Postdiagnostic Inflammatory, Hyperinsulinemic, and Insulin-Resistant Diets and Lifestyles and the Risk of Prostate Cancer Progression and Mortality

This article is featured in Highlights of This Issue, p. 1669

Prostate cancer is the most commonly diagnosed cancer among men in the United States (1). Over the past decade, research has identified various dietary and lifestyle factors associated with survival following prostate cancer diagnosis (2). However, much of the evidence regarding diet, in particular, remains mixed, leading to uncertainty about the role these factors play in improving outcomes following a diagnosis. Many past studies have examined single dietary factors in isolation, which does not adequately represent the combined impact of dietary intake on biological responses or the complex interactions in whole diets (3, 4). Therefore, it is important to examine dietary patterns to try to understand diet–prostate cancer relationships.

Inflammation and insulin pathways have been linked to cancer development and progression (5), including in the setting of prostate cancer (6–12). Postdiagnostic behaviors that overactivate these pathways may therefore lead to poorer prostate cancer outcomes. The empirically derived inflammatory, hyperinsulinemic, and insulin resistance dietary indices—calculated from food frequency questionnaire (FFQ) data—and the associated lifestyle indices provide an opportunity to study the role of diet- and lifestyle-related inflammation- and insulin-promoting behaviors in prostate cancer outcomes (13, 14). All five indices were developed in the Nurses’ Health Study (NHS) and validated in both the NHS-II and the Health Professionals Follow-Up Study (HPFS; ref. 13, 14). These novel indices have been studied in relation to risk of diabetes (15), colorectal cancer (onset and progression; ref. 16, 17), pancreatic cancer (18), hepatocellular carcinoma (19), and prostate cancer (among previously healthy men; ref. 20); however, they have not been examined in men diagnosed with prostate cancer. Because they predict plasma concentrations of circulating markers of inflammation (IL6, C-reactive protein, and tumor necrosis factor α receptor 2; ref. 13), hyperinsulinemia (C-peptide; ref. 14), and insulin resistance (triacylglycerol to high density lipoprotein cholesterol; TAG:HDL; ref. 14), the indices allow for the measurement of the inflammatory and insulinemic potential of whole diets and associated lifestyles without the necessity for biomarker data.

Here, we used these indices to examine associations between the inflammatory and insulinemic potentials of dietary patterns and lifestyle habits after a prostate cancer diagnosis and the risk of disease progression (primary outcome) and disease-specific mortality (secondary outcome) among men in the Cancer of the Prostate Strategic Urologic Research Endeavor (CaPSURE) cohort. Given the role that increased adiposity may play in activating these pathways (5, 21), we also examined whether obesity modified associations with the dietary indices.

CaPSURE is a longitudinal observational cohort of 15,310 men with biopsy-proven prostate cancer. Participants were enrolled at 43 urology practices across the United States starting in 1999. Data were collected on diagnostic and other clinical features, treatments, and clinical follow-up. Additional details of the CaPSURE cohort have been previously reported (22). All participants provided written informed consent, and the study was conducted in accordance with the Belmont Report and the U.S. Common Rule under local Institutional Review Board approval.

The CaPSURE Diet and Lifestyle (CDL) sub-study—consisting of a comprehensive lifestyle questionnaire and full-length FFQ—was administered at three time points between 2004 and 2016; a total of 2,891 men participated in at least one administration. For the subset of men who participated in more than one administration (n = 443), only the first administration (closest to diagnosis date) was used. Men with a last clinical follow-up (n = 160) or documented progression (n = 391) prior to completion of their first CDL questionnaire were excluded. Those with unknown or extreme caloric intake (<800 kcal/day or >4,200 kcal/day; n = 153) and/or missing ≥70 FFQ items (n = 20) were also excluded, consistent with the recommended approach to address implausible energy intakes (23–25). Finally, men with undocumented or unknown clinical T-stage (n = 100) or T-stage >T3a (n = 8) and those with death from an unknown cause (n = 3) were excluded, resulting in a sample size of 2,056 men with nonmetastatic disease.

We collected data on education, family history of prostate cancer, smoking history, medical history, supplement use, and height and weight [used to calculate body mass index (BMI)] via questionnaire. Self-reported dietary intake was collected via a validated (23, 26–28) semiquantitative FFQ that asked about average consumption of approximately 140 foods/beverages over the past year. Participants reported how often over the prior year they had consumed a specified portion size of each of the items. They could choose from 9 frequency options ranging from never or less than one serving per month to six or more servings per day.

