Postdiagnosis Body Mass Index, Weight Change, and Mortality From Prostate Cancer, Cardiovascular Disease, and All Causes Among Survivors of Nonmetastatic Prostate Cancer

PURPOSE

To investigate the association of postdiagnosis body mass index (BMI) and weight change with prostate cancer–specific mortality (PCSM), cardiovascular disease–related mortality (CVDM), and all-cause mortality among survivors of nonmetastatic prostate cancer.

METHODS

Men in the Cancer Prevention Study II Nutrition Cohort diagnosed with nonmetastatic prostate cancer between 1992 and 2013 were followed for mortality through December 2016. Current weight was self-reported on follow-up questionnaires approximately every 2 years. Postdiagnosis BMI was obtained from the first survey completed 1 to < 6 years after diagnosis. Weight change was the difference in weight between the first and second postdiagnosis surveys. Deaths occurring within 4 years of the follow-up were excluded to reduce bias from reverse causation. Analyses of BMI and weight change included 8,330 and 6,942 participants, respectively.

RESULTS

Postdiagnosis BMI analyses included 3,855 deaths from all causes (PCSM, n = 500; CVDM, n = 1,155). Using Cox proportional hazards models, hazard ratios (HRs) associated with postdiagnosis obesity (BMI ≥ 30 kg/m²) compared with healthy weight (BMI 18.5 to < 25.0 kg/m²) were 1.28 for PCSM (95% CI, 0.96 to 1.67), 1.24 for CVDM (95% CI, 1.03 to 1.49), and 1.23 for all-cause mortality (95% CI, 1.11 to 1.35). Weight gain analyses included 2,973 deaths (PCSM, n = 375; CVDM, n = 881). Postdiagnosis weight gain (> 5% of body weight), compared with stable weight (± < 3%), was associated with a higher risk of PCSM (HR, 1.65; 95% CI, 1.21 to 2.25) and all-cause mortality (HR, 1.27; 95% CI, 1.12 to 1.45) but not CVDM.

CONCLUSION

Results suggest that among survivors of nonmetastatic prostate cancer with largely localized disease, postdiagnosis obesity is associated with higher CVDM and all-cause mortality, and possibly higher PCSM, and that postdiagnosis weight gain may be associated with a higher mortality as a result of all causes and prostate cancer.

INTRODUCTION

In the United States, prostate cancer is the most commonly diagnosed cancer among men,¹ and survivors of prostate cancer may continue to experience excess mortality for up to 15 years after diagnosis.² Current evidence supports the link between obesity and risk of advanced prostate cancer,³ potentially through alterations in hormones, adipocytes, and inflammatory factors.^4,5 As these factors are also implicated in tumor progression pathways, it is important to understand the potential consequences of obesity and weight gain on long-term survival after a prostate cancer diagnosis.

Some previous studies,^6-11 although not all,¹² suggest that obesity measured before or during the first year after a prostate cancer diagnosis may be associated with higher prostate cancer–specific mortality (PCSM). However, survivors cannot change prediagnosis behaviors, and the initial hardships of a cancer diagnosis and primary treatment may influence body weight¹³ and make efforts toward modifying weight around the time of diagnosis difficult. Postdiagnosis (defined here as > 12 months after cancer diagnosis) obesity and weight change may be more easily modified and relevant to inform recommendations for survivors of cancer.

Only 2 previous studies investigated the effect of postdiagnosis body mass index (BMI) or weight gain on mortality among survivors of prostate cancer, the findings from which are conflicting.^6,14 Therefore, the goals of this study were to investigate associations of postdiagnosis BMI and weight change with cause-specific and all-cause mortality among men diagnosed with nonmetastatic prostate cancer.

METHODS

Study Population

Among 86,402 male participants in the Cancer Prevention Study (CPS) II Nutrition Cohort, we identified 11,788 patients with prostate cancer diagnosed between 1992 and 2013. The CPS-II Nutrition Cohort is a prospective study of cancer incidence and mortality, initiated in 1992/1993, that enrolled participants from 21 US states and is a subgroup of the larger CPS-II mortality cohort initiated in 1982 by the American Cancer Society.¹⁵ At baseline, a 10-page survey was mailed to all CPS-II Nutrition Cohort participants to collect information on demographics, medical conditions, lifestyle, and other factors. Follow-up surveys were sent to participants in 1997 and every 2 years thereafter to ascertain newly diagnosed cancers. All aspects of the CPS-II Nutrition Cohort are reviewed and approved by the Emory University Institutional Review Board.

