Study design and population
This 10-year follow-up study was conducted at the Aarhus University Hospital in Aarhus County, Denmark. The county has a population of approximately 650,000 (12% of the Danish population). The Aarhus University Hospital has the county's largest departments of gastroenterology and surgery, and most upper endoscopies in the county are performed there. We obtained data on all patients who underwent upper endoscopy at the hospital between January 1, 1992 and December 31, 1993, with a 10-year follow-up period extending until December 31, 2003.
Linkage between registries
Records in all registries used in this study contain the Danish Civil Registration System's unique 10-digit civil registration number, which is assigned to all Danish citizens at birth [17]. Use of the civil registration number allows valid linkage between registries.
UCEP patients
The Aarhus University Hospital Endoscopy Registry contains both paper files and electronic medical records for all patients who underwent upper endoscopy since 1976. Since 1977 the electronic record has been maintained by the Hospital Administrative Patient Registry. Each record includes information on the patient's civil registration number, dates of admission and discharge, date and type of procedures performed, and diagnoses coded by physicians according to the International Classification of Diseases (ICD). ICD-8 codes were used in 1977–1993, and ICD-10 codes thereafter (ICD-9 was never used in Denmark) [18]. Hard-copy medical records consist of referral notes (nearly 90% of the patients are referred from general practitioners as outpatients) and endoscopy records written by the physicians who performed the procedures. The latter include information on presenting symptoms (indication for the procedure), diagnoses made during the endoscopy, biopsies taken, and description of subsequent pathological findings. This information is both standardized by means of a checklist and described in free text. General practitioners' referral notes are not standardized and mainly include information on the patients' history and symptoms. We ascertained patients' symptoms both from the endoscopy records and from the referral notes, and for the majority of patients the descriptions of the presenting symptoms from the two sources were in agreement. One of the study physicians (EMM) coded and entered data from the hard-copy medical records into an electronic research database.
The physicians who performed the upper endoscopy did not take part in the evaluation of the data of the study, selection of patients for the study, study analyses, or interpretation of the results in the study.
During the recruitment period we identified 1,799 patients with a first-time normal upper endoscopy. These patients were classified into four groups according to their symptoms: (a) only chest/epigastric pain, (b) reflux-like symptoms (e.g., heartburn and/or acid reflux), (c) neither chest/epigastric pain nor reflux-like symptoms, and (d) both chest/epigastric pain and reflux-like symptoms. The subcohort of interest was comprised of the first group: 410 (23%) patients with only chest/epigastric pain and a first-time normal upper endoscopy. Thus, we excluded patients with symptoms such as specified/unspecified dyspepsia, heartburn and/or acid reflux or with other symptoms listed in the endoscopy record and in the referral note. Our study subcohort of patients with only chest/epigastric pain and a first-time normal upper endoscopy was defined and chosen a priori [7, 8].
Through linkage to the nationwide Danish Hospital Discharge Registry (HDR), we identified patients with discharge diagnoses of ischemic heart disease (IHD) (myocardial infarction, angina, and/or heart failure [19]) prior to the date of upper endoscopy, coded according to the ICD diagnoses in Appendix 1. The HDR, established in 1977, electronically tracks all non-psychiatric hospitalizations throughout Denmark, including dates of admission and discharge, procedures performed, and up to 20 discharge diagnoses coded by medical doctors at the time of discharge. Data also on out-patients were included from 1995. We excluded 24 patients with a discharge diagnosis of IHD prior to the date of upper endoscopy. The remaining 386 patients comprised the study cohort of UCEP patients, who may resemble patients with upper FGIDs [20, 21]. The study cohort was identical to the study cohort used in two recently published studies on other prognostic outcomes among UCEP patients [7, 8].