Physical activity was assessed via a validated physical activity questionnaire developed primarily to capture leisure-time physical activity (29). Participants were asked to report the average weekly time spent doing various aerobic and resistance exercises. Participants could choose from 11 frequency options ranging from 0 minutes to 11 hours per week. A metabolic equivalent of task (MET) value was assigned to each activity based on the energy required by that activity relative to the resting metabolic rate (30). MET-hours per week were calculated by multiplying the MET value for an activity by the reported time per week spent doing that activity.

Development and validation of the empirical dietary inflammatory pattern (EDIP), empirical dietary index for hyperinsulinemia (EDIH), and empirical dietary index for insulin resistance (EDIR) have been detailed previously (13, 14). Briefly, each index was developed in the NHS and subsequently validated in both the NHS-II and HPFS cohorts. Blood samples taken at the beginning of each cohort were used (EDIP development specified samples were only taken in participants free of cancer, diabetes, heart disease, or stroke diagnosis). Diets were measured via FFQs (similar to the FFQ administered in CaPSURE) based on 39 predefined food groups that were updated at regular intervals in each cohort. FFQs completed closest to the date of blood draw were used for development and validation of each index. The EDIP was created using reduced-rank regression and stepwise linear regression to identify a dietary pattern that was predictive of three plasma inflammatory markers (IL6, C-reactive protein, and tumor necrosis factor α receptor 2), resulting in the inclusion of 9 pro-inflammatory (red meat, processed meat, organ meat, other fish, tomatoes, other vegetables, refined grains, low-energy beverages, and high-energy beverages) and 9 anti-inflammatory (coffee, tea, fruit juice, wine, beer, leafy green vegetables, dark yellow vegetables, snacks, and pizza) food groups. Multivariable models found the EDIP was significantly associated with all three inflammatory biomarkers, with significant linear trends for each biomarker across quintile of the EDIP observed (13). These results were replicated in validation studies, which also found significant associations with adiponectin, a marker of overall inflammation, which was notably not used in the development of the EDIP (13).

Using a similar method, the EDIH and EDIR were created by identifying dietary patterns that were most predictive of plasma C-peptide and TAG:HDL, respectively. Although similar metrics, hyperinsulinemia is a consequence of prolonged insulin resistance (5) due to diminished cellular response to insulin, resulting in additional insulin secretion and subsequently high levels of insulin relative to glucose. The EDIH included 13 proinsulin secretion (red meat, processed meat, other fish, poultry, eggs, margarine, butter, cream soups, low-fat dairy, French fries, tomatoes, high-energy beverages, low-energy beverages) and 5 anti-insulin secretion (coffee, wine, high fat dairy, leafy green vegetables, whole fruit) food groups. The EDIR included 10 proinsulin resistance (red meat, processed meat, other fish, tomatoes, cream soups, other vegetables, refined grains, margarine, fruit juice, and low-energy beverages) and 8 anti-insulin resistance (coffee, wine, liquor, beer, high fat dairy, nuts, leafy green vegetables, and dark yellow vegetables) food groups. Multivariable models found the EDIH and EDIR were significantly associated with C-peptide and TAG:HDL biomarkers, with the results replicated in validation studies (14).

The resulting EDIP, EDIH, and EDIR are weighted sums of 18 index-specific food groups (some overlapping), with higher indices reflecting diets with greater inflammatory (EDIP) or insulinemic (EDIH, EDIR) potential. A detailed list of the specific food items included in each food group for each of these indices can be found in Supplementary Table S1. Weights are available in the original publications describing the creation of the indices (13, 14).

We also considered how two related indices—the empirical lifestyle index for hyperinsulinemia (ELIH) and the empirical lifestyle index for insulin resistance (ELIR), developed using the same methodology as the EDIH and the EDIR—related to prostate cancer outcomes. A lifestyle index for inflammation has not yet been created. Both lifestyle indices included BMI and physical activity in addition to dietary factors (Supplementary Table S1). Details of their development and validation, as well as the weights needed to calculate the indices, are available in the original publication describing the creation and validation of these indices (14). Multivariable models found the ELIH and ELIR were significantly associated with C-peptide and TAG:HDL biomarkers, with the results replicated in validation studies (14).

