LAUSR

Dear Sir, We thank Liu et al. for their interest in our recent study in International Journal of Cancer on statins and pancreatic cancer risk in patients with chronic pancreatitis. Liu et al. raises concern about some potential flaws in our study design, to which we would like to reply. Liu et al. suggest that starting follow-up 1 year after chronic pancreatitis is insufficient to account for misdiagnosis of pancreatic cancer as chronic pancreatitis. We agree with Liu et al. that the choice of a 1-year lag period between chronic pancreatitis and pancreatic cancer is arbitrary. There is no consensus on the optimal choice of lag period. Two years may be preferable to 1 year to mitigate the impact of misdiagnosis of pancreatic cancer by chronic pancreatitis, but this comes at the expense of fewer observed outcomes and reduced precision. To investigate if the 2 years suggested by Liu et al. could explain our results, we conducted an additional analysis, restricting to patients who were alive and at risk of pancreatic cancer 2 years after the date of chronic pancreatitis diagnosis (reducing our study population from 8,311 to 7,742). This restriction reinforced our finding of a null association (adjusted HR: 1.03 [95% CI: 0.60–1.76] compared to 0.90 [95% CI: 0.56–1.44] in the original analysis). This was also the case when including diabetes in the multivariable model (adjusted HR: 0.93 [95% CI: 0.54–1.61] compared to 0.74 [95% CI: 0.46–1.19] when using the original 1-year lag period). These estimates suggest that choosing 1 or 2 years as lag periods between chronic pancreatitis and pancreatic cancer could not explain our observed null finding. Liu et al. also suggest that adjusting for a summary score of comorbidity burden, such as the Gagne Comorbidity Index, may be insufficient to account for potential confounding. In our study-setting with relatively few observed outcomes, adjusting for a wide range of comorbidity conditions would be unfeasible. Instead, we chose to adjust for the Gagne Comorbidity Index. We agree that this may be insufficient. Therefore, we also adjusted for diabetes, which we considered to be a strong potential confounder, to further reduce the potential for residual confounding. The remaining concerns are all related to the variables included in the multivariable Cox model in our study. Liu et al. suggest that the multivariable model needs to be adjusted for some covariates that they argue are confounders. To be considered as a confounder, a variable needs to be (i) associated with the exposure (i.e., imbalanced between exposure groups), (ii) a risk factor for the outcome and (iii) not on the causal pathway between exposure and outcome. Drawing a directed acyclic graph (as in the Supporting Information in the original article) can be a useful means to visualize if a given variable fulfills these requirements. Liu et al. argue that treatment of chronic pancreatitis, cardiovascular and cerebrovascular disease, and exposure to some other prescription drugs (nonsteroidal antiinflammatory drugs, proton pump inhibitors and antidiabetic medications) are confounders and should be adjusted for in the multivariable model. We did not have information on treatment of chronic pancreatitis. However, this is of limited concern. Liu et al. state that surgery reduces risk of pancreatic cancer in patients with chronic pancreatitis. This statement is based on a study by Ueda et al. of 506 chronic pancreatitis patients, of which only one patient underwent surgery and was subsequently diagnosed with pancreatic cancer. This study reported a HR of 0.11 (95% CI: 0.014–0.80). Contrary to this finding, another study of 1,656 chronic pancreatitis patients reported that both endoscopic and surgical treatment was associated with a higher risk of pancreatic cancer compared to no treatment. However, estimates from this study were equally imprecise as in the study by Ueda et al. Accordingly, we find it questionable that surgery should reduce pancreatic cancer risk in chronic pancreatitis. Even if this is the case, we consider it unlikely that surgery would be associated with the exposure (statin prescriptions). In contrast, comorbid cardiovascular and cerebrovascular diseases are related to statin exposure. However, there is no evidence available suggesting that these diseases are risk factors for pancreatic cancer. Accordingly, they do not fulfill requirements to be defined as confounders and should therefore not be included in the multivariable model. Likewise, Liu et al. suggested that the use of nonsteroidal antiinflammatory drugs, proton pump inhibitors and antidiabetic medications could have confounded our results. The use of nonsteroidal antiinflammatory drugs or proton pump inhibitors does not seem to be associated with pancreatic cancer risk in the general population. However, as this association may be different in chronic pancreatitis patients and the general population, they could potentially act as confounders. We therefore conducted an analysis, including these drugs as covariates International Journal of Cancer IJC