Main findings
In this prospective study that included 3,658,417 participants and 202,837 cancer cases, we found that a higher BMI was associated with risk of 18 of 26 cancer types, although these relations differed in terms of direction, shape, and smoking status at baseline. BMI was positively associated with risk of cancers of the corpus uteri, kidney, gallbladder and biliary tract, thyroid, colorectum, breast post-menopausal, multiple myeloma, leukemia, and non-Hodgkin lymphoma (in descending order of linear effect sizes). After restricting the analyses to never smokers to account for incomplete adjustment for smoking, BMI was also positively associated with Hodgkin lymphoma and cancers of the head and neck, and brain and CNS. BMI was associated in an inverse U-shaped manner with the risk of prostate cancer and in an L-shaped fashion with the risk of four cancers (head and neck, esophagus, larynx, and trachea, bronchus, and lung) in the overall cohort likely indicating residual confounding by smoking since the shape of these associations drastically changed among never smokers, except for prostate cancer.
In a subsample of 291,305 participants with a WC measurement and 27,837 cancer cases, we compared cancer risk estimates of WC and BMI. The 99% CIs of the WC and BMI effect estimates consistently overlapped, indicating that WC provides risk associations similar to BMI across a wide range of cancer types in our population.
Strengths and limitations of this study
This study has several strengths. Firstly, to our knowledge, this is the first study to systematically compare both BMI and WC indicators in relation to the risk of a wide variety of cancers, including less frequently occurring ones. Secondly, owing to the large scale of the SIDIAP database, we were able to investigate the association between BMI and numerous cancer types in a Southern European region, increasing the external validity of results previously reported in Northwestern European countries [
3,
4]. Lastly, we previously demonstrated the high quality of cancer diagnoses in the SIDIAP data and we conducted sensitivity analyses in regions where we could include cancer cases confirmed by population-based cancer registries (Additional file
1: Table S6) [
18].
This study also has limitations. Firstly, the inclusion of individuals with a BMI measurement (62% of the SIDIAP adult population) could result in selection bias. However, the study participants were not substantially different from the overall SIDIAP population (Additional file
1: Table S2). Secondly, although we cannot exclude the possibility of exposure misclassification, we were empirically reassured that this was not a serious bias. The distribution of BMI in the SIDIAP was similar to population-based survey data and representative studies of the Spanish population (Additional file
1: Table S11). Thirdly, outcome misclassification could have biased our results towards the null because modest positive predictive values have been reported in a validation study of SIDIAP cancer diagnoses [
18]. Fourth, residual confounding is an inherent limitation of observational studies; an example in our study was residual confounding for smoking status at baseline. Fifth, we did not have data on factors in the possible causal path between obesity and cancer, such as specific reproductive variables (e.g., parity, breastfeeding history), physical activity, and diet. Neither did we have information on cancer subtype or stage at diagnosis, which could have helped sharpen the analyses for certain cancers (e.g.,
prostate cancer). Fifth, while the magnitude of this study’s sample size has its advantages, some of the significant findings of this study could have been related to the large sample size. Another limitation was the missing covariate data which ranged from 10% (for the MEDEA deprivation index) to 39% (for alcohol intake risk). However, the results from our main analysis did not differ when we performed multiple imputations of these data (Additional file
1: Table S5). Finally, we had information for both BMI and WC for only 10% of the study participants. This limited our interpretation of the comparison of adiposity measures associated with cancer risk to individuals with both indicators and does not enable us to extrapolate the WC effect estimates to the general population.
Interpretation and comparison with previous studies
The observed positive associations between BMI and different cancer types are in line with previous studies. The increased risk of breast post-menopausal and corpus uteri cancers has been consistently reported in the literature [
25,
26]. Furthermore, our non-linear analyses showed that the higher the BMI, the greater the magnitude of risk of corpus uteri cancer which concurs with previous studies [
4,
27]. The positive association between BMI and cancers of the colorectum, kidney, thyroid, and gallbladder and biliary tract is well recognized in the literature; however, nuances by subtype (kidney) [
2,
28], histology (thyroid) [
29], and sex (colorectal and gallbladder and biliary tract) have been reported [
25,
30,
31]. In our data, we observed a stronger effect of BMI for gallbladder and biliary tract cancer in women and colorectal cancer in men, which is in line with previous studies (Additional file
1: Table S12) [
25,
31]. Further, our results showed a clear pattern in the association between BMI and hematological cancers. The association observed between BMI and higher risk of leukemia and multiple myeloma has been consistently reported in the literature [
25,
32‐
34], but the association between BMI and the lymphomas is less well established. Although our results for non-Hodgkin lymphoma are supported by two meta-analyses [
25,
35], other studies have only reported a link with the subtype of diffuse large B cell lymphoma [
36]. For Hodgkin lymphoma, we observed a J-shaped association with BMI, which concurs with a large study from the United Kingdom (UK) [
37]. The positive association observed between BMI and cancers of the brain and CNS might have been driven by the inclusion of meningioma in this broad cancer group [
2].
We also observed that the associations between BMI and respiratory tract cancers (head and neck, esophagus, larynx, and trachea, bronchus and lung) were L-shaped, suggesting that low BMIs are an approximation of heavy smoking. In the linear analyses restricted to never smokers, the associations between BMI and cancers of the larynx and esophagus became null, likely due to the opposite effects of BMI in adenoma and squamous cell carcinoma [
25]. Also, among never smokers, BMI became positively associated with cancer of the head and neck and remained negatively associated with cancer of the trachea, bronchus, and lung, which concurs with other meta-analyses [
25,
38‐
40]. For prostate cancer, we found an attenuated inverse U-shaped association which coincided with a large UK study [
4]. The shape of this association could be explained by the dual effect of BMI on prostate cancer (inversely and positively associated with localized and advanced prostate cancer, respectively) [
41]. Unfortunately, we did not have data on prostate cancer subtypes to test this hypothesis.
There were also differences between our results and those of previous studies. Despite the evidence supporting the inverse association between BMI and risk of breast pre-menopausal cancer [
25], we observed a negative trend only with BMI values greater than 27 kg/m
2. In addition, some studies described a positive association between BMI and cancers of the liver and stomach [
42,
43]. Our results suggest these associations are non-linear and similarly shaped to a large UK study (U- and L-shaped for liver and stomach cancers, respectively) [
4]. We noted that the non-linear association for stomach resembled the one for respiratory tract cancers, suggesting residual confounding by smoking status for this cancer as well.
In a post hoc analysis, modeling height and weight in mutually adjusted models, we found that the nine and five cancer types that were positively and negatively, respectively, associated with BMI (in linear models) were also all associated with weight in the same directions. On the other hand, height was positively associated with 14 cancer types (and only negatively associated with corpus uteri cancer) (Additional file
1: Table S9). This suggests that the associations observed for BMI (our main analysis) were driven by excess body weight rather than height. Height is a complex exposure and likely reflects the fact that more stem cells are at risk of acquiring driver mutations during cell division over time. A second possible explanation is that a common factor (such as insulin-like growth factor (IGF) 1) directly affects cancer risk as well as increasing height [
44].
Finally, our results indicate that BMI and WC have a comparable relationship with cancer risk. The effect estimates of BMI and WC were similar although we observed moderate differences for cancers of the bladder, larynx, and trachea, bronchus, and lung. Contrarily to BMI, WC was not negatively associated with the risk of cancers of the larynx and trachea, bronchus, and lung. We hypothesized that this could be explained by smoking since smokers tend to have a higher WC, more visceral adipose tissue, and leaner body mass [
5].