Abstract
Background:
Prostate cancer is ubiquitous in older men; differential screening patterns and variations in biopsy recommendations and acceptance will affect which man is diagnosed and, therefore, evaluation of cancer risk factors. We describe a statistical method to reduce prostate cancer detection bias among African American (n=3398) and Non-Hispanic White men (n=22,673) who participated in the Selenium and Vitamin E Cancer Prevention trial (SELECT) and revisit a previously reported association between race, obesity and prostate cancer risk.
Methods:
For men with screening values suggesting prostate cancer but in whom biopsy was not performed, the Prostate Cancer Prevention Trial Risk Calculator was used to estimate probability of prostate cancer. Associations of body mass index (BMI) and race with incident prostate cancer were compared for observed versus imputation-enhanced outcomes using incident density ratios.
Results:
Accounting for differential biopsy assessment, the previously reported positive linear trend between BMI and prostate cancer in African American men was not observed; no BMI association was found among Non-Hispanic White men.
Conclusions:
Differential disease classification among men who may be recommended to undergo and then consider whether to accept a prostate biopsy leads to inaccurate identification of prostate cancer risk factors. Imputing a man’s prostate cancer status reduces detection bias. Covariate adjustment does not address the problem of outcome misclassification. Cohorts evaluating incident prostate cancer should collect longitudinal screening and biopsy data to adjust for this potential bias.
Keywords: risk factor, prevention, prostate cancer, differential misclassification
INTRODUCTION
The identification of risk factors and preventive strategies for prostate cancer by epidemiologic and other studies has a major impact, affecting how clinicians select patients for prostate biopsy and treatment and what steps patients take to reduce their disease risk. As detectable prostate cancer is almost ubiquitous in aging men, illustrated by the 15% risk of prostate cancer in men over age 55 undergoing a 6-core prostate biopsy with low PSA and a normal digital rectal exam, biases in selecting whom to biopsy could have profound and potentially adverse implications on populations of men at risk for prostate cancer.1,2 While seemingly a trite concept, as prostate cancer is so common and asymptomatic during a decade or more of sojourn,3,4 for a man to receive a prostate cancer diagnosis, he must be screened with PSA, recommended to undergo biopsy, and agree to have a biopsy. We previously observed that a broad range of factors affected patients’ likelihood of biopsy.5 Younger or married men, those with a family history of prostate cancer or with BPH, were more likely to undergo biopsy. A higher BMI, diabetes, or previous or current smokers were less likely to undergo biopsy. When accounted for, using the unique resource of the PCPT in which all men were recommended to undergo biopsy, we found that the magnitude and, in some cases, the direction of risk factors were significantly changed. An important example was aspirin that, without adjustment for biopsy, was associated with a 34% reduction of prostate cancer risk; by comparison, after adjustment, prostate cancer risk was actually increased by 20–23%. Following our observation, the 19,114-subject ASPREE randomized trial confirmed a 31% increased risk of cancer death in subjects receiving aspirin.6 Unidentified confounders in observational cohorts can make it difficult to obtain unbiased estimates of risk factor – cancer associations. In a recent comparison, investigators found no significant correlation between the hazard ratio estimates reported by observational studies and randomized trials.7
In 2015, Barrington et al examined the joint BMI and race association in the Selenium and Vitamin E Cancer Prevention Trial (SELECT), a study evaluating vitamin E and selenium for cancer prevention. No association between BMI and prostate cancer incidence was found in non-Hispanic white (NHW) men, but a positive linear trend for BMI was seen in African American (AA) participants (NWH trend p=0.63, AA trend p=0.03)8. While investigators acknowledged possible bias due to screening and election of biopsy, possible direction or magnitude was not discussed.
Because prostate cancer is highly-prevalent in aging men, if any given risk factor increases the likelihood of biopsy, even if there is no true association, men with the risk factor will appear to have increased risk simply due to the bias caused by a greater likelihood of biopsy.
