Pancreatitis Is a Risk Factor for Pancreatic Cancer

GASTROENTEROLOGY 1995;109:247-251 Pancreatitis Is a Risk Factor for Pancreatic Cancer PRADEEP BANSAL and AMNON SONNENBERG Division of Gastroenterology and Division of Epidemiology, Department of Veterans Affairs Medical Center and Medical College of Wisconsin, Milwaukee, Wisconsin Background & Aims: The Department of Veterans Affairs (VA) maintains a computerized file of all hospital discharges since 1970. In taking advantage of this large database, the present study aimed to determine whether pancreatitis is a risk factor for pancreatic cancer. Methods: A case control study compared the occurrence of pancreatitis in 2639 patients with pancreatic cancer and a matched control group of 7774 subjects using multivariate conditional logistic regression. Results: The odds ratio associated with pancreatitis increased from 2.04 (95% confidence interval [CI], 1.53-2.72) 7 or more years before the first diagnosis of cancer to 2.14 (CI, 1.68-2.72) 3 or more years before cancer diagnosis and to 2.31 (el, 1.87-2.86) 1 or more years before cancer diagnosis. In a multivariate analysis, all types of pancreatitis or chronic pancreatitis alone were associated with a significant risk, their odds ratios being 3.42 (CI, 1.98-5.91) and 2.23 (Cl, 1.43-3.49), respectively. No influence was exerted by other variables, such as history of gallstone disease or alcoholism, frequency of hospital discharges with pancreatitis, and length of coverage in the VA system. Conclusions: A history of pancreatitis constitutes a significant risk for subsequent development of pancreatic cancer. The increase of the risk with decreasing time before the diagnosis of cancer may indicate that a fraction of pancreatic cancers are initially misdiagnosed as pancreatitis. I t has been suggested that pancreatitis may be a risk factor for the development of pancreatic cancer. *-2 This concept assumes that pancreatic cancer derives from, among other causes, chronic inflammation caused by a preexisting disorder. Analogies can be drawn with the development of esophageal cancer from Barrett's epithelium and hepatic cancer from cirrhosis. However, although the latter associations are well recognized, the link between pancreatitis and pancreatic cancer has been restricted to anecdotal evidence until recently. 3 Prospective cohort studies of patients with chronic pancreatitis have shown an excess of pancreatic cancer compared with the general population. 4-7 The length of follow-up in some studies has raised doubts whether pancreatic cancer was masquerading as pancreatitis during the study pe- riod. 8 The aim of the present study was to determine whether pancreatitis is a risk factor for pancreatic cancer. In addition, whether the severity of pancreatitis and its cause affect the risk of developing cancer was tested. Materials and Methods Data Source The patient treatment file (PTF) comprises a multitude of individual data files. The data files are managed by the Department of Veterans Affairs (VA) central Automation Center at Austin, Texas. Permits to use the data and the computational facilities at Austin are granted to interested investigators by the Automation Center. The main files contain records of all inpatient treatments from all VA hospitals distributed throughout the United States. These files were started in 1970 and are available for each fiscal year since. Although the contents of the files and their structure have changed during the past 2 decades, essential demographic and health characteristics are available in similar form from all main files. Each patient record contains one primary and up to nine secondary discharge diagnoses. From 1970 until 1980, the diagnoses were coded according to the 8th revision of the Clinical Modification of the International Classification of Diseases (ICD); since 1981, the 9th revision of the ICD has been used. Individual subjects can be identified by their social security number. Extraction of Cases Veterans in whom exocrine pancreatic cancer (ICD codes 157.0-157.3 and 157.8-157.9) was initially diagnosed between 1988 and 1992 were extracted from the PTF, individual veterans being identified by their social security number. Patients followed up by the VA system for 1 year or less before the diagnosis of pancreatic cancer were excluded from the case population. Each case was followed up retrospectively through the annual PTF files of previous years to accumulate all previous discharge diagnoses between 1970 until 1 year before the diagnosis of pancreatic cancer. For example, if the patient was found to have pancreatic cancer in 1990, the follow-up period included the annual files from 1970 until 1988, leaving out the 1989 and 1990 discharge diagnoses. Cases were flagged if Abbreviations used in this paper: CI, confidence interval; ICD, International Classification of Diseases; PTF, patient treatment file; VA, Veterans Affairs. 1995 by the American Gastroenterological Association 0016-5085/95/$3.00

