between 3.0 and 8.0, corresponding to a population prevalence of 1% and 0.5%, respectively. As the true underlying disease model is unknown, simple models were simulated, such that the disease penetrance was not allowed to depend on age.

RESULTS

Pedigrees were drawn and data entered for 70 eligible families. Four representative families are presented in fig 1. Among these families, 957 individuals for whom at least present age, age at death, year of birth, or age of diagnosis was available were used for analyses. A total of 173 individuals had a reported diagnosis of BO or OAC/OGJAC (). Of these, 119 (68%) were confirmed by medical record, pathology report, and/or tissue review. The remaining 54 affected individuals remain unconfirmed due to unavailability of the appropriate medical or pathology records. Individuals for whom medical or pathology records refuted the reported diagnosis were not considered affected. If the refuted diagnosis made the family ineligible based on the above criteria, the family was not included in the analyses (this happened in only three instances). Of the 173 affected individuals, 101 had BO only, 52 had OAC/OGJAC, and 20 had both BO and OAC/OGJAC. In total, 133 affected individuals were male, and 40 were female, for a male:female ratio of 3.3:1.

Figure 1

Four of 70 representative familial BO/OAC families. (A) Single affected male diagnosed with OAC at 42 years of age with multiple family members with GORD; (B) proband with BO and OAC at 75 years, with maternal grandmother diagnosed with OAC at 79 years, and maternal cousin diagnosed with OGJAC at 69 years; (C) proband diagnosed with BO at 21 years, brother with BO at 40 years, father with OAC at 44 years, daughter and paternal cousin with GORD; (D) proband diagnosed with BO at 69 years, dizygotic twin brother diagnosed with BO at 67 years, sister and parents with GORD.

DNA samples were obtained from peripheral blood lymphocytes and/or paraffin embedded or fresh frozen tissue for 174 individuals. Of these, 103 had a diagnosis of BO and/or OAC/OGJAC, and 71 were unaffected. An additional 64 individuals are reasonably expected to be recruitable because they have expressed verbal or written interest in the study, or have already been sent recruitment materials but have not yet returned blood and/or released tissue samples.

Age of diagnosis data were available for 154 of the 173 affected individuals and these individuals were used to calculate average ages of diagnosis of BO and OAC among males and females (Table 1). Twelve individuals were recruited on the basis of young age at diagnosis alone (under 40 years with a diagnosis of BO, or under 50 years with a diagnosis of OAC) and these individuals were excluded from the mean age of diagnoses calculations. Among males, the average ages of diagnosis of BO and OAC were 50.6 and 57.4 years, respectively. Among females, the average ages of diagnosis of BO and OAC were 52.1 and 63.5 years, respectively. The weighted average age for BO was 51.0 years, and for OAC was 58.2 years.

Table 1

Number of affected and unaffected male and female subjects and average ages of diagnosis for those accrued irrespective of age of diagnosis

Probands were asked to report cancers other than OAC occurring in themselves or their relatives ( Table 2). In total, 124 (12.9%) participants had a reported diagnosis of cancer other than OAC. Of these, 109 were unaffected (did not have a diagnosis of BO or OAC) and 15 were affected (did have a diagnosis of BO or OAC). Further, 13.9% of unaffected and 8.7% of affected individuals had a cancer other than OAC.

Table 2

Number and percentage of affected (diagnosed with BO or OAC/OGJAC) and unaffected participants diagnosed with cancers other than OAC/OGJAC

The SIR of observed non-OAC invasive cancer incidence to expected non-OAC invasive cancer incidence among the 70 FBE probands was calculated (excluding the diagnosis of non-melanoma skin cancer). Seven probands were diagnosed with a cancer other than OAC/OGJAC or non-melanoma skin cancer. The cancers reported among these seven individuals were colon, breast, pancreatic, testicular, prostate, and tongue cancer, and leukaemia. The number of diagnoses of other invasive cancers, including gastric cancer, among FBE probands was less than that expected, based on US and Connecticut invasive cancer rates (observed=7; expected=9.9; SIR=0.71). However, using rates based on cancer incidence in Connecticut from 1935 to 1939 for the time period prior to the 1930s may have overestimated the number of expected invasive cancers, decreasing the SIR.

