Serum cholesterol and the risk of developing hormonally driven cancers: A narrative review

Abstract Although cholesterol has been hypothesized to promote cancer development through several potential pathways, its role in the risk of developing hormonally driven cancer is controversial. This literature review summarizes evidence from the highest quality studies to examine the consistency and strength of the relationship between serum cholesterol parameters and incidence of hormonally driven cancer. Articles were identified using EMBASE. Longitudinal observational studies published between January 2000 and December 2020 were considered for inclusion. The endpoint of interest was incident prostate, ovary, breast, endometrium, and uterine cancers. In total, 2732 reports were identified and screened; 41 studies were included in the review. No associations were found for ovarian cancer. Most endometrial cancer studies were null. The majority (76.9%) of studies reported no association between cholesterol and prostate cancer. Data on breast cancer were conflicting, associations limited, and effect sizes modest. Our results do not provide evidence for a clear association between cholesterol and different types of incident, hormonally driven reproductive cancers. Future studies should investigate the impact of lipid‐lowering therapy.

several mechanisms, including via the elevated expression of low-density lipoprotein receptor, the upregulation of cholesterol synthesis, and in some instances by triggering lipoprotein accumulation and cholesterol deposition. 13 Cancer cells tend to have high cholesterol content, 14,15 with the accumulation of cholesterol identified in several different human cancers, such as breast, prostate, colon, and others. [16][17][18][19] Furthermore, cholesterol metabolites, such as 27-hydroxycholesterol, have the capacity to signal through the estrogen receptor (ER) or liver X receptor, impacting breast cancer pathophysiology. 20,21 Lastly, in mouse models, the inhibition of proprotein convertase subtilisin/ kexin type 9 (PCSK9), a key protein in the regulation of cholesterol metabolism, may augment immune checkpoint therapy via mechanisms involving the promotion of intratumoral T-cell infiltration, suggesting additional potential roles for cholesterol in the tumor microenvironment which may modify (constrain) cancer growth. 22 The role of cholesterol in the risk of developing hormonally driven cancers is debated; epidemiological and Mendelian randomization studies have shown conflicting evidence regarding the association between cholesterol and hormonally driven cancers; some support a positive association, [23][24][25][26][27] others an inverse association or no association. [28][29][30][31] Reverse causation may contribute to these conflicting findings, as cancer cells are known to decrease serum cholesterol levels through several mechanisms. In addition, a long lag-time between exposure and outcome, protopathic bias (inclusion of cancers that may be preexisting prior to exposure), inadequate sample sizes, the use of cross-sectional study designs, and the inability to fully account for confounding factors may all contribute to the inconclusive evidence available to date.
The objective of this work was to summarize the available literature assessing the link between cholesterol and hormonally driven cancers. Further studies are needed to assess the role of cholesterol in the development of other cancer types.

| METHODS
We conducted a literature review to find studies that assessed the association between serum cholesterol levels and the development of hormonally driven reproductive cancers in humans.

| Eligibility criteria
Studies reporting on adults (≥18 years of age) in the general population who were at risk of developing hormonally driven cancers (and without the cancer of interest at baseline) were included. Studies enrolling both those <18 years old and adults (≥18 years old) were included if >80% of the population were adults, or if separate data were presented for adults. Studies enrolling patients who already had cancer at the start of the study were excluded. The exposure of interest was high serum cholesterol levels versus low serum cholesterol levels. High versus low cholesterol could be assessed incrementally (tertiles and quartiles) or using a single cutoff point. Cholesterol measurements of interest included total cholesterol, high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C). Outcomes of interest were the incidences of breast, endometrial, or ovarian cancer (for women) and the incidence of prostate cancer (for men). Studies reporting on the risk of testicular cancer in men or cervical cancer in women were excluded, since hormone levels have not been previously associated with the incidence of these cancers. Observational studies designed to assess the temporal association between cholesterol and cancer development were considered, including cohort, case-cohort, nested case-control, and Mendelian randomization studies. Both prospective and retrospective study designs were considered. Published systematic literature reviews and/or meta-analyses of studies with relevant designs, as noted above, were also included. The following study designs were excluded: case control studies, case series, and case reports. Interventional studies (i.e., trials) were excluded, due to additional effects that may be independent from lipid lowering. All articles and abstracts reviewed were required to be in English. Non-English papers (even if the abstract was in English) were excluded.

