For all models, we checked the proportional hazard assumption by using the Schoenfeld test of proportionality and by visual inspection of the scaled Schoenfeld residuals [22].

We assessed the robustness of our primary objective findings by performing six sensitivity analyses. First, we accounted for residual selection bias by additionally adjusting model 2 for the number of GP consultations in the year of the index date because participants who see their GP more often may have different health behaviors than those who see their GP less often. Second, we explored potential outcome misclassification by restricting the analyses to specific regions of Catalonia where we had access to population-based or hospital cancer registries. We considered as cancer cases only those who had the same diagnosis in the SIDIAP and a cancer registry. Third, we addressed potential reverse causality (i.e., undiagnosed cancer affecting BMI) by extending the minimum follow-up time (of 1 year in the main analyses) to 2 and 4 years. Fourth, we strengthened the validity of our results by performing multiple imputations (using the fully conditional specification approach, with 10 imputed data sets created) to deal with missing values of model 2 covariates [23, 24]. Fifth, we avoided confounding in the analyses of BMI and specific cancer types by re-running model 2, additionally adjusting for HRT use in post-menopausal women [women-only cancers] and excluding participants with a diagnosis of chronic hepatitis B/C [liver cancer risk factor] or a helicobacter pylori infection [stomach cancer risk factor]). Finally, to investigate to which extent the relationships between BMI and risk of each cancer type represents an effect of weight, height, or both weight and height, we re-ran the multivariable-adjusted models (model 2) with height and weight as the main exposures, mutually adjusted for each other.

To assess the robustness of our secondary findings, we performed two sensitivity analyses. We re-ran the analyses that compared BMI and WC in relation to cancer risk with mutual adjustment for both adiposity indicators using residuals of WC and BMI (e.g., we regressed WC on BMI, and we included the residuals from this analysis in the model using BMI as an indicator of general adiposity) to assess if this added valuable information to fully capture adiposity [9]. Finally, we added height as an adjustment variable to the analyses that compared BMI and WC in relation to cancer risk.

The a priori level of statistical significance was set at a 2-sided P value of 0.01 for all analyses. We used STATA version 15.1 (College Station, TX, USA) for data analysis and R version 3.5.0 for data visualization.

We obtained approval from the Clinical Research Ethics Committee of the IDIAPJGol (project code: P14/074) to perform this study. 

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.