We performed power simulations to determine whether we would have sufficient power to localise a susceptibility gene in a genetic linkage analysis (figure 2). Simulations were performed using the identified pedigree structures and eight different autosomal dominant or recessive models that are consistent with the observed population prevalence and sibling risk ratio (λs) from the study of Chak et al.3 In this simulation, assuming α=1.0 for Group 3, if a 400 marker genome scan was performed (average intermarker distance 9 cM for a desired ϑ=0.05), a linkage study would have at least a 70% probability of identifying a unique susceptibility locus (lod score>3.0) in all of the hypothetical autosomal recessive and autosomal dominant models examined. If the genome scans were performed on Group 2, assuming α=1.0 at a desired ϑ=0.05, a linkage study would have at least a 70% probability of identifying the unique susceptibility locus (lod>3) in all the autosomal recessive models, and in autosomal dominant models DOM2 and DOM4 where λS was highest. The power to detect linkage (lod score>3) under the other five models at ϑ=0.05 ranged between 55 and 67%. When simulations were performed for Group 1, assuming α=1.0 at a desired ϑ=0.05, a linkage study would have at least a 70% probability of identifying the unique locus only under the two autosomal recessive inheritance models in which λS was highest, REC2 and REC4. The power to detect linkage (lod score>3) under the other six models at ϑ=0.05 ranged between 20% to 62%. However, when simulations were performed assuming that the sample was heterogeneous with respect to the susceptibility locus (α=0.5), the power to detect linkage (lod score>3) at ϑ=0.05 was less than 10% in all models for group 1, less than 16% in all models for group 2, and less than 36% in all models for group 3. As an example, fig 3 illustrates the power to detect linkage for model DOM3. Therefore, although we have sufficient power assuming sample homogeneity, our goal is to continue accrual of families with BO and/or OAC for the purpose of identifying gene(s) causing susceptibility to these conditions.

Figure 2

Estimated probabilities of maximum lod scores: data from representative autosomal dominant model (DOM 3) are shown. (A) Estimated power of linkage study for identifying linkage, lod>2. (B) Estimated power of linkage study for identifying linkage, lod>3.

DISCUSSION

It is widely accepted that the majority of BO and OAC are sporadic, although familial clustering of BO and OAC has been recognised for at least 3 decades. In a hospital based clinical epidemiological series, Chak et al3 showed that up to 20% of all BO presentations were familial and therefore might have a genetic component, although a causative gene has yet to be identified. Based on the reported familial patterns of BO/OAC (table 3) as well as the patterns of inheritance of the families in our series, it is likely that inheritance of BO/OAC predisposition follows an autosomal dominant model of inheritance. The function of the putative susceptibility gene(s) for BO/OAC may be broad, ranging from that of a tumour suppressor gene to that of a controller of muscle tone of the lower oesophageal sphincter. In the tumour suppressor gene model, germline mutations in the gene predispose to neoplasia, but cancer does not develop until the second allele is lost, mutated or silenced by whatever mechanism.15 This “second hit” may be influenced by environmental and other genetic factors.

Table 3

Summary of published familial BO/OAC cases

The inheritance pattern in our series of families is consistent with an autosomal dominant mode of inheritance. We examined pedigrees from published reports of familial aggregation of BO and/or OAC to gather further evidence of mode of inheritance, ages of diagnosis and gender ratio (table 3). These reports range from single sibling pairs to multiple families in which at least two members were affected with BO/OAC. The families reported include both those with BO only and those with BO and OAC. We noted the inheritance pattern with which each report was most consistent: autosomal dominant inheritance (with complete or incomplete penetrance), or autosomal recessive inheritance. All of the families reported, as well as most in our series, are consistent with autosomal dominant inheritance with incomplete penetrance. The single young affected individuals and sibling pairs in our study and the sibling pairs reported by Gelfand et al and Prior and Whorwell5,6 could also be consistent with autosomal recessive inheritance, as neither parent appeared to be affected. However, an autosomal dominant inheritance with incomplete penetrance in one of the parents is also plausible. The average ages of onset for BO and OAC ranged from 24 to 67 years, and 60 to 74 years, respectively. Interestingly, the gender ratios of affected individuals among these families range from being equal (1:1) to being completely male or completely female. The sum of gender ratios across all reported families is 4:3 male:female.

Although it would be tempting to perform a segregation analysis to help determine a genetic aetiology for familial BO, the fact that nearly all of the affected individuals in this study were identified as part of the pedigree ascertainment means that little information for a segregation analysis is available from these families once the ascertainment scheme has been taken into account. Our consortium plans to systematically and prospectively recruit consecutive persons identified with BO and OAC at participating institutions to enable a segregation analysis in the future.

The reported average ages of diagnosis of sporadic BO and OAC are estimated to be approximately 63 and 64 years, respectively.16,17 The average ages of diagnosis of BO seen in our series and in some of the previously reported families are considerably younger than these (tables table 1 and 3 ). Younger ages of onset are consistent with an inherited germline predisposition and the two hit hypothesis. Individuals born with one “hit” are one step closer to the development of precancer or cancer and are likely to develop their second “hit” earlier in life than an individual without an inherited predisposition.