Given these observations, we re-examined the relationship between race and BMI and risk of prostate cancer in SELECT, to assess whether accounting for differential biopsy referral and acceptance rates changed associations between BMI, race and prostate cancer as earlier reported.
METHODS
Details of the SELECT trial design and results have been previously published.9,10,11 From 2001–2004, 35,533 men 55 years or older (≥ 50 for AA men) were recruited from 427 study sites in the US, Canada and Puerto Rico and randomized to selenium, vitamin E, the combination or neither in a double-blind fashion. Eligibility included a PSA less than 4 ng/ml and a normal DRE. At the time of study closure, the median follow-up was 5.5 years. Participants underwent annual PSA and DRE screening and received biopsy recommendations according to local standard of care and the participant’s preference. The primary endpoint for SELECT was time from randomization to prostate cancer diagnosis.
We first evaluated whether receipt of a biopsy (regardless of outcome) varied by race and BMI. Using a proportional hazards model, we estimated the hazard ratio (HR) for each BMI and race group compared to NHW lowest BMI group, the same reference group as the original report, in terms of time to first on-study biopsy. The model adjusted for study arm, age, family history of prostate cancer and longitudinal PSA and DRE as time-dependent covariates. Men without a biopsy were censored at last contact.
Next, we evaluated the association of BMI and race with prostate cancer accounting for differential biopsy patterns. Significant differences between the originally reported associations of BMI and race and prostate cancer based on observed biopsies compared to those that mimic uniform biopsy recommendations and acceptance would suggest that detection bias is present.
For these analyses, we assumed a man who had a negative biopsy during SELECT was cancer-free for the remainder of the trial. For those without a prostate cancer diagnosis or negative biopsy, we used a man’s estimated probability of prostate cancer under two scenarios to account for lack of biopsy. Because AA men could enter SELECT at 50 years versus 55 for NHW, we adjusted for age as the single covariate in our models. Although more complex models could be considered, we wanted to keep the models relatively simple to focus on the comparisons across approaches.
To estimate probability of prostate cancer for those without a biopsy we used the PCPT risk calculator.12 In that trial, from 1994–1997, 18,880 eligible men were randomized to receive finasteride or placebo. The primary endpoint was period prevalence of prostate cancer at 7 years. The design13, conduct14 and primary results15 have been published. Men who had not been diagnosed with prostate cancer during the trial were recommended to undergo an end-of-study prostate biopsy regardless of PSA or DRE results. Due to this unique feature, differential biopsy detection bias was minimized15. We used the risk calculator developed for the placebo arm.12
The data inputs for the PCPT risk calculator are: age, PSA and DRE status within the prior year, family history, prior negative biopsy, and race. Men with larger BMI may have hemo-dilution of PSA17. Liang et al estimated BMI-adjusted PSA factors using PCPT data. We used adjusted PSA values in the PCPT risk calculator18.
Our approach differed slightly from Barrington et al8. NHW men in the lowest BMI category remained the reference group with the same BMI cut-points. However, incidence density ratio (IDR) is the measure of association instead of a HR from a proportional hazards model. The ID endpoint allows a man’s outcome to be something other than a 0 (no cancer) or 1 (prostate cancer). The ID for a group is the sum of incident prostate cancers divided by the sum of total person-years at risk19. For data from an exponential distribution, the ID is an efficient estimator for the hazard parameter, but it can also more generally serve as a useful descriptive statistic for time-to-event data20. An ID was calculated for each of the 30 BMI (5), race (2) and age group (3) combinations, and ID ratios (IDRs) were formed using ID of NHW lowest BMI as the reference group. IDRs can be interpreted as a measure of association analogous to a HR.