Most prostate cancers were self-reported and subsequently verified through medical records (93.8%) or linkage with state cancer registries (6.1%). Additional cancers were identified during the process of verifying another cancer (n = 5). Two analytic cohorts, the postdiagnosis BMI (n = 8,330) and weight change (n = 6,942) cohorts, were constructed to maximize sample size because fewer men completed 2 postdiagnosis surveys. Men were excluded based on the criteria shown in Figure 1. To minimize reverse causation bias (ie, preexisting disease leading to both weight loss and mortality), person-time and deaths occurring within 4 years of completing the postdiagnosis surveys were excluded from all analyses, similar to previous studies.^16-18 A comparison of eligible participants included and excluded in the analyses is described in the Data Supplement.

FIG 1.

Exclusion flowchart among male participants diagnosed with prostate cancer from 1992 to 2013 in the Cancer Prevention Study II Nutrition Cohort. (^a) Exclusion as a result of implausible diagnosis dates included individuals with self-reported diagnosis dates that were > 6 months before the diagnosis date obtained from medical records or cancer registries and those diagnosed at death. (^b) Body mass index (BMI) < 15 or > 60 kg/m².

Assessment of BMI and Weight Change

Postdiagnosis BMI was based on height from the 1982 baseline survey in the original CPS-II cohort and weight from the first questionnaire completed 1 to < 6 years after diagnosis and categorized as healthy weight (BMI, 18.5 to < 25 kg/m²; referent group), overweight (BMI, 25 to < 30 kg/m²), and obese (BMI, ≥ 30 kg/m²). Weight change was the difference between postdiagnosis weight and the weight reported on the next biennial survey. For participants who did not return the next biennial survey or did not report their weight (5%), we calculated weight change based on weight reported on the earliest subsequent biennial survey (up to 4 surveys after). Relative weight change was calculated as

$$\left(\frac{weigh{t}_{2nd postdiagnosis survey}-weigh{t}_{1st postdiagnosis survey}}{weigh{t}_{1st postdiagnosis survey}}\right)\times 100\%$$

and categorized as moderate gain (≥ 5%), small gain (3% to < 5%), maintenance (± < 3%), small loss (3% to < 5%), and moderate loss (≥ 5%), according to expert recommendation.¹⁹ We also calculated absolute weight change and categorized it as follows: gain of ≥ 10 lb, gain of 5 to < 10 lb, maintenance (± < 5 lb), loss of 5 to < 10 lb, and loss of ≥ 10 lb.

Assessment of Outcomes

The primary outcome of interest was PCSM, defined as deaths for which the underlying cause was listed as prostate cancer, according to International Classification of Diseases 9th (ICD-9) and 10th Revisions (ICD-10) codes (ICD-9: 185; ICD-10: C61). Secondary outcomes of interest included cardiovascular disease–related mortality (CVDM; ICD-9: 390-459; ICD-10: I00-I99) and deaths as a result of all causes. Vital status, cause of death, and date of death were ascertained through linkage with the National Death Index, updated through December 31, 2016.

Statistical Analyses

Cumulative incidence functions²⁰ for PCSM, CVDM, and all-cause mortality were produced within strata of postdiagnosis BMI and relative weight change. Cox proportional hazards regression models were used to produce cause-specific hazard ratios (HRs) and 95% CIs to estimate associations of postdiagnosis BMI and weight change on mortality outcomes.

To mitigate the potential for bias as a result of reverse causation, follow-up started 4 years after completing the first postdiagnosis survey in BMI analyses and 4 years after completing the second postdiagnosis survey in weight change analyses using delayed-entry models. For all analyses, follow-up ended on the death date or December 31, 2016, whichever came first.

Postdiagnosis BMI multivariable models controlled for age, education, smoking status, physical activity, American Joint Committee on Cancer primary tumor (T) category, Gleason score, initial treatment, and year of diagnosis. Postdiagnosis weight change models additionally adjusted for the first postdiagnosis BMI. All models adjusted for age by stratifying on single year of age at diagnosis. We used multiple imputation procedures to address missing covariate data (see Data Supplement).

For postdiagnosis BMI analyses, we considered an interaction with risk of disease progression category based on the National Comprehensive Cancer Network (NCCN) guidelines, with T1-2 tumors that have Gleason scores ≤ 7 classified as lower-risk tumors (NCCN low- and intermediate-risk groups) and T3-4 tumors or tumors with Gleason scores ≥ 8 classified as high-risk tumors.²¹ We were unable to consider interactions with weight change as a result of the limited prostate cancer deaths within strata.