In this cohort of men with nonmetastatic prostate cancer at diagnosis, the primary outcome was time to prostate cancer progression. Progression was defined as biochemical recurrence, secondary treatment, bone metastases, or death attributed to prostate cancer [prostate cancer–specific mortality (PCSM)]. Biochemical recurrence was defined as two consecutive prostate-specific antigen (PSA) readings ≥0.2 ng/mL after radical prostatectomy or two consecutive PSA levels at least 2.0 ng/mL greater than the postradiation nadir following radiotherapy (31). The date of recurrence was recorded as the date of the second elevated PSA. Secondary treatment was defined as any treatment started ≥6 months following completion of primary treatment. Bone metastases were attributed to prostate cancer if an urologist reported prostate cancer progression to bone or advancement to TNM stage M1b, the patient had a positive bone scan, or the patient underwent radiation to treat bone metastases. Cause of death was determined by the registry data coordinating center and through confirmation by either the vital statistics official death certificate from the state in which the death occurred or by the National Center for Health Statistics National Death Index. Deaths were attributed to prostate cancer if the death certificate included ICD-9 code 185 [(metastatic) malignant neoplasm of prostate] as the primary or secondary cause of death. For men with multiple progression events, the earliest event date was recorded as the date of progression.

Time to progression was measured from the date of completion of the first questionnaire to the date of progression. However, the exact date of progression is unlikely to have occurred on the date of the clinic visit at which it was recorded. To account for this uncertainty, we used an interval rather than a precise date of progression. For men with documented biochemical recurrence, secondary treatment, or bone metastases, the censoring interval was bound by the last normal clinical visit (left limit) and the first clinic visit documenting evidence of progression (right limit). For men with only a progression event of PCSM, the left and right limit were both date of death. Men without documented progression were censored at their last date of follow-up or death (other cause); thus, the right limit was undefined (i.e., censored). Clinical follow-up was last consistently assessed across all CaPSURE sites on January 31, 2019. The 26 men who had a last known clinical follow-up date beyond this date were administratively censored on January 31, 2019.

Pearson's r was used to report correlations between each of the 5 indices. Parametric survival models with a Weibull distribution were used to accommodate interval censoring. We fit survival models using both continuous indices (per 1-SD increase in index) and cohort-specific quintiles. All models were clustered by CaPSURE clinical site, with robust standard errors used to calculate confidence intervals (CI). Simple models were adjusted for age at diagnosis (continuous) and time between diagnosis and first questionnaire (continuous). Fully adjusted models additionally adjusted for T-stage (T1, T2, T3a), Gleason score (<7, 7, >7), and PSA (≤6 ng/mL, >6 to 10 ng/mL, >10 ng/mL) at diagnosis; primary treatment (radical prostatectomy, radiation, hormonal therapy, watchful waiting/active surveillance, other); self-reported race (white, other); total energy intake (continuous, kcal/day); smoking status (current, former, never); family history of prostate cancer in a brother or father (yes/no); total alcohol intake (continuous, servings/day); use of supplements (multivitamins, calcium, vitamin E, or selenium; yes, no); BMI (continuous; models for dietary indices only); and physical activity (continuous, MET-hours/week; models for dietary indices only). We further considered adjustment for height, household income, education, intake of fatty fish and cruciferous vegetables, walking pace, and history of diabetes or heart disease, but estimates were qualitatively unchanged, so these variables were not included in the final models. We examined the goodness of fit of survival models using plots of Cox-Snell residuals. Fully adjusted models were also run using exponential distributions, which produced Cox-Snell residual plots that demonstrated poorer fit than Weibull models; thus, Weibull models were used.

We assessed interactions between each of the dietary indices (EDIP, EDIH, EDIR) and obesity in two ways. First, we created a cross product between each of the indices (continuous) and BMI (<30 vs. ≥30 kg/m²). We then used likelihood ratio tests based on models with and without the interaction terms to look for statistically significant multiplicative interactions. To assess additive interaction, we used the BMI thresholds (<30 vs. ≥30 kg/m²) and a dichotomized version of each index (above and below median) to create a 4-level variable (high index-high BMI, high index-low BMI, low index-high BMI, low index-low BMI) and added it to the fully adjusted model. Low index/low BMI was used as the referent and HR estimates were used to calculate the relative excess risk due to interaction (RERI; ref. 32, 33). The delta method was used to calculate CI that indicated whether RERI results were different from zero (RERI ≠ 0 is evidence of additive interaction; ref. 34).