| Literature databases and search strategy
We searched the EMBASE electronic bibliographic database from January 1, 2000, through to December 14, 2020. In addition, we searched congress abstracts listed in EMBASE over a 2-year period (January 1, 2019, through to December 14, 2020). A full-text screen was performed on the references identified from the title and abstract screening. Systematic reviews and meta-analyses were identified during the first pass, and manual bibliographic searching was conducted to identify any additional relevant studies. The search criteria are provided in Table S1. In brief, articles relating to cholesterol, HDL-C, or LDL-C and prostate, ovary, breast, endometrium, or uterus with cancer, carcinoma, tumor, malignancy, or neoplasm terms were identified.

| Screening and abstraction
The identified reports, based on the above eligibility criteria, were screened in a two-step process: (1) title/ abstract screening and (2) full-text screening. Two researchers performed the screening process, with each record screened once by one of the two reviewers. After removing duplicates, citations were screened for inclusion using Microsoft Excel workbooks (Microsoft Corporation, Redmond, WA). Reasons provided for inclusion/exclusion of studies were documented during both stages of screening. Citations were excluded for the following reasons: non-English, non-human study, incorrect patient population, no comparison of cholesterol groups, outcome of interest not reported, incorrect study design, and other. The results of the full-text screen and abstraction were assessed by a second independent reviewer for inclusion in the review. All relevant information was extracted and summarized manually; a final quality-control step was completed to ensure the accuracy of all extracted information summarized in the review. Relevant statistical measures abstracted included relative risk (RR), odds ratio (OR), hazard ratio (HR), or incident rates, as well as the corresponding 95% confidence intervals (CI) and p values.

| Overview of records identified by the literature review
A flow diagram of the screening results is shown in Figure S1. In total, 2732 records were identified and screened, of which 41 references passed both screening steps and were included in the review. An overview of the 41 studies is presented in Table S2 (longitudinal studies) and Table S3 (Mendelian randomization). A summary of the number and directionality of studies by cancer type is presented in Table 1.

| Prostate cancer findings across longitudinal studies
In total, 17 longitudinal studies reporting on the incidence of prostate cancer met the inclusion criteria, including cohorts from Korea, Scandinavia, Western Europe, the USA, and the UK (Table S2). A summary of the outcomes reported is presented in Table 2. The association between total serum cholesterol levels and prostate cancer was reported in 13 studies; the majority ( 63 b An association between total cholesterol and risk of endometrial cancer risk was observed when analyzed by quartiles of total cholesterol, but not when using medical cutoff points for total cholesterol. 74 (assessed by quartiles, quintiles, and clinically relevant cutoff points) and risk of incident prostate cancer. 24,[32][33][34][35][36][37][38][39][40] In contrast, two studies (2/13; 15.4%) reported a positive association between total serum cholesterol and the incidence of prostate cancer, 41,42 and a single study reported an inverse association between total serum cholesterol and the incidence of prostate cancer. 43  Only three studies assessed the association between LDL-C and incidence of prostate cancer. 33,42,44 None of these publications found an association between LDL-C levels (assessed by quartiles and per mmol/L increase) and risk of prostate cancer. Although non-significant, two studies reported estimates that trended toward an association between higher LDL-C and increased risk of prostate cancer. 33,44 Interestingly, in a small study (only 40 cases), an association between higher levels of LDL-C and aggressive prostate cancer was found (HR [95% CI]: 1.83 [95%CI: 1.15-2.90]) in men with triglyceride (TG) levels ≤4.52 mmol. 42 Lastly, the association between low HDL-C levels and prostate cancer was reported in seven studies 33,34,42,[44][45][46][47] ; the majority (6/7) reported no association between HDL-C levels (assessed by quartiles, per standard deviation [SD] and per mmol/L increments, and by cutoff points) and risk of prostate cancer. 33,34,42,[45][46][47] The studies were heterogenous in sample sizes (ranging from 43 to 5409 cases) and in the variables controlled for in the models. One prospective cohort study from the Apolipoprotein MOrtality RISk (AMORIS) database reported an inverse association between serum HDL-C levels (assessed by quartiles) and increased prostate cancer risk (Q4 vs. Q1, HR [95% CI]: 0.81 [0.70-0.94]); similar results were obtained when the first 3 years of follow-up were excluded. 44