248 BANSAL AND SONNENBERG GASTROENTEROLOGY Vol. 109, No. 1 they had pancreatitis as a primary or secondary diagnosis. The pancreatitis was classified according to the ICD code as acute (ICD codes 577.0 and 577.09) or chronic (ICD codes 577.1 and 577.19). Pancreatic pseudocyst (ICD code 577.2)was considered a manifestation of chronic pancreatitis. The occurrence of the following discharge diagnoses was also marked: alcoholism and alcoholic liver disease (ICD codes 303.0-303.93, 305.0-305.03, and 570.0-571.09) and cholelithiasis (ICD codes 573.9-574.50). For each case subject, we also extracted the frequency of hospitalizations, the length of time between the first diagnosis of pancreatitis and pancreatic cancer, and the length of coverage in the VA system. Extraction of Control Patients For each case subject, 4 random control patients were selected from the identical annual PTF as the case. The controls were matched to the case by age, sex, and race. Patients with any previous discharge diagnosis of pancreatic cancer between 1981 and 1992 were excluded from the complete annual files of 1988 until 1992. After these exclusions, the entire patient records from each year between 1988 until 1992 were randomized. For each case subject, the randomized PTF was scanned sequentially until a control subject of the same age, race, and sex who had been discharged 2 weeks before or after the first diagnosis pancreatic cancer in the case was identified. Ten-year age categories were used for matching, i.e., 25 years or younger, 25-34 years, 35-44 years, etc. Individual control subjects were traced retrospectively through files of previous years to find their first appearance in the annual files since 1970 and to accumulate their previous discharge diagnoses. Subjects were eliminated from the control population if they had not been followed up in the VA system for at least 1 year. The profile of each control patient was marked for the same parameters as the case subjects. Statistical Analysis To study the effect of time on the relationship between pancreatitis and pancreatic cancer, the time between the first diagnosis of pancreatitis and pancreatic cancer was separated into three distinct periods. The first period covered the time of 7 or more years, the second period covered the time of 3 or more years, and the third period covered the time of 1 or more years before the first diagnosis of pancreatic cancer. For each time period, the number of cases and controls with any previous diagnosis of pancreatitis were compared by calculating an unadjusted odds ratio and its 95% confidence interval (CI). For all other univariate comparisons, odds ratios and their 95% CI were calculated by the PHREG procedure of SAS using an adjustment for matching of cases and controls. 9 For univariate comparisons between two quantitative parameters, such as age and length of coverage, we calculated the 95 % CI of the mean; for univariate comparisons between qualitative parameters, such as sex and race, we used the X 2 test. In the multivariate analysis, the data from all three time periods were analyzed jointly. The occurrence or absence of pancreatic cancer represented the outcome variable. Any previ- ous diagnosis of pancreatitis or chronic pancreatitis served as the two predictor variables of primary interest. In addition, the confounding influence of the following variables was tested: presence of alcoholism or gallstone disease, frequency of hospital discharges with a diagnosis of pancreatitis, time since the first hospital discharge of pancreatitis, and length of time since the subject's first appearance in the PTF files, i.e., length of VA coverage. Case and control patients were compared by multivariate conditional logistic regression, again using the PHREG procedure of SAS. All predictor variables were included in the model, which also adjusted for matching between cases and controls. Results General Characteristics Each annual main file contained about 1700 hospital discharges with a diagnosis of pancreatic cancer. In 4589 patients, pancreatic cancer was first diagnosed between 1988 and 1992. We excluded 1930 patients who had been followed up in the VA system for 1 year or less before the diagnosis of pancreatic cancer. The final case population consisted of 2639 patients with pancreatic cancer. The initial case population was matched to a random sample of 18,352 subjects without pancreatic cancer. A total of 10,578 subjects were excluded from the control population because they or their matched case had been followed up in the VA system for 1 year or less. The final control population consisted of 7774 subjects. Between 1988 and 1992, 1,667,000 men and 43,000 women were treated in VA hospitals in the United States, their average age being 57.0 years (SE, 0.01). Of all hospitalized veterans, 1,264,000 were white, 326,000 were black, and 120,000 were of other ethnicity. Compared with the total veterans population, the case popula- tion contained relatively more men and fewer women, i.e., 2611 