We evaluated PCSM as our secondary outcome given the small number of PCSM events (n = 73) in this cohort of men initially diagnosed with nonmetastatic disease. For these analyses, we used Cox proportional hazards models rather than parametric survival models because date of death was known. Men who were originally excluded due to documented progression prior to completion of the questionnaire were included in these secondary analyses (as death could not occur prior to completing the questionnaire), resulting in a sample size of 2,447 men. Proportional hazards assumptions were assessed graphically by plotting the scaled Schoenfeld residuals against follow-up time.

We were interested in understanding how deaths due to causes other than prostate cancer (i.e., competing risks) may have impacted our primary results. Methods to address competing risks in the presence of interval censoring are not readily available or accessible. Thus, we ran Cox proportional hazards analyses on our fully adjusted models of progression and compared these results to Fine-Gray analyses accounting for other deaths as a competing risk (35).

Although there was no missingness for any of the indices, missingness in covariates resulted in a loss of events in our fully adjusted models (n = 17). To understand the impact of this missingness on our primary results, we performed a sensitivity analysis using multiple imputation to handle missing data (36). We performed multiple imputation via chained equations using the chained command in Stata to first generate 25 imputed datasets. We then fit survival models across all 25 imputed datasets and pooled the results using Rubin's Rules (37). Our imputed model included all variables without missingness (EDIP, EDIH, ELIH, EDIR, ELIR, BMI, physical activity, CaPSURE clinical site, age at diagnosis, vital status, total energy intake, days in follow-up, race, clinical T stage, and family history of prostate cancer) and variables with incomplete values (diagnostic PSA and Gleason score, smoking status, supplement use, total alcohol intake, and primary treatment).

All statistical analyses were performed using Stata version 17 (StataCorp, College Station, TX). A two-sided alpha level of 0.05 was used to assess statistical significance.

Data can be made available on request.

Participant characteristics by quintile of the inflammatory and insulinemic dietary and lifestyle indices are shown in Tables 1 and 2. The participants had a mean (SD) age of 64.4 (7.9) years at diagnosis, and most (n = 1,953; 95%) identified as white race. These characteristics were similar to the larger CaPSURE cohort [mean age: 66.0 (8.6) years; 86% identified as white race]. Characteristics were fairly balanced across quintiles of each index, although men consuming more inflammatory and insulinemic diets tended to have higher BMI and lower levels of physical activity. This was also true for men with more insulinemic lifestyles, with a more pronounced increase in BMI and decrease in physical activity, as expected (both are components of these indices). Correlations between the indices are shown in Supplementary Table S2.

During a median follow-up of 6.4 years [interquartile range (IQR), 1.3–12.7] after completion of the questionnaire, 192 progression events were documented, including 168 (88%) biochemical recurrences, 7 (4%) bone metastases, and 17 (9%) deaths related to prostate cancer as the first recorded event (a total of 73 men had documented PCSM, most with another progression event prior to PCSM). Secondary treatment did not account for any of the progression events.

Participants with higher inflammatory diet indices (EDIP) had an increased risk of prostate cancer progression (HR_{per 1-SD}, 1.27; 95% CI, 1.17–1.37), amounting to a 2.61-fold (95% CI, 1.75–3.90; P_trend < 0.01) higher risk in those in the highest versus the lowest quintile of EDIP (Table 3). Those with more insulinemic diets (EDIH) also had a higher risk of progression (HR_{per 1-SD}, 1.24; 95% CI, 1.05–1.46), amounting to a 1.63-fold (95% CI, 0.93–2.86; P_trend = 0.05) higher risk among those in the highest versus lowest quintile. The hyperinsulinemic lifestyle index (ELIH) was similarly associated with progression (HR_{per 1-SD}, 1.34; 95% CI, 1.17–1.54), with a 2.81-fold (95% CI, 1.78–4.43; P_trend < 0.01) higher risk of progression among those in the highest versus lowest quintile of the index. There was suggestive evidence that the insulin resistance dietary index (EDIR) was associated with prostate cancer progression (HR_{per 1-SD}, 1.22; 95% CI, 1.00–1.48), though results from the models with EDIR modeled as quintiles were not statistically significant (HR_{Q5 vs. Q1}, 1.38; 95% CI, 0.62–3.11; P_trend = 0.45). The insulin resistance lifestyle index (ELIR) was statistically significantly associated with a higher risk of prostate cancer progression (HR_{per 1-SD}, 1.36; 95% CI, 1.12–1.64), reflecting a 2.43-fold (95% CI, 1.45–4.07; P_trend < 0.01) higher risk of progression for those with the highest versus lowest quintile of the index (Table 3).