Mendelian randomization studies
Three Mendelian randomization studies assessing the association between circulating lipids and prostate cancer were identified (Table 2; Table S3). [48][49][50] None of the identified studies found an association with genetic markers of lipid levels and prostate cancer risk. All three studies had very large sample sizes with >1300 cases, used two sample Mendelian randomization designs, and combined single nucleotide polymorphisms (SNPs) to form genetic scores as the main predictor in the model. The number of SNPs evaluated varied in each study. The numbers of SNPs included in the genetic scores for TGs, LDL-C, and HDL-C were 15, 11, and 36 for the Bull et al. study, 48

| Breast cancer findings across longitudinal studies
The literature review identified 17 records reporting on the association between serum cholesterol levels and risk of breast cancer, including cohorts from Austria, Denmark, France, Korea, Italy, Norway, Sweden, and the USA (Table S2). A summary of the outcomes reported for breast cancer is presented in Table 3. Overall, studies evaluating various serum lipids provided conflicting evidence for breast cancer. Twelve studies evaluated the effect of total serum cholesterol levels on incident breast cancer risk; of these, seven (58.3%) reported no association, 51-57 two (16.7%) reported a positive association, 41,58 and three (25%) reported an inverse relationship 32,33,59 between total cholesterol levels and risk of breast cancer. A more limited number of studies evaluated the association between LDL-C levels and breast cancer; of the five records, none reported an association. 33,51,53,54,56 Lastly, 12 studies reported on the association between HDL-C levels and breast cancer; of these, six (50%) reported no association 51,[53][54][55][56]60 and six (50%) an inverse association. 33,47,59,[61][62][63] Of note, the inverse association observed in the Kucharska-Newton et al. study was limited to the premenopausal cohort; no association was found in the postmenopausal cohort. 63 The effect of total serum cholesterol levels on incident breast cancer risk varied among the studies identified. An analysis of the Women's Health Study cohort, including  51 Furthermore, in stratified analyses by age (<55 vs. ≥55 years) and BMI (<30 vs. ≥30), no association between the highest quartile compared to the lowest quartile of total cholesterol levels and breast cancer was observed. 51 Similarly, an analysis within the prospective Nurses' Health Study reported no association between quintiles of total serum cholesterol levels and breast cancer risk in either pre-or postmenopausal women. 52 Additionally, there was no association after excluding cases diagnosed in the first 2 years of follow-up. 52 A nested case-control study within the prospective de Epidémiologique auprès des Femmes de la Mutuelle Générale de l'Education Nationale (E3N) study also reported no association between tertiles of total serum cholesterol (T1: <4.87, T2: 4.87-5.55, T3: ≥5.56 mmol/L) and risk of breast cancer; furthermore, the risk was not modified by BMI, menopausal status, waist circumference, time between blood donation and diagnosis, or ER status of the tumor. 53 No association between quartiles of total serum cholesterol levels (Q1: <4.80; Q2: 4.80-<5.50, Q3: 5.50-<6.30, Q4: ≥6.30 mmol/L) and risk of breast cancer was also reported following an analysis of the Swedish AMORIS database. 56 In addition, when excluding those with follow-up less than 3 years, no link was observed. 56 In a prospective study of 24,208 postmenopausal women in the Women's Health Initiative, quartiles of total cholesterol showed no association with cancers of the breast even after accounting for possible sources of confounding and bias. 54,55 In a prospective, nested casecontrol study, Lucht et al. also demonstrated no association between tertiles of total cholesterol (T1: 117.2-202.0; T2: 202.0-232.9; T3: 233.0-351.8 mg/dl) and risk of breast cancer among postmenopausal women; a sensitivity analysis after excluding cases diagnosed within 12 months also showed null results. 54 64 In comparison, a positive association between serum total cholesterol levels and risk of breast cancer was reported in three studies. 41 41 Of note, three studies reported an inverse association between quartiles or quintiles of total serum cholesterol levels and breast cancer. 32,33,59 From an analysis of women from a Norwegian cohort (n = 38,823; 708 cases), an inverse association between total serum cholesterol was reported for those with cholesterol in the highest quartile (>6.82 mmol/L) compared with the lowest quartile (<5. 24 33 Likewise, an analysis of participants in the Metabolic syndrome and Cancer Project (Me-Can) reported an inverse association between total cholesterol levels (by quintiles and per 1 mmol/L increment) and breast cancer incidence, with associations persisting for  Total cholesterol, LDL-C, and HDL-C No significant association was detected between any lipid biomarkers and breast cancer risk overall Risk was not modified by menopausal status Analyses adjusted for age at menarche, number of children and age at first full-term pregnancy, menopausal status, age at menopause and current menopause hormone therapy use at blood collection, use of progesterone alone within the last 24 hours before blood collection, family history of breast cancer in first-degree relatives, personal history of benign breast disease, daily alcohol intake, daily glycemic load, daily lipid intake, daily energy intake without alcohol, BMI, and education level    comparisons between the fifth to the first quintile when excluding the first year of follow-up. 32 LDL-C levels were not associated with breast cancer in the five reported studies. 33,51,53,54,56 This agrees with the associations between total cholesterol and breast cancer reported from these studies, as summarized above. The only exception was the analysis from participants among the SU.VI.MAX cohort which reported no association between derived LDL-C levels (Planella's formula) and breast cancer (e.g., per 1 mmol/L increment, HR [95% CI]: 0.83 [0.66-1.05]; p trend 0.1); interestingly, the same study reported that total cholesterol and HDL-C were associated with breast cancer risk (results summarized above and in Table 3). The results were consistent when excluding cases diagnosed during the first 3 years of follow-up. 33 In total, 12 studies reported on the association between HDL-C levels and breast cancer. A prospective study evaluating the association of breast cancer with lipid biomarkers in participants from the Women's Health Study reported no association between HDL-C levels and breast cancer risk ( Abbreviations: BMI, body mass index; CI, confidence interval; ER, estrogen receptor; GWAS, genome-wide association study; HDL-C, high-density lipoprotein cholesterol; HR, hazard ratio; HRT, hormone replacement therapy; IBC, inflammatory breast cancer; IV, instrument variable; LDL-C, low-density lipoprotein cholesterol; MetS, metabolic syndrome; MR, Mendelian randomization; non-HDL-C, non-high-density lipoprotein cholesterol; OR, odds ratio; PMD, percent mammographic density; PR, progesterone receptor; RR, rate ratio; SD, standard deviation; SE, standard error; TG, triglyceride. a Age at baseline or recruitment.