We conducted supplemental analyses to address concerns regarding competing causes of death (eg, nonprostate cancer deaths in PCSM models); the effects of hormone therapy on weight; overcompensation for reverse causation through our exclusion of follow-up within 4 years of survey completion; and the influence of prediagnosis BMI on outcomes. See Data Supplement for additional details on all supplemental analyses, covariates, and imputation procedures. Statistical analyses were performed using SAS (version 9.4; SAS Institute, Cary, NC) and R (version 3.5.3; R Foundation, Vienna, Austria) software.

RESULTS

Participant characteristics are presented by postdiagnosis BMI (Table 1) and weight change category (Data Supplement). In our postdiagnosis BMI cohort, approximately 36% of men were healthy weight, 49% were overweight, and 16% were obese. Obese men were less likely to have graduated from college or engage in physical activity than healthy weight men.

TABLE 1.

Participant Characteristics by Postdiagnosis BMI Among Men Diagnosed with Nonmetastatic Prostate Cancer in the CPS-II Nutrition Cohort

We observed 3,855 deaths in postdiagnosis BMI analyses and 2,973 deaths in weight change analyses (500 and 375 prostate cancer–specific deaths, respectively). In the postdiagnosis BMI cohort, the median time from diagnosis to survey completion was 2.2 years (interquartile range [IQR], 1.2 years). In the weight change cohort, the median time from diagnosis to completion of the first postdiagnosis survey was 2.2 years (IQR, 1.2 years), and the second postdiagnosis survey was completed a median of 2.0 years (IQR, 0.2 years) after the first postdiagnosis survey. The median follow-up time was 7.3 years (IQR, 7.7 years) and 5.7 years (IQR, 6.1 years) in postdiagnosis BMI and weight change cohorts, respectively. Graphs depicting the cumulative incidence of PCSM, CVDM, and all-cause mortality over time within strata of postdiagnosis BMI and weight change, with corresponding estimates, are provided in the Data Supplement.

Postdiagnosis BMI

Results from multivariable cause-specific models examining associations of postdiagnosis BMI with all mortality outcomes are presented overall and by risk of disease progression category in Table 2. Overall, compared with healthy weight men, obese men had a higher but nonsignificant hazard of PCSM (HR, 1.28; 95% CI, 0.96 to 1.67) and a higher hazard of CVDM (HR, 1.24; 95% CI, 1.03 to 1.49) and all-cause mortality (HR, 1.23; 95% CI, 1.11 to 1.35). Overweight men had a higher hazard of PCSM (HR, 1.23; 95% CI, 1.00 to 1.50), but not CVDM or all-cause mortality. The association of BMI with mortality outcomes seemed to be somewhat stronger among those diagnosed with lower-risk tumors, particularly for PCSM, although the interaction between risk category and continuous BMI was not statistically significant (P = .63). For example, postdiagnosis obesity was associated with a higher hazard of PCSM among men diagnosed with lower-risk tumors (HR, 1.58; 95% CI, 1.13 to 2.22) but not high-risk tumors (HR, 1.00; 95% CI, 0.63 to 1.58).

TABLE 2.

Cox Proportional HRs and 95% CIs for the Association of Postdiagnosis BMI With Mortality as Result of Prostate Cancer, Cardiovascular Disease, and All Causes, for the Total Cohort and Stratified by Risk of Disease Progression Category

Weight Change

Results from multivariable cause-specific models of the association of weight change with mortality outcomes are listed in Table 3. Compared with men who maintained their postdiagnosis weight, men who gained > 5% had a higher hazard of PCSM (HR, 1.65; 95% CI, 1.21 to 2.25), whereas all other weight change categories had similar hazards. The hazard of all-cause mortality was higher among men who gained > 5% (HR, 1.27; 95% CI, 1.12 to 1.45), lost 3%-5% (HR, 1.15; 95% CI, 1.02 to 1.31), and lost > 5% (HR, 1.30; 95% CI, 1.16 to 1.46), but not among men who gained 3%-5%. No associations between weight change and CVDM were observed. Models examining absolute weight change yielded similar results.

TABLE 3.