There was no convincing evidence of associations with PCSM (Table 3), though power for these analyses was limited. There was also no evidence of interaction between any of the dietary indices and BMI.

Results from the Cox proportional hazards models for progression were very similar to those from the Parametric (Weibull) survival models, and there was no evidence that competing events impacted the results (Supplementary Table S3). Multiple imputation resulted in 2,053 complete cases across all covariates and retainment of all 192 events in the multivariable models (Supplementary Table S4). The results were qualitatively unchanged from the primary analysis. The supplemental material also includes results from the simple (i.e., not fully adjusted) models (Supplementary Table S5).

In these analyses, we evaluated associations of three dietary (EDIP, EDIH, EDIR) and two lifestyle (ELIH, ELIR) indices – previously developed to estimate concentrations of biomarkers for the underlying inflammatory and insulin pathways – with prostate cancer progression and PCSM. Findings from this study suggest that diets with high inflammatory or insulinemic potential following a prostate cancer diagnosis are associated with a 2.61-fold and 1.63-fold higher risk of prostate cancer progression, respectively, for those in the highest versus lowest quintiles. The evidence was weaker, but still consistent with a positive association, for diets promoting insulin resistance.

The hyperinsulinemic and insulin resistance lifestyle indices also demonstrated strong associations with prostate cancer progression. Individuals in the highest versus lowest quintile of the ELIH and ELIR had a 2.8-fold and 2.4-fold higher risk of progression, respectively. These results are consistent with prior work demonstrating that the correlation between the lifestyle indices and circulating biomarkers was more than twice the correlation observed with the diet-only indices (14). These findings are also consistent with the World Cancer Research Fund/American Institute for Cancer Research report, which found adiposity to be the single most consistent factor predisposing men to higher risk of fatal prostate cancer (38). Therefore, lifestyle changes that include more physical activity and achieving a healthy weight, in addition to low insulinemic and inflammatory diets, may lower risk of progression.

Although, to our knowledge, no study has examined these dietary and lifestyle indices in men with prostate cancer, these findings are consistent with our current understanding of the role of the inflammation and insulin pathways in promoting cancer growth and development. This report adds to our understanding of how these pathways may promote prostate cancer progression. Specifically, insulin is a potent growth factor that promotes cell metabolism and mitogenic processes, and cancer cells have been shown to have a disproportionally higher expression of insulin receptors than normal cells (39, 40). The EDIH has also been associated with a higher risk of prostate cancer development among previously disease-free men (20, 41). Thus, it is plausible that higher levels of circulating insulin would promote prostate cancer progression (5). Inflammation can also act to promote insulin production (5), and has been independently linked to prostate cancer risk (42). For example, IL6, a prominent inflammatory biomarker, has been shown to promote proliferation of prostate cells and inhibit cell death, and may be involved in the transition to metastatic disease (7). The EDIP has also been associated with increased risk of incident lethal prostate cancer among men under 65 years of age (20). Thus, it is also plausible that diets promoting inflammatory processes would promote cancer progression.

We observed a correlation between all three dietary indices, which is not surprising given the inflammatory and insulin pathways are interrelated (5, 43). Indeed, prior research found that the EDIP was associated with biomarkers of hyperinsulinemia and that the EDIH was associated with biomarkers of inflammation (41, 44). Similarly, although the insulin resistance and hyperinsulinemic indices were developed to predict different biomarkers (14), hyperinsulinemia is a consequence of prolonged insulin resistance (5). Recent work found that the EDIH is predictive of both insulin secretion and insulin resistance (45), which may explain why the EDIR was not as strongly associated with prostate cancer progression as the EDIH in these analyses.

