PATIENTS AND METHODS

The methodology used for the record linkage study based on the Swedish Inpatient Register has been described in detail elsewhere.⁶ In brief, the National Board of Health and Welfare started collecting data on individual hospital discharges in the Inpatient Register in 1964–65. In addition to the national registration number, a unique personal identifier assigned to all Swedish residents, each record contains administrative and medical data, such as hospital department and discharge diagnoses. Record linkage of the study cohort to the nationwide Register of Causes of Death allowed us to obtain information on date of death among those deceased up to 1995. Corresponding linkage to the Emigration Register identified dates of emigration, when applicable. The National Swedish Cancer Register, founded in 1958 and close to 98% complete,⁷ was used to ascertain all incident cancers. To remove records with incorrect national registration numbers, for which no matches would be expected on record linkages and which would contribute person time without any outcome, we also linked the cohorts to the Register of the Total Population.

We excluded records: (1) with erroneous or incomplete national registration numbers; (2) with inconsistencies uncovered during record linkage; (3) of patients who died during the index hospitalisation; and (4) of patients with prevalent cancers at entry. Records deleted as a result of the first two reasons listed were unlikely to be followed up, as indicated by the fact that only four pancreatic cancer cases were ascertained among 10 749 deleted records. Following these exclusions, we identified 178 688 patients with a discharge diagnosis of alcoholism (ICD-7=307, 322; ICD-8=291, 303; ICD-9=291, 303, 305A), 8452 for chronic pancreatitis (ICD-7=587.10, 587.19; ICD-8=577.10, 577.19; ICD-9=577B), 13 553 for alcoholic liver cirrhosis (ICD-7=581.10; ICD-8=571.00; ICD-9=571C, 571D), and 7057 for non-alcoholic liver cirrhosis (ICD-7=581.00, 583.10; ICD-8=571.01, 571.90, 571.98; ICD-9=571A, 571F, 571G), during 1965–1994. The chronic pancreatitis cohort was further subdivided into alcoholic or non-alcoholic subgroups according to a documented hospitalisation for alcoholism.

We only considered first primary cancers. Follow up time (person years) was therefore calculated from the first hospitalisation for the disease under study until the occurrence of a first cancer diagnosis, emigration, death, or the end of the observation period (31 December 1995), whichever occurred first. As these five cohorts partly overlapped, 10 075 patients were transferred from the alcoholism cohort to the appropriate alcoholism plus complication cohort when the complication was first diagnosed. Patients with both chronic pancreatitis and liver cirrhosis were allocated to the chronic pancreatitis cohorts.

To minimise the possible influence of selection bias, we discarded person time and cancer cases occurring within the first year of follow up. To avoid possible ascertainment bias associated with differential autopsy rates between alcoholics and the general population, we did not count the first primary cancers found incidentally at autopsy. The expected number of cancers was calculated by multiplying the number of observed person years in age (five year groups), sex, and calendar year strata by the corresponding stratum specific cancer incidence rates, derived from the entire Swedish population. The incidence rates were aggregated by five calendar years to avoid instability of rare cancers. The relative risk of cancer was estimated as the standardised incidence ratio (SIR), defined as the ratio of the observed number of cancers to that expected. The 95% confidence interval (CI) of the SIR was calculated on the assumption that the observed number follows a Poisson distribution.^8,9 We also stratified the analyses by selected cohort characteristics that may influence risk patterns, including age at first hospitalisation for the disease under study and follow up duration. A χ² statistic was used to test any monotonic trend of SIRs.⁹

We also used an indirect method to evaluate the confounding effect of smoking. Let I be the pancreatic cancer incidence rate among non-smokers; RR the relative risk of pancreatic cancer associated with ever smoking; PY the observed person years; P1 the percentage of ever smokers among alcoholics; and P2 that among the general population. Assuming similar distributions of other risk factors for pancreatic cancer among alcoholics and the general population, and the fact that the ratio of observed person years in non-smoker and smoker subgroups can be reasonably approximated by the percentage of non-smokers and ever smokers, the observed number of cases can be approximated as I×PY×(1+RR×P1−P1), and the expected number of cases as I×PY×(1 RR×P2−P2). The SIR associated with confounding of smoking can be approximated as (1+RR×P1−P1)/(1+RR×P2−P2).