and 28, respectively (X 2 = 28.83; df = 1; P < 0.0001). The case population was also characterized by relatively more blacks and fewer whites, i.e., 1909 whites, 599 blacks, and 125 others (X2 = 38.74; df = 2; P < 0.0001). Lastly, patients with pancreatic cancer were older than the population of all hospitalized veterans, their average age being 67.0 years (z = 865.74; P < 0.0001). However, the overall differences and their clinical importance are small. Highly significant results were obtained primarily because of the large number of subjects. Comparison of Case and Control Patients The case population of 2639 patients with pancreatic cancer was compared with a population of 7774 matched controls. Table 1 shows the breakdown of both populations by type ofpancreatitis and any previous diag-

July 1995 PANCREATITIS AND PANCREATIC CANCER 249 Table 1, Patient Characteristics Patients with cancer Control subjects Variable (n = 2639) (n = 7774) Odds ratio a 95% CI No. of patients with any type of pancreatitis (%) 157 (5.9) 207 (2.7) 2.35 Chronic pancreatitis 93 (3.5) 99 (1.3) 2.94 Acute pancreatitis 64 (2.4) 108 (1.4) 1.76 No. of patients with history of alcoholism (%) 674 (25.5) 1765 (22.7) 1.17 No. of patients with history of gallstone disease (%) 131 (5.0) 349 (4.5) 1.12 No. of hospitalizations 396 485 -- Length of history of pancreatitis (yr) b 8.2 (7.3-9.1) 9.0 (8.3-9.8) -- Length of coverage by VA system (yr) b 11.6 (11.4-11.9) 11.5 (11.3-11.6) -- 1.89-2.92 2.19-3.94 1.28-2.41 1.05-1.30 0.91-1.38 acalculated with adjustment for matching of case and control subjects but without multivariate adjustment for other confounding variables. Walues are expressed as mean with the 95% CI in parentheses. nosis of alcoholism or gallstones. A total of 396 hospitalizations were recorded in the annual files from 1970 to 1 year before the diagnosis of cancer in the case population. For the controls, the total was 485 hospitalizations. The average frequency of hospitalizations per patient with pancreatitis was similar in both the case and the control subjects. The length of history of pancreatitis was slightly shorter among case than control subjects (Table 1). The length of coverage in the VA system was not significantly different among case and control subjects. The overall number of individuals with pancreatitis was 157 in the case population and 207 in the control subjects, the univariate odds ratio being 2.35 (Table 1). In the univariate analysis, chronic and acute pancreatitis were both associated with an increased risk for pancreatic cancer, whereas a history of alcoholism or gallstone disease did not exert an independent influence. Only 5.9% of the patients with pancreatic cancer cases had a history of pancreatitis. The data of Table 1 can be used to calculate an attributable risk of 3.38% (95% CI, 2.38-4.37). l This relatively low value indicates that overall pancreatitis contributed to only very few cases of pancreatic cancer. As shown in Table 2, the odds ratios of each separate time period were also significantly different from unity. The program for the multivariate conditional logistic regression analysis adjusted for matching and the confounding influence of other variables. In the first multivariate model, using "any type of pancreatitis" as predictor variable, only pancreatitis exerted a significant influence (Table 3). In the second multivariate model, "chronic pancreatitis" was the only predictor variable to be significantly associated with an increased risk of pancreatic cancer. No significant influence was exerted by any of the other predictor variables of both models. Alcoholism is prevalent among veterans, as evidenced by the high average prevalence rate of 23.4% (Table 1). Alcoholism alone is not a risk factor for pancreatic cancer. If alcohol-induced pancreatitis behaved differently from other types ofpancreatitis with respect to pancreatic cancer, alcoholism would increase the number of subjects with pancreatitis in the control but not in the case population. Because of the excess of pancreatitis in the control population, the outcome of the analysis would be biased Table 2. Distribution of Case and Control Patients in Different Time Periods Length of follow-up before diagnosis of pancreatic cancer Variable ~1 yr ->3 yr ->7 yr Case patients Total no. of patients 2639 2401 1882 No. of patients with any type of pancreatitis 157 118 81 Chronic pancreatitis 93 46 31 Acute pancreatitis 64 72 50 Mean length of followup (SE) 8.2 (0.5) 10.2 (0.5) 12.8 (0.4) Range of follow-up 1.1-21.6 3.0-21.6 7.1-21.6 Control patients Total no. of patients 7774 6996 5417 No. of patients with any type of pancreatitis 207 165 117 Chronic pancreatitis 99 80 52 Acute pancreatitis 108 85 65 Mean length of followup (8E) 9.0 (0.4) 10.7 (0.4) 13.0 (0.4) Range of follow-up (SE) 1.2-21.8 3.1-21.8 7.0-21.8 Odds ratio a 2.31 2.14 2.04 95% CI 1.87-2.86 1.68-2.72 1.53-2.72 acalculated without adjustment for confounding variables or matching between case and control patients.