Main conclusions/outcomes
Strong evidence that HDL may be associated with an increased risk of breast cancer, whereas LDL may not be related to breast cancer risk (all models adjusted for age, study site, or country, and principal components for European ancestry) One SD increase in genetically predicted HDL (~15 mg/dl) was associated with a 12% increased risk of breast cancer (OR [95% CI]: 1.12 [1.08-1.16]; p = 1.7 × 10 −9 ) No association was found between genetically predicted LDL and breast cancer risk • LDL, per 1 SD increase in genetically predicted LDL (~37 mg/dl): • OR (95% CI): 1.00 (0.96-1.04); p = 0.88 No significant associations were found between genetically predicted total cholesterol and breast cancer risk Total cholesterol, per 1 SD increase in genetically predicted total cholesterol (~42 mg/dl): OR (95% CI): 1.05 (0.99-1.11); p = 0.11 Genetically elevated plasma HDL and LDL levels appear to be associated with increased breast cancer risk Single-trait MR for HDL, per 1 SD increase: OR (95% CI): 1.08 (1.04-1.13); p < 0.001 Multivariable MR (adjusted for effects of LDL, TGs, BMI, and age at menarche) for HDL per 1 SD increase supported the single-trait observation for HDL: OR (95% CI): 1.06 (1.03-1.10); p = 4.93 × 10 −4 Multivariable MR analysis for LDL per 1 SD increase showed a relationship between LDL and breast cancer risk: OR (95% CI): 1.03 (1.01-1.07); p = 0.02 No difference in the relationships for HDL or LDL were observed when stratified by breast tumor ER status Genetically raised LDL was associated with higher risk of breast cancer: OR (95% CI) per SD: 1.09 (1.02-1.18); p = 0.020, and higher risk of ER+ breast cancer: OR (95% CI) per SD: 1.14 (1.05-1.24); p = 0.004, with no association with ER-breast cancer (p = 0.577) Genetically raised HDL had no nominally significant association with overall breast cancer risk: OR (95% CI) per SD: 1.07 (0.97-1.19); p = 0.171, or ER-breast cancer: OR (95% CI) per SD: 1.09 (0.91-1.30); p = 0.365, but appeared associated with higher risk of ER+ breast cancer: OR (95% CI) per SD: 1.13 (1.01-1.26); p = 0.037 LDL-raising variants in PCSK9 were associated with increased breast cancer risk: OR (95% CI) per SD: 1.10 (1.02-1.19); p = 0.014, but not with ER+ (p = 0.099) or ER-(p = 0.089) breast cancer HDL-C tertiles (T3 vs. T1, OR [95% CI], 0.98 [0.74-1.30]; p = 0.84); the results were not modified when assessed by menopausal status. 53 Similarly, no association between breast cancer and HDL-C levels was reported in cohort studies from Korea (assessed by HDL-C < 40 vs. ≥40 mg/ dl) 60 and Sweden (assessed by HDL quartiles). 56 Another study that reported no association between HDL-C levels (assessed by HDL-C tertiles) and breast cancer was among postmenopausal women who were enrolled in two large prospective cohorts in the USA. 55 In contrast, five studies reported an inverse association between HDL-C levels and breast cancer risk, including a prospective nested case-control study among postmenopausal women of the ORDET cohort (HDL-C ≤ 55 vs. >55 mg/dl, RR [95% CI]: 1.60 [1.10-2.33]). 61 A second nested case-control study among participants from the USA reported a negative association between low HDL-C levels (assessed using cutoff points [<50.00, 50.00-64.99, and ≥65.00 mg/dl (reference)] or as a continuous variable) and the increased risk of inflammatory breast cancer (OR [95% CI]: for <50.00 mg/dl: 2.6 [1.7-4.0]; for 50.00-64.99 mg/dl: 2.0 [1.3-3.0]); no significant difference in results was observed when assessed by ER status of cases. 62 An analysis among two Norwegian population-based screening surveys reported an inverse association between quartiles of HDL-C and postmenopausal breast cancer risk (combined cohort, Q4 vs. Q1, RR [95% CI]: 0.75 [0.58-0.97]; p trend 0.02); no significant association was observed in the premenopausal cohort. 59 Two further prospective studies among cohorts based in France and Denmark also reported an inverse association between HDL-C levels (assessed by quartiles, per 1 SD [19 mg/dl] increment, and per one unit increment) and breast cancer risk. 33,47 Of note, the inverse association between HDL-C and breast cancer risk in the French cohort was borderline non-significant when assessed according to menopausal status (premenopausal: per one unit increment in HDL-C, HR [95% CI]: 0.31 [0.10-1.00], p trend 0.05; postmenopausal: per one unit increment in HDL-C, HR [95% CI]: 0.53 [0.28-1.01], p trend 0.05). 33 Finally, two studies reported borderline inverse associations between HDL levels and risk of breast cancer. An analysis among postmenopausal women of the Women's Health Initiative cohort reported a borderline association between HDL quartiles and breast cancer risk (