After estimating ID’s for each BMI, race and age cross-classification, the corresponding estimated covariance matrix of the cell-specific ratio estimates was computed via a first-order Taylor series approximation of the deviations of the estimates from their expected values21. Nine BMI/race IDRs adjusted for age group were estimated, along with standard errors and 95% confidence intervals. The GENMOD procedure in SAS was used to estimate IDRs by specifying Poisson regression. (SAS version 9.4).
Three approaches were used for estimating a man’s ID; a standard approach and two sensitivity analysis strategies.
Standard (unadjusted) Approach: This is analogous to a proportional hazards analysis of observed incident prostate cancer where, in the absence of a diagnosis, a man is assumed to be disease free. The numerator equals 1 for a prostate cancer diagnosis; otherwise equals 0. His time at risk (denominator) is days from randomization to diagnosis or last contact.
Sensitivity analysis #1: For-cause Imputation of Biopsy Result: We estimate a man’s probability of prostate cancer at his first for-cause screening value (i.e., PSA≥ 4.0 ng/ml or suspicious DRE) but no biopsy. The man’s ID indicator (numerator) is the estimated probability of prostate cancer, and his at-risk time (denominator) is the time to first for-cause screening. A man with a diagnosis of prostate cancer, a man with only a negative biopsy on study, or a man who had no “for-cause” screening had their ID handled the same as the Standard Approach.
Sensitivity analysis #2: Imputation for All Non-Biopsied: If a man had no study biopsy and no for-cause screening, we estimate his probability of prostate cancer when he exited the trial. Others follow sensitivity analysis #1.
For estimation of prostate cancer probability, we performed multiple imputation of the predicted probability, using the standard error provided from the PCPT risk calculator.
RESULTS
There were 22,673 NHW and 3398 AA men from SELECT that were evaluable for this analysis. AA men tended to be younger (median 58 years) compared to NHW men (median 62). Other characteristics are summarized in Table 1, stratified by race and BMI. AA men had higher BMI (p< 0.001) than NHW men, and were more likely to be hypertensive, diabetic, and current smokers. Both hypertension and diabetes increase with BMI. The numbers of PSA and DRE screens were similar across BMI and race categories. Approximately 85% of all SELECT men had PSA testing and 70% underwent DRE annually7. More NHW men underwent a negative prostate biopsy prior to entering SELECT (range: 8.8%−11.3%) compared to AA men (range: 4.9% – 7.4%), suggesting more intensive pre-trial screening in the NHW group.
Table 1.
Baseline and On-study Characteristics of SELECT Participants Stratified by BMI and Race for N=26,071 Men
| BMI Category of Subject | ||||||
|---|---|---|---|---|---|---|
| Race | Participant Factors At Study Entry | BMI 18.0 – 24.9 | BMI 25.0–27.4 | BMI 27.5–29.9 | BMI 30.0 –34.9 | BMI 35.0 – 50.0 |
| NHW | Sample size N (%) | 4555 (20.1%) | 6140 (27.1%) | 5153 (22.7%) | 5092 (22.4%) | 1733 (7.6%) |
| AA | 638 (18.8%) | 749 (22.0%) | 688 (20.2%) | 897 (26.4%) | 426 (12.5%) | |
| NHW | Current Smoker % | 7.3% | 5.5% | 4.8% | 4.6% | 5.4% |
| AA | 32.4% | 18.7% | 16.0% | 12.2% | 10.8% | |
| NHW | Heavy Alcohol Use% | 2.0% | 2.3% | 2.5% | 2.0% | 2.3% |
| AA | 3.3% | 2.8% | 1.3% | 1.9% | 1.9% | |