Cox Proportional HRs and 95% CIs for the Associations of Postdiagnosis Body Weight Change With Mortality as a Result of Prostate Cancer, Cardiovascular Disease, and All Causes

Supplemental Analyses

Supplemental analyses accounting for the influence of other causes of death on PCSM and CVDM yielded similar results to primary analyses (Data Supplement). Results were similar after excluding men initially treated with hormone therapy, except that associations of obesity and weight gain with PCSM were somewhat stronger, although less precise as a result of smaller sample size (Data Supplement). After excluding follow-up occurring within 2 years of survey completion, results were similar except that associations of weight loss with CVDM and all-cause mortality were stronger, suggesting presence of reverse causation bias (Data Supplement). Compared with men who maintained a healthy weight both before and after diagnosis, men who were obese at both time points had a higher hazard of CVDM and all-cause mortality but not PCSM (Data Supplement). A higher hazard of PCSM was observed among men with a combined prediagnosis BMI category of overweight and a postdiagnosis BMI category of obese (HR, 1.69; 95% CI, 1.12 to 2.57). See Data Supplement for additional details and further discussion of results.

DISCUSSION

In this large cohort study of US men diagnosed with nonmetastatic prostate cancer, which was localized in the vast majority of patients, we found that postdiagnosis obesity was associated with a higher hazard of all-cause mortality and CVDM and possibly an elevated hazard of PCSM, although the latter was statistically nonsignificant. In addition, we found higher overall mortality and PCSM among men who gained a modest to high amount of weight during the interval between their first postdiagnosis questionnaire, typically 2 years after diagnosis, and their second postdiagnosis questionnaire, typically 4 years after diagnosis.

Our observation of a higher overall mortality and marginally higher hazard of PCSM, although it included the null, among postdiagnosis obese men is in contrast to the only previous study that examined BMI ≥ 1 year after prostate cancer diagnosis and survival.¹⁴ Farris et al¹⁴ found that among 829 Canadian survivors of prostate cancer, obesity, assessed approximately 2-3 years after diagnosis, was not associated with PCSM or all-cause mortality. The observed higher hazard of PCSM among postdiagnosis obese survivors of prostate cancer in our study was not statistically significant and inconsistent across analyses and could be a result of chance. However, obesity is an accepted risk factor for all-cause mortality in the general population,²² although there is some debate on whether this association attenuates with age,²³ and there are several possible explanations for the null findings by Farris et al,¹⁴ including limited statistical power; inclusion of men diagnosed with higher-risk tumors (all were diagnosed with ≥ T2 cancers), a subgroup in which we observed no association with PCSM, potentially because of selection bias (see Data Supplement for further discussion); and bias as a result of reverse causation through inclusion of patients with metastatic disease.

Several other studies have examined the association of BMI, measured < 1 year after prostate cancer diagnosis, with survival and largely suggest obesity is associated with higher PCSM and all-cause mortality.^6-9,12,24,25 The largest of these studies, conducted by Chalfin et al⁷ among 11,152 men (PCSM, n = 245) who underwent radical prostatectomies at a single tertiary referral center, had similar findings to our study. The remaining studies were mostly conducted in clinical populations, with sample sizes ranging from 945 to 7,274 patients (61-220 prostate cancer deaths), and most studies,^6,8,9,24 although not all,^12,25 suggested a positive relationship between obesity and PCSM. The link between obesity and PCSM is further supported by evidence suggesting that obesity may be associated with an increased risk of biochemical recurrence,²⁶ which often precedes PCSM, although there is some debate on whether residual confounding by disease severity could account for positive results.²⁷

In our study, men who experienced modest to high weight gain (> 5% of their body weight or > 10 lb) after a prostate cancer diagnosis were more likely to die of prostate cancer and all causes than men who maintained their weight. Similarly, Bonn et al⁶ found that survivors of prostate cancer who gained > 5% of their weight (from diagnosis to a median of 7.3 years later) were almost twice as likely to die of prostate cancer and somewhat more likely to die of all causes, although the latter was not statistically significant. However, estimates were imprecise as a result of few deaths (PCSM, n = 96) and were based on retrospective report of weight change. In contrast, Farris et al¹⁴ observed no association between weight gain ≥ 2.6 lb (from diagnosis to 2-3 years after diagnosis) and PCSM but did not report the potential effects of greater amounts of weight gain, as was done in the current study. Two previous studies reported no relationship between prediagnosis adult weight gain and lethal prostate cancer,^14,28 although one reported positive findings among never-smokers.²⁸ This is in contrast to our findings, and differences may be a result of the extent of collider bias,²⁹ the timing in which weight gain occurred, or residual confounding by smoking, disease severity, or treatment. In our study, we attempted to mitigate the potential for collider bias to influence weight change results by examining postdiagnosis weight change and controlling for the first postdiagnosis BMI measure (Data Supplement). Evidence suggesting that weight gain before diagnosis to approximately 1 year after diagnosis may be associated with an increased risk of prostate cancer recurrence³⁰ supports our current findings. Our findings regarding higher risk of all-cause mortality, but not PCSM, among men who lost weight are supported by similar previous studies.^6,14 Although weight loss could serve as a useful predictor, a causal interpretation should not be drawn because reverse causation as a result of underlying diseases likely biased the results.