Environment and lifestyle undoubtedly play a role in development of BO and OAC, and may affect the penetrance of the putative inherited factor. With strictly autosomal dominant inheritance with 100% penetrance and no gender specific environmental factors, an equal gender ratio of affected individuals would be expected. While this has been seen in some previously reported familial cases (table 3), the gender ratio of affected individuals in this family series is 3.3:1 male:female, which is similar to the gender ratios reported with BO/OAC in the general population. This may represent biased ascertainment of male probands in our series of families, dilution of our familial population with truly sporadic cases of BO/OAC, or additional genetic and environmental factors playing a larger role in the development of BO/OAC in males.

An association between Barrett’s oesophagus and colorectal cancer has been proposed. Sontag et al18 reported that of 65 patients with sporadic BO, 29 (44.6%) were also diagnosed with colonic tumours (19 benign and 10 malignant). This number was significantly higher than their control populations in which only 7.7% of symptomatic individuals seen for colonoscopy and 11.3% of symptom free patients with occult blood in the stool had been diagnosed with colonic tumours. Howden and Horhnung19 performed a systematic review of the literature and concluded that patients with BO have an increased risk of developing colon cancer with an odds ratio of 5.19 (p<0.0001), but other researchers have not found an association.20 If such an association does exist, it may be due to shared risk factors in the development of BO and colon cancer such as increasing age, excessive alcohol consumption, smoking, obesity, and dietary factors.21 However, oesophageal and colorectal adenocarcinoma also share somatic genetic changes, such as inactivating mutations in the APC and TP53 tumour suppressor genes and activating mutations in the SRC oncogene, which lead to increased expression of the cyclo-oxygenase 2 (COX-2) enzyme.

In our series of families, 13 (1.3%) were diagnosed with colon cancer. However, the SIR of cancers other than OAC or OGJAC among 70 FBE probands was 0.71, indicating that these probands were not more likely to develop non-OAC/OGJAC cancers than expected. The sample size of this analysis is relatively small, making this a crude comparison, but it suggests that other component cancers may not be a part of a hereditary BO/OAC syndrome and that the putative correlation with colon cancer may be due primarily to environmental factors. However, 26 of the 70 probands were diagnosed with OAC/OGJAC, and may not have lived long enough to develop a second cancer. In addition, the expected number of cancers for this analysis may be falsely inflated by using average annual Connecticut cancer incidence rates (based on the period from 1935 to 1939) for the time period prior to 1935, which may have decreased the resulting SIR. It is also possible that gastroenterologists perform screening colonoscopies on persons affected with BO or OAC, and that these persons consequently may be at decreased risk for colon cancer compared to the general population. Therefore, rejection of the null hypothesis may be premature, and more study in this area is warranted.

Recruitment of a series of families with familial BO/OAC is the first step in identifying the causative genetic predisposition. One large family with multiple affected individuals may be sufficient to identify a causative gene through linkage analysis, but most recruited families are small with few affected members. Our Familial Barrett’s Oesophagus Consortium has the potential to perform informative linkage analysis with the recruited families, if genetic homogeneity is assumed. However, because familial Barrett oesophagus is more likely to be a complex disease with genetic heterogeneity, our aim is to continue recruitment until nearly double the present accrual before we initiate formal linkage studies. The results of this study show that the pedigree structures we have identified have sufficient power for future success in this goal.

Following the paradigm of other well characterised hereditary cancer syndromes, identifying a gene or genes that predispose to the development of these conditions will enable cancer genetics professionals to offer genetic counselling and testing for families and individuals at risk. Asymptomatic predisposition testing is a powerful tool to use in making decisions about lifestyle and screening behaviours, and would allow prevention to be targeted to those at high risk.

Acknowledgments

This is a Familial Barrett Oesophagus Consortium study. We thank Shannon Edwards BS and Jill Griesbach BS for technical assistance. This study is partially funded by US National Institutes of Health Grants R03DK61426 (to AC, WMG, GF and CE), HG01577 (to KABG), and K24DK02800 (to AC); National Cancer Institute Grants R21CA030722 (to CE and WMG), Vanderbilt Physician Scientist Development Award (to WMG), and P30CA16058 (to The Ohio State University Comprehensive Cancer Center). Some of the results of this paper were obtained by using the program package SAGE, which is supported by a U.S. Public Health Service Resource Grant (1 P41 RR03655) from the National Center for Research Resources. CE is a Doris Duke Distinguished Clinical Scientist.

REFERENCES

View Abstract

  • .

    Publisher Correction
    Please note that the abstract shown in the Full Text, PDF and Abstract versions is incorrect and not that prepared by the authors. The correct version is provided below as a PDF (printer-friendly file)

    The error is much regretted

    Files in this Data Supplement:

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.

Linked Articles

  • Correction
    BMJ Publishing Group Ltd