| NHW | Hypertension % | 19.5% | 25.3% | 32.1% | 41.7% | 52.6% |
| AA | 44.7% | 43.7% | 51.9% | 57.2% | 69.0% | |
| NHW | Diabetes % | 3.6% | 4.9% | 7.7% | 11.9% | 18.2% |
| AA | 10.7% | 14.0% | 16.0% | 19.8% | 31.0% | |
| NHW | Family History of Prostate Cancer % | 18.7% | 20.7% | 20.0% | 20.0% | 19.8% |
| AA | 16.3% | 20.4% | 18.3% | 17.7% | 19.7% | |
| NHW | Prior Negative Biopsy Before SELECT % | 11.3% | 11.3% | 10.8% | 11.2% | 8.8% |
| AA | 4.9% | 7.3% | 5.8% | 7.4% | 6.3% | |
| Participant Factors On-Study | ||||||
| NHW | # of DREs per Person-Year | 0.89 | 0.88 | 0.87 | 0.86 | 0.82 |
| AA | 0.86 | 0.86 | 0.86 | 0.86 | 0.84 | |
| NHW | # of PSA Screens per Person-Year | 1.20 | 1.20 | 1.19 | 1.18 | 1.15 |
| AA | 1.19 | 1.18 | 1.18 | 1.17 | 1.17 | |
| NHW | Median PSA just prior to biopsy | 3.50 | 3.58 | 3.64 | 3.40 | 3.34 |
| AA | 3.89 | 3.40 | 3.70 | 3.85 | 3.80 | |
| NHW | % with at least 1 on-study biopsy | 17.4% | 17.4% | 16.5% | 15.1% | 13.7% |
| AA | 17.6% | 20.2% | 17.7% | 17.5% | 18.5% | |
NHW = non-Hispanic white, AA=African American, DRE=digital rectal exam, PSA= prostatic specific antigen, BMI = body mass index
Table 2 provides the likelihood of getting a prostate biopsy by race and BMI. AA men were more likely to undergo an on-study biopsy, but no correlation with BMI. In the NHW group, the highest two BMI groups were significantly less likely to undergo a biopsy compared to the lowest BMI group. After accounting for age, family history and longitudinal PSA values and digital rectal exam status, race and BMI played a role in who was recommended for and accepted a prostate biopsy.
Table 2.
Time to First Biopsy Regardless of Prostate Cancer Outcome in a Cox Regression Model, Stratified by Race and BMI Category
| (A) Time to First Biopsy, Censor at Last Contact Date | ||
|---|---|---|
| Risk Factor | Hazard Ratio* (95% CI) | p-value* |
| Race × BMI Classification | ||
| NHW BMI 1 | Reference group 1.0 | |
| NHW BMI 2 | 1.02 (0.98, 1.07) | 0.35 |
| NHW BMI 3 | 0.98 (0.94, 1.03) | 0.40 |
| NHW BMI 4 | 0.90 (0.86, 0.94) | <0.0001 |
| NHW BMI 5 | 0.81 (0.76, 0.87) | <0.0001 |
| AA BMI 1 | 1.19 (1.09, 1.30) | 0.0001 |
| AA BMI 2 | 1.17 (1.07, 1.27) | 0.0004 |
| AA BMI 3 | 1.09 (1.00, 1.20) | 0.049 |
| AA BMI 4 | 1.03 (0.95, 1.12) | 0.48 |
| AA BMI 5 | 1.15 (1.03, 1.28) | 0.015 |
Adjusted for selenium, vitamin E supplement arms, age, family history and longitudinal PSA and DRE status;
Grp 1: BMI 18.0 – 24.9, Grp 2: BMI 25.0–27.4, Grp 3: BMI 27.5–29.9, Grp 4: BMI 30.0 –34.9, Grp 5: BMI 35.0 – 50.0
In Table 3 we provide the number of prostate cancer diagnoses and person-time at risk for each BMI and race group. In column 3, the covariate adjusted BMI and race HRs for incident prostate cancer from the Barrington et al report show the positive linear trend between BMI and prostate cancer in the AA group (p=0.03), but not in the NHW group (p=0.63).8 In column 4, the same model is adjusted for only age group to have a comparator for our IDR analysis. Columns 5–7 report the IDRs using the Standard Approach and 2 sensitivity analyses. As expected, the corresponding HRs (column 4) and Standard Approach IDRs (column 5) are very comparable and found an inverse linear association between BMI and prostate cancer in the NHW group (p=0.035), and among AA men there was a higher risk of prostate cancer in the top four BMI groups but no obvious linear trend (IDRs range 1.45–1.60, trend p=0.15).