We did not find statistically significant evidence of associations between any of the indices and PCSM. While other mechanisms for the lack of associations cannot be ruled out, our results may reflect the relatively small number of cause-specific deaths in this cohort of men diagnosed with nonmetastatic prostate cancer. Further research is needed to understand whether the relevant biological mechanisms are associated with PCSM among men with prostate cancer.

The inflammatory, hyperinsulinemic, and insulin resistance dietary indices were developed to predict inflammatory and insulinemic biomarkers associated with whole diet, and were not developed specific to any type of cancer. Recently, these indices have been associated with colorectal cancer incidence and mortality (16, 46), highlighting the role of inflammatory and insulin pathways across cancers (5). Readers should focus on the importance of tailoring whole diets following a prostate cancer diagnosis to collectively minimize consumption of inflammatory foods and those known to overstimulate insulin secretion, and avoid focusing on the role of any given component of these indices.

There are several limitations of our study. Measurement error is a known limitation of self-reported diet data. However, the FFQ used in this proposal has been reported to be valid and reproducible and to perform well compared with 1-week diet records and in repeated FFQs 1-year apart (23). Men who self-select to participate in diet and lifestyle studies may also be relatively healthy compared with prostate cancer survivors who opt out. Men in our study predominately identified as white race, and 77% reported having at least some college-level education. While the generalizability of our results may thus be limited, the dietary indices have been applied in multiethnic samples and found to predict risk of developing type 2 diabetes with heightened risk among African American and Hispanic women compared with European American women (47). In addition, although multiple imputation was used to address missingness in covariates, these methods rely on the assumption that data are missing at random, which cannot readily be assessed.

In conclusion, in this cohort of men diagnosed with nonmetastatic prostate cancer, diets with higher inflammatory and insulinemic potential were associated with higher risk of prostate cancer progression. Insulinemic lifestyle indices that included diet, physical activity, and BMI, were also associated with risk of disease progression. These findings add to the evidence that inflammation and insulin pathways influence prostate cancer progression and suggest that modifiable health habits may improve prostate cancer clinical outcomes.

C.S. Langlais reports grants from NCI; and grants from UCSF Prostate Cancer Program during the conduct of the study. E.L. Van Blarigan reports grants from NIH during the conduct of the study. S.A. Kenfield reports personal fees and nonfinancial support from Fellow Health, Inc. outside the submitted work. J. Neuhaus reports grants from NIH during the conduct of the study. P. Carroll reports personal fees from Alessa, Nutcracker Therapeutics, Francis Medical; and personal fees from Progenics, Exact Biosciences, Biopharma Communications outside the submitted work. J.M. Chan reports grants from Cancer League Foundation and grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. JMC is the Steven & Christine Burd-Safeway Distinguished Professor.

C.S. Langlais: Conceptualization, formal analysis, funding acquisition, methodology, writing–original draft, writing–review and editing. R.E. Graff: Conceptualization, funding acquisition, methodology, writing–review and editing. E.L. Van Blarigan: Conceptualization, funding acquisition, methodology, writing–review and editing. S.A. Kenfield: Conceptualization, funding acquisition, methodology, writing–review and editing. J. Neuhaus: Methodology, writing–review and editing. F.K. Tabung: Methodology, writing–review and editing. J.E. Cowan: Data curation, writing–review and editing. J.M. Broering: Data curation, writing–review and editing. P. Carroll: Funding acquisition, writing–review and editing. J.M. Chan: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.

Research reported in this publication was supported by the University of California, San Francisco Prostate Cancer Program, the NCI of the NIH: F31CA247093 (to C.S. Langlais) and K07CA197077 (to E.L. Van Blarigan), and the UCSF Goldberg-Benioff Program in Translational Cancer Biology. J.M. Chan is the Steven & Christine Burd-Safeway Distinguished Professor. S.A. Kenfield is the Helen Diller Family Chair in Population Science of Urologic Cancer. R.E. Graff is supported by a Young Investigator Award from the Prostate Cancer Foundation. CaPSURE was funded by the United States Department of Defense Prostate Cancer Research Program (W81XWH-13-2-0074 and W81XWH-04-1-0850).

The authors wish to thank the CaPSURE participants for making this research possible and the research team who diligently worked on ensuring data quality. The authors would also like to thank all funders of this work.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.