Population controls
For each UCEP patient, controls residing in Aarhus County were identified from the Civil Registration System and matched by age and gender (N = 4,100). The controls were selected on the date of the corresponding patient's first-time normal upper endoscopy (the index date). Ten controls per UCEP patient were randomly chosen to achieve statistical precision [22]. On the basis of information from HDR, 67 controls with discharge diagnoses of IHD prior to the index date were excluded. The remaining 3,793 controls were included in the analyses [7, 8].
Risk of IHD and mortality
Data on hospitalizations for IHD (defined as a discharge diagnosis of myocardial infarction, angina and/or heart failure) during the 10 years of follow up were obtained from the HDR. Mortality was ascertained from the Civil Registration System, which tracks Danish citizens' births, deaths, and migrations. In addition, death certificates, available from the Danish Causes of Deaths Registry, provided information on cause-specific mortality up to December 31, 2003. Since 1970, death certificates have included information on cause and manner of death (natural death, accident, suicide, or unknown) for 100% of deceased Danish residents. Because few patients and controls died of unnatural causes, we did not consider manner of death in our analyses.
In our cohort of UCEP patients, we focused on the seven most common causes of deaths occurring after the date of normal upper endoscopy: IHD, pneumonia, stroke, arteriosclerosis (in the absence of IHD or stroke), lung cancer, alcohol dependence, and chronic obstructive pulmonary disease.
Confounding factors
HDR data were used to compute a comorbidity index score – the Charlson Index [23, 24] – for each UCEP patient and control [7, 8]. The Charlson Index, covering 19 major disease categories weighted according to their prognostic impact on patient survival, has been adapted for use with hospital discharge registry data. We computed the Index based on diagnoses recorded during all previous hospitalizations since 1977. We used discharge diagnoses of alcohol- and smoking-related diseases as proxies for alcohol abuse and tobacco smoking (ICD codes provided in Appendix 1) [7, 8]. Alcohol- and smoking-related diagnoses were excluded from the Index to reduce the risk of residual confounding from these diseases. Three index levels were defined to capture increasing degrees of comorbidity: no comorbidity (Charlson Index 0), comorbidity level 1 (Charlson Index 1–2), and comorbidity level 2 (Charlson Index > 2) [7, 8].
Statistical analysis
Demographic and clinical variables such as gender, age, presence of alcohol- and smoking-related diseases, level of comorbidity, subsequent discharge diagnosis of IHD, overall mortality, and cause-specific mortality were presented as proportions or means, as appropriate.
Follow up began on the date of normal upper endoscopy or the corresponding index date for controls, and ended on the date of initial diagnosis of IHD, the date of death, the date of emigration, or at the end of the study period on December 31, 2003, whichever came first.
We constructed Kaplan-Meier survival curves and used life table techniques to estimate the risk of hospitalization for IHD and death and to summarize risk over time [7, 8]. Cox regression was used to calculate the incidence rate ratio as an estimate of the relative risk and associated 95% confidence interval (CI) of hospitalization for IHD among UCEP patients compared to that for controls, while adjusting for alcohol- and smoking-related diseases and level of comorbidity [7, 8]. Cox regression was also used to estimate the mortality rate ratio (MRR) and associated 95% CI for UCEP patients, relative to controls, while adjusting for alcohol- and smoking-related diseases and level of comorbidity. Similarly, Cox regression was used to estimate the MRR for cause-specific deaths. All-cause MRRs also were calculated after <1 year, 1–2 years, 3–4 years, and ≥ 5 years of follow up. For pneumonia, cause-specific MRRs were estimated for the following time periods: within <7 days, 7–31 days, and ≥31 days after the index date.
Separate analyses were performed for each type of IHD (myocardial infarction, angina, and heart failure) and stratified by time elapsed since the index date (<1 year, 1–2 years, 3–4 years, and ≥ 5 years).
Proportional hazards assumptions for the models within time periods were assessed graphically and found to be adequate. Analyses were performed using STATA version 9.1 SE (StataCorp, College Station, Texas, USA). The study was approved by the Danish Data Protection Agency (# 2001-41-1590).