Table 3.
Adjusted Incidence Density Ratios and Adjusted Hazard Ratios from the SELECT Trial, Modeling Time to Prostate Cancer
| Col. | 1 | 2 | 3 | 4 | 5 | 6 | 7 | |
|---|---|---|---|---|---|---|---|---|
| Observed Incident Cases and Person Years at-risk^^ | Hazard Ratio from original JAMA Oncology, Barrington et al** | Hazard Ratio from Proportional Hazards Model* | Standard Approach Incidence Density Ratio*, | Sensitivity Analysis #1: Imputing at 1st For-Cause Screening * | Sensitivity Analysis #2 Imputing at 1st For-Cause screening or End of Study* | |||
| Race | BMI Group | Prostate Cancer(N) | PY | HR (95% CI)** | HR* (95% CI)* | IDR* (95% CI)* | IDR* (95% CI)* | IDR* (95% CI)* |
| NHW | 1 (ref) | 289 | 24,599 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 |
| 2 | 438 | 33,194 | 1.12 (0.97, 1.30) | 1.12 (0.97, 1.30) | 1.12 (0.97, 1.30) | 1.09 (0.96, 1.23) | 1.07 (0.99, 1.16) | |
| 3 | 333 | 27,802 | 1.04 (0.89, 1.22) | 1.02 (0.87, 1.19) | 1.02 (0.87, 1.19) | 0.99 (0.87, 1.13) | 1.02 (0.94, 1.11) | |
| 4 | 299 | 27,572 | 0.96 (0.82, 1.13) | 0.92 (0.78, 1.08) | 0.92 (0.78, 1.08) | 0.90 (0.79, 1.03) | 1.02 (0.94, 1.11) | |
| 5 | 94 | 9,284 | 0.94 (0.74, 1.19) | 0.87 (0.69, 1.10) | 0.87 (0.69, 1.10) | 0.84 (0.69, 1.01) | 1.11 (0.99, 1.24) | |
| Linear Trend Across NHW BMI Groups | P=0.63 | p=0.035 | p=0.035 | p=0.004 | P=0.54 | |||
| AA | 1 | 39 | 3167 | 1.28 (0.91, 1.80) | 1.14 (0.81, 1.59) | 1.12 (0.80, 1.57) | 0.99 (0.74, 1.32) | 1.02 (0.85, 1.23) |
| 2 | 63 | 3797 | 1.67 (1.27, 2.21) | 1.53 (1.16, 2.01) | 1.51 (1.15, 1.98) | 1.32 (1.04, 1.67) | 1.17 (0.99, 1.37) | |
| 3 | 57 | 3531 | 1.64 (1.23, 2.19) | 1.47 (1.10, 1.95) | 1.45 (1.09, 1.93) | 1.21 (0.94, 1.55) | 1.14 (0.96, 1.35) | |
| 4 | 73 | 4593 | 1.68 (1.29, 2.18) | 1.47 (1.14, 1.90) | 1.46 (1.12, 1.89) | 1.25 (1.00, 1.57) | 1.17 (1.01, 1.36) | |
| 5 | 37 | 2116 | 1.90 (1.34, 2.70) | 1.63 (1.15,2.29) | 1.60 (1.14, 2.26) | 1.30 (0.95, 1.77) | 1.28 (1.05, 1.57) | |
| Linear Trend Across AA BMI Groups | P=0.03 | p=0.15 | p=0.15 | p=0.24 | P=0.092 | |||
Reported from paper, adjusted for age, education, diabetes mellitus, family history of prostate cancer, current smoking and trial supplement arm, PY = person-years at risk, HR=hazard ratio, IDR= incidence density ratio, AA=African American, NHW=Non-Hispanic White
Adjusted for age group,
time to prostate cancer or last contact date summed across race and BMI categories
BMI Categories: Grp 1: 18.0 – 24.9, Grp 2: 25.0–27.4, BMI 27.5–29.9, BMI 30.0 –34.9, BMI 35.0 – 50.0
Imputing at the first for-cause screening to mimic uniform prostate cancer ascertainment across groups (sensitivity analysis #1, column 6), AA and NHW men in the lowest BMI group (< 25.0) have similar risk of prostate cancer (IDR=0.99). AA men in BMI groups 2–5 have elevated risk of prostate cancer (IDRs 1.32, 1.21, 1.25, 1.30, respectively) compared to NHW in the lowest BMI group but no linear trend is apparent (trend p=0.24), and IDRs are closer to the null compared to the Standard Approach. NHW men had a significant inverse association between BMI and prostate cancer (trend p=0.004). The relationship between BMI and prostate cancer was the opposite direction for NHW men. Figure 1 shows the trends for the unadjusted and for-case screening adjustment by race and BMI group.