250 BANSAL AND SONNENBERG GASTROENTEROLOGY Vol. 109, No. 1 Table 3. Multivariate Analysis of the Risk for Developing Pancreatic Cancer df Odds ratio 95% Cl Wald X 2 Probability X 2 Any type of pancreatitis Any type of pancreatitis History of alcoholism History of gallstone disease No. of hospitalizations Length of history of pancreatitis Length of VA coverage Chronic pancreatitis Chronic pancreatitis History of alcoholism History of gallstone disease No. of hospitalizations Length of history of pancreatitis Length of VA coverage 1 3.42 1.98-5.91 19.3968 0.0001 1 1.08 0.97-1.21 1.8687 0.1716 1 1.01 0.82-1.25 0.0072 0.9325 1 1.04 0.96-1.11 0.9299 0.3349 1 0.82 0.67-1.01 3.3634 0.0667 1 1.02 0.98-1.07 1.1070 0.2927 1 2.23 1.43-3.49 12.3450 0.0004 1 1.09 0.98-1.22 2.3794 0.1229 1 1.04 0.85-1.29 0.1628 0.6866 1 0.99 0.92-1.07 0.0318 0.8584 1 1.10 0.97-1.24 2.1998 O. 1380 1 1.02 0.98-1.06 0.7547 0.3850 NOTE. In the first and second multivariate model, either any type of pancreatitis or chronic pancreatitis alone, respectively, was used as predictor variables. towards the null hypothesis. To test for this possibility, a second set of multivariate analyses was calculated, after all subjects with a discharge diagnosis of alcoholism had been removed from the case and control population. The adjusted odds ratios were 2.77 (0.92-8.29) for any pancreatitis and 3.57 (1.45-8.80) for chronic pancreatitis. They were of a magnitude similar to the odds ratios shown in Table 3, and they do not support the hypothesis that frequent alcoholism biased the outcome of the analysis. Discussion The results of this case-control study strongly suggest that a history of pancreatitis represents a risk factor for development of pancreatic cancer. The risk is particularly associated with chronic but not acute pancreatitis. The association appears to be unrelated to the severity or the extent of the pancreatitis, because the length of its history and the number of previous hospital discharges do not increase the risk. Despite the clear-cut association between pancreatiris and pancreatic cancer, the risk concerns only a small fraction of 6% of all pancreatic cancers among veterans. Pancreatitis has been shown to represent a risk factor for the development of pancreatic cancer. I-2 However, this association has so far been described for only chronic pancreatitis. Except for the study by Ekbom et al., acute pancreatitis has not been studied as an independent risk factor. 3 It is also unclear whether the type ofpancreatitis plays any significant role. Case reports suggest that unusual forms of pancreatitis, such as tropical pancreatitis and chronic calcific pancreatitis, are associated with the development of pancreatic cancer. 11-14 In the present study, a history of alcoholism or gallstone disease were not associated with an independent risk for pancreatic cancer. The data suggest that chronic pancreatitis represents a more important risk factor than acute pancreatitis. In two previous studies, the maximum occurrence of pancreatic cancer was noted within the most recent years of follow-up of the cohort of patients with pancreatitisj 2 The temporal proximity between pancreatitis and pancreatic cancer raised the possibility of a spurious association. It was speculated that pancreatic cancer had masqueraded as pancreatitis during the period of 1-2 years preceding its definitive diagnosis. The period of follow-up in our study ranged from 1 to 22 years with a mean of 8.2 years. The data were analyzed in separate time periods; even as far back as 7-22 years before pancreatic cancer, a significant risk could be shown (Table 2). However, it is obvious from our data that the risk increased in the later years, possibly because a fraction of pancreatic cancers were initially misdiagnosed as pancreatitis or because diagnostic ascertainment of pancreatitis has been most reliable during recent time periods. There are reasons to believe that the odds ratios reported in the present study represent conservative estimates of the true strength of the association between pancreatitis and pancreatic cancer. Lowenfels et al. have reported a higher relative risk in their cohort study. 1 The present analysis was restricted to inpatients. Although treatment of pancreatitis or pancreatic cancer requires hospital admission in the majority of cases, some patients may have been treated as outpatients. Outpatient treatment would have particularly affected the outcome of the analysis if it were more common in pancreatitis than pancreatic cancer. Our analysis could have also underestimated the frequency of pancreatitis if veterans with this diagnosis were frequently treated outside the VA system. They may have entered the VA system only after devel-