Mendelian randomization studies
Four Mendelian randomization studies were identified that evaluated the association between circulating lipids and breast cancer (Table 3; Table S3). 49,[65][66][67] All four studies had very large sample sizes with >1000 cases, used two-sample Mendelian randomization designs, and combined SNPs to form genetic scores as the main predictor in the model. The number of SNPs included in the genetic scores for TGs, LDL-C, and HDL-C were 4, 44, and 28 for the Nowak et al. study 67  Four studies reported the association between genetically predicted increase in LDL-C and breast cancer. 49,[65][66][67] Two of these found a significant association between genetically predicted LDL-C and breast cancer, 66,67 reporting ORs (95% CI) of 1.03 (1.01-1.07); p = 0.02 and 1.09 (1.02-1.18); p = 0.020, respectively. The other two studies found no association between genetically predicted LDL-C and breast cancer. 49,65 Of the two studies examining relationships between LDL-C and ER+ or ER-breast cancer, 66,67 one found an association. An association between genetically elevated LDL-C (per SD increase) and a higher risk was observed for ER+ breast cancer (OR [95% CI]: 1.14 [1.05-1.24]; p = 0.004) but not for ER-breast cancer (p = 0.577). 67 Furthermore, PCSK9 genetic variants known to result in elevated LDL-C levels were associated with increased breast cancer risk overall (per SD, OR [95% CI]: 1.10 [1.02-1.19]; p = 0.014). 67 Notably, although Orho-Melander et al. found no significant associations with the identified SNPs in the genetic scores for HDL-C or LDL-C, they found a significant decrease in the odds of breast cancer for each copy of HMGCR rs12916 allele, which mimics the effect of statins (OR [95% CI]: 0.89 [0.82-0.96] per LDL-lowering T-allele). 49 Similar to the LDL-C findings, half of the studies examining the relationship between HDL-C and breast cancer observed a significant association. 65,66 In contrast, one study found no association with overall breast cancer, 49 and one study found no association with overall breast cancer risk or ER-breast cancer. 67 Only a modest effect was observed in the studies which found a significant association. These publications found between a 6% and 12% increased risk of breast cancer (OR [95% CI]: 1.06 [1.03-1.10]; p = 4.93 × 10 −4 and 1.12 [1.08-1.16]; p = 1.7 × 10 −9 ), respectively. 65,66 In the two studies examining ER status, one found no association 66