Our findings should be considered in the context of our study limitations. First, despite efforts to reduce bias as a result of reverse causation through exclusion of follow-up ending within 4 years of completing the postdiagnosis surveys, this bias cannot be ruled out, although it is expected to bias results downward and is unlikely to account for any positive findings. Second, although restricting the study population to those who survived at least 4 years after the postdiagnosis survey reduces the potential for reverse causation to bias results, it could induce selection bias in the presence of uncontrolled risk factors for the outcome.²⁹ However, if our causal diagram is correct (Data Supplement), this bias is unlikely to affect our weight change results, after controlling for the first postdiagnosis BMI measure. Third, BMI based on self-reported data may have been misclassified. Bias as a result of misclassification of BMI in this study is likely nondifferential because of the study’s prospective design, and results are expected to be biased toward the null. Although BMI can be a poor proxy for excess body fatness, weight gain largely reflects an increase in body fat, given that height likely remains constant in our population. Most survivors of prostate cancer included in our study were elderly and white; therefore, results may not be generalizable to a nonwhite or younger population. In addition, we only had data on initial treatment and lacked details on completion dates and whether the patient received additional treatment, limiting our ability to better control for confounding. For example, if patients received hormone therapy after initial treatment as a result of disease recurrence, they would have a higher risk of both dying and gaining weight, potentially biasing results. Further research will be needed to determine whether there is an association between postdiagnosis weight gain and PCSM, independent of secondary treatments. Finally, because of the deterministic relationship between BMI and weight change, it is difficult, if not impossible, to separate out their direct effects.

The current study has several noteworthy strengths. To our knowledge, our study was based on the largest number of prostate cancer deaths (n = 500) to date and the first to examine the association of postdiagnosis BMI and weight change with CVDM among survivors of prostate cancer. Our study sample was selected from a longitudinal cohort that enabled us to examine weight change prospectively. Multiple imputation procedures were used to address missing data, and we were able to consider lifestyle factors such as physical activity and smoking status. Finally, we conducted several sensitivity analyses indicating that the associations of postdiagnosis obesity with CVDM and all-cause mortality, as well as associations of weight gain with PCSM and all-cause mortality, were robust to several assumptions.

Our results suggest that among survivors of nonmetastatic prostate cancer, postdiagnosis obesity may be associated with higher CVDM and all-cause mortality and that postdiagnosis weight gain may be associated with a higher mortality as a result of all causes and prostate cancer. Survivors of prostate cancer are advised to maintain a healthy weight and avoid weight gain,^31,32 and our findings provide additional evidence to follow these recommendations. This may be especially important given evidence that survivors of prostate cancer may be more prone to weight gain compared with men in the general population.³³ Our results do not support promotion of weight loss among survivors of prostate cancer, although the observed positive association between weight loss and all-cause mortality is likely a result of underlying disease rather than a true causal relationship. Future studies are needed to determine whether intentional weight loss provides health benefits among overweight and obese survivors of prostate cancer. Clinicians should be vigilant about identifying moderate weight loss or gain in survivors of prostate cancer because both have poor prognostic implications.

AUTHOR CONTRIBUTIONS

Conception and design: Alyssa N. Troeschel, Terryl J. Hartman, Ying Wang

Collection and assembly of data: Alyssa N. Troeschel, Eric J. Jacobs, Ying Wang

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Postdiagnosis Body Mass Index, Weight Change, and Mortality From Prostate Cancer, Cardiovascular Disease, and All Causes Among Survivors of Nonmetastatic Prostate Cancer

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).

W. Dana Flanders

Employment: Epidemiologic Research & Methods

Consulting or Advisory Role: Epidemiologic Research & Methods

Travel, Accommodations, Expenses: Epidemiologic Research & Methods

No other potential conflicts of interest were reported.