Figure 1.
Incidence Density Ratio Estimates for Prostate Cancer Incidence, Stratifying on Race and BMI Categories, Lowest Non-Hispanic White BMI Group (W1) is the Reference Group
*Bias-adjusted for those with elevated PSA or suspicious DRE but no biopsy (sensitivity analysis #1)
Imputing for everyone without an on-study biopsy (sensitivity analysis #2, column 7) the inverse linear trend in the NHW group disappears (trend p=0.54), suggesting biopsy bias appears to be operational in the NHW group too, and most IDR estimates for NHW and AA BMI groups moved closer to the null compared to “for-cause” only imputation (sensitivity analysis #1).
Because the multiple imputation approach had almost no effect on confidence intervals, only the single imputation results are reported.
DISCUSSION:
One could argue that differences in physician recommendations for biopsy or patient acceptance of biopsy are completely appropriate. For the former, they may be based on unmeasured risk factors (e.g., subtleties in DRE or risk of biopsy complications) and, for the latter, reflecting patient preferences. Unfortunately, these biases will affect the interpretation of other risk factors for prostate cancer. We previously reported that, accounting for PSA and DRE screening, men on SELECT with co-morbid conditions were less likely to undergo a biopsy7. These health conditions were more prevalent among AA men, and the prevalence often correlated with increasing BMI. Although screening was relatively uniform for all men on SELECT, it appears that screening prior to SELECT was more intensive among NHW men perhaps causing a delay in the conduct of on-study biopsies in the NHW group. AA men also had a greater risk of dropping out of the study early22. A number of these factors could differentially impact disease assessment, and it is difficult to anticipate the direction of the net bias.
The PCPT risk calculator was developed externally to SELECT, incorporating end of study biopsy results that minimized detection bias. It has been externally validated in a number of cohorts.23,24 An additional feature of the PCPT risk calculator is that it calculates a man’s risk of high grade prostate cancer. By applying the methods described in this paper, an investigator could use that endpoint instead of or in addition to all incident prostate cancer. Other prostate cancer risk calculators are based on groups of men who were recommended for and accepted biopsy based on abnormal DRE or elevated PSA. Use of these other risk calculators would propagate the biopsy detection bias. We evaluated the impact of substituting the probability of prostate cancer for men who did not undergo prostate biopsy on the association of BMI, race and prostate cancer. Our results suggest that the HRs from the multivariate PH model from the original Barrington et al paper8 over-estimate the risk of prostate cancer in AA men with higher BMI, and there is no evidence of a linear trend. However, we find that AA men with a BMI ≥ 25.0 have a 14–30% greater risk of prostate cancer compared to NHW men with BMI < 25.0, or alternatively, AA and NHW men with a BMI < 25 have similar risk of prostate cancer.