| Endometrial cancer findings across longitudinal studies
The risk of endometrial cancer has risen substantially over the years, mainly due to increases in BMI. 68 While there is a clear association between obesity and risk for endometrial cancer, it is still unknown whether lipoproteins, after controlling for obesity, have a major impact on the development of endometrial cancer. From a mechanistic perspective, high levels of adipose tissues increase estrogen by improving the efficiency of androstendione, leading to lower levels of peroxisome proliferator activated receptor alpha, which is the involved in the catabolism of lipoproteins. 69,70 To investigate the link between lipoproteins and endometrial cancer, we found five prospective cohort studies reporting on the association. 54,[71][72][73][74] A summary of the outcomes reported for endometrial cancer is presented in Table 4.
For the lipoproteins of interest, four of the five studies reported associations between the risk of endometrial cancer by total cholesterol, LDL-C, and HDL-C, 54,[72][73][74] while one reported only on total cholesterol. 71 Of the four studies reporting on LDL-C, none found an association between LDL-C and risk of endometrial cancer. 54 74 Of the four studies that assessed the relationship between HDL-C levels and incidence of endometrial cancer, only one reported an association. 72 The authors reported an inverse association between quartiles of HDL-C and risk of endometrial cancer ( 30-0.72]; p trend 0.0003); however, after further adjustment for sex steroid hormones and other obesity-related hormones, the observed association between HDL-C and endometrial cancer was no longer significant (p trend 0.09). 72

Mendelian randomization studies
One record reporting on a bidirectional, two-sample Mendelian randomization analysis to investigate the relationship between circulating lipids (LDL-C, HDL-C, and TGs) and endometrial cancer risk was identified. 75 Genetic variants associated with each of the lipid traits of interest were identified and assessed using data from two genome-wide association studies (Global Lipids Genetics Consortium, n = 188,578 76 ; and Endometrial Cancer Association Consortium, 12,906 cases and 108,979 controls). 77 Genetically raised LDL-C was reported to be associated with a lower risk of endometrial cancer, both overall and for endometrioid and non-endometrioid subtypes (Table 4; Table S3). In contrast, a higher genetically predicted HDL-C level was reported to be associated with an increased risk of non-endometrioid endometrial cancer (p = 0.036); however, this finding was not significant after adjusting for multiple comparisons.

| Ovarian cancer findings across longitudinal studies
Only two studies evaluated the relationship between circulating cholesterol levels and ovarian cancer. 54,56 A summary of the outcomes reported for ovarian cancer is presented in Table 5. An analysis among postmenopausal women from the Women's Health Initiative cohort (n = 24,208; 115 cases) reported no association between quartiles of total cholesterol, LDL-C, and HDL-C with ovarian cancer risk. 54 Similarly, an analysis within the Swedish AMORIS database (n = 27,394; 808 cases) showed no association between circulating levels of total cholesterol, LDL-C, and HDL-C with risk of ovarian cancer (assessed by quartiles). 56 There were no Mendelian randomization studies for ovarian cancer located in the literature.