July 1995 PANCREATITIS AND PANCREATIC CANCER 251 oping pancreatic cancer for terminal care or because they ran out of financial resources and other means of health insurance. To account for such a potential bias, we restricted our analysis to case and control subjects who could be traced retrospectively in the PTF for more than 1 year. In addition, the length of coverage in the VA system was used as an adjustment variable in the multivariate analysis. Because the hospitalization records come from all over the United States, there is no means to assure a similar standard of diagnostic accuracy as, for instance, in a prospective clinical trial. A medical history may be recorded in an incomplete fashion if it is of little relevance to the acute medical problem. A history of alcoholism or gallstone disease may be go unnoticed if it was previously dealt with outside the VA system. Alcohol consumption is heavy among veterans. The odds ratio of approximately 3 observed in a population of heavy drinkers may not apply to other populations less heavily exposed to alcohol. In conclusion, our study confirms earlier reports that pancreatitis represents a risk factor for the development of pancreatic cancer. The threefold risk found in the veterans population represents a conservative estimate; the true risk may be higher. On the other hand, the long-term follow-up period of the present study excludes the possibility that, in our population, most cases of pancreatic cancer had been previously misdiagnosed as pancreatitis. Because pancreatitis affects only 6% of patients with pancreatic cancer, we need to search for additional environmental risk factors that could explain the recent increase in the incidence of pancreatic cancer and would provide us with a handle at effective prevention. References 1. Lowenfels AB, Maisonneuve P, Cavallini G, Ammann RW, Lankisch PG, Andersen JR, Dimagno EP, Andr6n-Sandberg A, Domel- 16f L, International Pancreatitis Study Group. Pancreatitis and the risk of pancreatic cancer. N Engl J Med 1993;328:1433-1437. 2. Ekbom A, McLaughlin JK, Karlsson BM, Nyr~n O, Gridley G. Adami HO, Fraumeni JFJr. Pancreatitis and pancreatic cancer: a population-based study. J Natl Cancer Inst 1994;86:625-627. 3. ivy E J, Sarr MG, Reiman HM. Nonendocrine cancer of the pancreas in patients under age forty years. Surgery 1990;108:481-487. 4. Imperio DRH, Charafeddine N, Thuviath PJ. Natural history and long-term outcome of patients admitted to the Johns Hopkins University Hospital with chronic pancreatitis (abstr). Gastroenterology 1994;106:A298. 5. Ammann RW, Akovbiantz A, Largiader F, Schueler G. Course and outcome of chronic pancreatitis: longitudinal study of a mixed medical-surgical series of 245 patients. Gastroenterology 1984; 86:820-828. 6. Ammann RW, Knoblauch M, Mohr P, Deyhle P, Largiader F, Akovbiantz A, Schuler G, Schneider J. High incidence of extrapancreatic carcinoma in chronic pancreatitis. Scand J Gastroenterol 1980;15:395-399. 7. Lankisch PG, Lohr-Happe A, Otto J, Creutzfeldt W. Natural course in chronic pancreatitis. Digestion 1993;54:148-155. 8. Ekbom A, McLaughlin JK, Nyr6n O. Pancreatitis and the risk of pancreatic cancer (letter). N Engl J Med 1993;328:1502. 9. SAS Institute. SAS/STAT software: changes and enhancements, release 6.07. Cary, NC: SAS Institute, 1992:435-456. 10. Kahn HA, Sempos CT. Statistical methods in epidemiology. New York: Oxford University, 1989:72-84. 11. Thomas PG, Augustine P, Ramesh H, Rangabashyam N. Observations and surgical management of tropical pancreatitis in Kerala and Southern India. World J Surg 1990;14:32-42. 12. Augustine P, Ramesh H. Is tropical pancreatitis premalignant? Am J Gastroenterol 1992;87:1005-1008. 13. Misra SP, Thorat VK, Vij JC, Anand BS. Development of carcinoma in chronic calcific pancreatitis. Int J Pancreatol 1990;6:307-312. 14. Chari ST, Mohan V, Pitchumoni CS, Viswanathan M, Madanagopalan N, Lowenfels AB. Risk of pancreatic carcinoma in tropical calcifying pancreatitis: an epidemiologic study. Pancreas 1994;9:62-66. Received November 14, 1995. Accepted February 27, 1995. Address requests for reprints to: Amnon Sonnenberg, M.D., Gastroenterology 111-F, Department of Veterans Affairs Medical Center, 2100 Ridgecrest Drive Southeast, Albuquerque, New Mexico 87108. Fax: (505) 256-2803.