RESULTS

DISCUSSION

Compared with the general population, alcoholic patients without alcoholic chronic pancreatitis or alcoholic liver cirrhosis had a modest 40% excess risk for pancreatic cancer. Even among alcoholics who were diagnosed with chronic pancreatitis or liver cirrhosis, only twofold risks for pancreatic cancer were observed whereas a substantial but transient risk increase was found among non-alcoholic chronic pancreatitis patients. The remarkable excess risks during the first few years of follow up suggest that in some instances pancreatitis may be an early manifestation of an as yet undiagnosed pancreatic cancer. This would be an example of reversed causation or selection bias rather than a true causal association between pancreatitis and pancreatic cancer, as also indicated in our previous study.¹⁰ The relative risk of pancreatic cancer among non-alcoholic liver cirrhosis patients was close to unity.

As stated in several reviews,^1–,³ most previous epidemiological studies could not demonstrate an excess risk for pancreatic cancer associated with moderate alcohol intake. Recent studies corroborate these findings.^11–,¹⁸ In contrast, heavy alcoholic intake has been associated with an increased risk of pancreatic cancer in seven studies.^19–,²⁶ A summary estimate of the relative risk among alcoholics or populations with heavy alcohol intake, mostly based on mortality data and small sample sizes, indicates a non-significant standardised mortality ratio of approximately 1.2.³ More recent studies of alcoholics based on incident cases show higher relative risks (SIR 1.3–2.6).^27–,²⁹ In a cohort study among alcohol abstainers, a significantly reduced risk of pancreatic cancer was found, although the confounding effect of smoking was not controlled for in the analysis.³⁰ Selection bias due to low participation rate, recall bias, and chance finding as a result of small sample size may be some of the reasons for these discrepant results.

The most serious limitation of our study was the lack of information on smoking, an established risk factor for pancreatic cancer.³¹ The habit is clearly overrepresented in alcoholics. We estimated to what extent confounding by smoking might generate the 40% overall excess risk we found among patients with alcoholism. Assuming a relative risk of 2 for pancreatic cancer among current smokers,³² 80% prevalence of smokers among alcoholics,³³ and 30% in the Swedish general population above 18 years old in 1983,³⁴ confounding by smoking will move a true SIR of unity among alcoholics to approximately 1.4. Thus the observed excess risk in our alcoholics without complications may be almost totally attributable to the confounding effect of smoking. Even among alcoholics with chronic pancreatitis or liver cirrhosis, the excess risk is only about 100%. The confounding effect of smoking in these two cohorts may be more obvious, notably as smoking is also a risk factor for alcoholic chronic pancreatitis³⁵ or alcoholic liver cirrhosis.³⁶ There are no other conceivable confounders that would bias our results towards unity. If anything, lower socioeconomic status, lower intake of vegetables and fruits, deficiency of folate, etc, among alcoholics^31,37,38 would bias the relative risk estimates upwards.

The advantages of our study include the prospective study design, a near complete follow up, and a much larger sample size than in previous studies. Possible selection bias was minimised as the first year of follow up was excluded. Despite the advantages of our study, register based studies also have limitations apart from lack of information on possible confounding factors. A mixture of exposed subjects in the “unexposed” reference category by using the general population as reference could lead to underestimation of the true association.

In summary, the excess risk for pancreatic cancer among alcoholics was small, and may be totally attributable to confounding by smoking. Even among alcoholics who developed complications—that is, chronic pancreatitis or liver cirrhosis—the excess risk for pancreatic cancer was modest. Hence alcohol is unlikely to play a causal role in the aetiology of pancreatic cancer.