| CONCLUSIONS
The aim of our study was to summarize the association between serum cholesterol levels and incidence of hormonally driven cancers, with a focus on epidemiological studies with designs supporting the highest level of causal evidence. We identified 41 longitudinal or Mendelian randomization studies reporting on the association between circulating cholesterol levels and risk of prostate, breast, endometrial, or ovarian cancers. These results provide evidence for a complex association between cholesterol and different types of incident hormonally driven reproductive cancers. Even among the same types of cancer and serum cholesterol measures, often studies reported conflicting results and overall failed to convey evidence of an effect of plasma lipids.
Thus far, the evidence reviewed does not support an association between serum cholesterol variables (either total, HDL-C, or LDL-C levels) and incident ovarian cancer. Additionally, there were no observed associations between endometrial cancer and serum cholesterol variables among the longitudinal studies. However, the single Mendelian randomization study examining this relationship found that higher genetically predicted levels of HDL-C and lower genetically predicted levels of LDL-C were associated with incident endometrial cancer. The association with HDL-C is likely spurious as the confidence interval for the comparison with HDL included 1 [estimate 1.07 (1.00-1.14)] and the p-value was non-significant after adjusting for multiple comparisons. 75 As stated, this study also reported a protective effect of elevated LDL-C levels.
Despite this single finding, the preponderance of evidence suggests no association of serum cholesterol levels and endometrial cancer. Similarly, the evidence reviewed does not support an association between cholesterol and prostate cancer. Most of the longitudinal studies and all the Mendelian randomization studies assessing the relationship between cholesterol and prostate cancer found no association. In the few studies reporting an association, the direction is inconsistent. Two publications reported an association between higher levels of total cholesterol and an increased risk of prostate cancer 41,42 and one reported lower levels of total cholesterol were associated with an increased risk of prostate cancer. 43 The estimates from both Kitahara et al. and Kok et al. are derived from statistical models which are minimally adjusted; therefore, it is unclear whether either association would remain with additional adjustment. Finally, we cannot discount the results from the Heir et al. study due to the long duration of follow-up and statistical analysis considering competing risks (i.e., death). As this is the only study reporting an inverse association, this relationship needs further exploration with similar statistical methods.
Finally, the association between cholesterol levels and breast cancer requires further research. Interestingly, our summary of longitudinal studies indicates a potential effect of lower HDL levels and increased breast cancer risk in some subgroups (i.e., post-menopausal and obese). To examine these relationships often requires a large sample size and statistical evaluation of the interaction. These variables were not handled consistently across these studies. For example, menopausal status is often treated as a confounding variable 33,51,61 ; however, there may be an interaction between serum cholesterol, menopausal status, and breast cancer development. 54,58,59,63,64 In order to elucidate these potential interactions, future studies should be powered to evaluate potential interactions with menopausal status and obesity. Furthermore, only a modest effect of LDL and HDL on the risk of breast cancer was observed in the Mendelian randomization analyses, and potential pleiotropic effects cannot be ruled out. Therefore, these relationships require further exploration.
The limitations of this study include the inability to account for potential differences in risk factors related to cholesterol across geographic and ethnic populations. The data were not pooled for meta-analysis due to the wide variety of populations, study designs, and serum cholesterol parameters included in the studies. Studies in this review were not formally assessed with a quality tool, but common weaknesses across included studies are insufficient sampling, short follow-up time, and potential misclassification, such as pre-clinical bias. In cases of pre-clinical bias, undiagnosed cancer may already be affecting serum cholesterol as the cancer may utilize cholesterol for cell replication and therefore result in decreased serum levels. This could result in a mistaken association between low serum cholesterol and cancer diagnosis. The strengths of this analysis include the robustness of the review of studies and selection of those with the strongest epidemiologic study designs.
In summary, these results show no clear signal between serum cholesterol (total, HDL-C, or LDL-C) and incident, hormonally driven cancer. The apparent lack of a role for cholesterol in the risk of developing the hormonally driven cancers in the studies included in this review may be related to confounders, modifiers, or the design of such studies (limitations discussed above) that prevent clear patterns of association from emerging. While serum cholesterol levels do not necessarily reflect the status of intracellular cholesterol trafficking, these cells are reliant on circulating cholesterol for cancer development and growth. This need may be particularly elevated for malignant and fast-growing tumors and for steroid hormone-dependent cancers.
Thus, additional research into the role of cellular cholesterol is warranted. In addition, we would suggest conducting a rigorous life course epidemiologic study examining both the genetic and longitudinal association between serum lipids (in adolescence, young adulthood, and later adult life) and cancer risk. Such a study would help to elucidate biological and behavioral processes that operate in combination across an individual's life that influence cancer risk, accounting for multiple confounders such as diet, exercise, and use of lipid-lowering therapy.
Pharmaceuticals, Inc. according to Good Publication Practice guidelines (http://annals.org/aim/artic le/24248 69/ good-publi catio n-pract ice-commu nicat ing-compa ny-spons ored-medic al-resea rch-gpp3). The authors were responsible for all content and editorial decisions, and received no honoraria related to the development of this publication.

FUNDING INFORMATION
This analysis was funded by Regeneron Pharmaceuticals, Inc.