There are possible limitations to this study. Although we adjust for age group, there may be other useful covariates. The Barrington analysis8 adjusted for age, education, diabetes mellitus, family history of prostate cancer, current smoking status, and SELECT trial arm. However, such extensive covariate control has the potential to increase net bias even when the covariates would have been confounders under perfect classification, and even if covariates are determinants of classification. Bias due to misclassification cannot be adequately dealt with by the methods used for control of confounding.”25
We defined a for-cause screening rule to impute probability of prostate cancer. The PSA cutoff for biopsy level of 4.0 ng/mL could be disputed. We selected this level as it was the cutoff used in SELECT and PCPT and the most commonly-employed threshold for biopsy recommendation. Other for-cause PSA cutoffs could be employed to best match the screening patterns of a given cohort.
There are several approaches for missing cohort data. Structural modeling approaches to selection bias have been proposed where “selection bias” can encompass bias from inappropriate selection of controls, bias from differential loss-to-follow-up, non-response bias, among others26, 27, 28. Sometimes correction can be accomplished by a generalization of inverse-probability weighting or stratification, assuming outcomes are missing at random (MAR): missingness can be modeled using some observed covariates. There are situations in which MAR is questionable, and the missingness may be non-ignorable (NI): the missing data mechanism is related to the missing values and cannot be modeled with observed covariates. Some aspects of differential biopsies may introduce a NI pattern of missingness. Our method using an external calculator to estimate prostate cancer probability for those without a biopsy is one approach to minimize the NI missing data challenge. We do not use the observed covariable correlations with prostate cancer from SELECT to calculate biopsy outcomes as this would not reduce bias.
Our analysis identified factors that are differentially distributed by race and BMI. Most (e.g., smoking, hypertension, and diabetes) are related to life expectancy (and the perceived potential benefit of an earlier detection of prostate cancer, and its subsequent treatment). These factors could also be related to risk of biopsy complications. The source of the bias could be the patient’s decision, the treating physicians’ decision, or both. These biases likely confound previous attempts to link obesity with the risk of prostate cancer. Our analysis confirms AA men’s higher risk of prostate cancer, but, after accounting for differential detection bias, the role of BMI is not as clear.
Differential biopsy ascertainment may play a role in evaluating the association of many potential risk factors and prostate cancer including diabetes, statin use, vitamin D levels and physical activity, to name a few. To analytically minimize this differential detection bias, cohorts need to collect more detailed individual screening and biopsy data.
Highlights.
Factors unrelated to cancer risk can impact which men receive a prostate biopsy.
Prostate cancer is prevalent so factors that increase the likelihood of a biopsy will look like a risk factor.
Men with higher BMI and co-morbid conditions are less likely to get biopsied.
If all men in the cohort received biopsies uniformly, increasing BMI would not be associated with higher prostate cancer risk in this analysis.
Acknowledgments
Supported by grants: UM1 CA182883, P30 CA065174, U01 CA086402
SELECT ClinicalTrials.gov Identifier:
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Conflicts of interest
The authors declare no potential conflicts of interest.
References:
- 1).Jahn JL, Giovannucci EL, & Stampfer MJ The high prevalence of undiagnosed prostate cancer at autopsy: implications for epidemiology and treatment of prostate cancer in the Prostate-specific Antigen-era. International journal of cancer, 137(12), 2795–802, 2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2).Thompson IM, Pauler DK, Goodman PJ, et al. : Prevalence of prostate cancer among men with a prostate-specific antigen level, or = 4.0 ng per milliliter. N Engl J Med 350:2239–2246, 2004. [DOI] [PubMed] [Google Scholar]
- 3).Pashayan N, Duffy SW, Pharoah P, Greenberg D, Donovan J, Martin RM, Hamdy F, Neal DE Mean sojourn time, overdiagnosis, and reduction in advanced stage prostate cancer due to screening with PSA: implications of sojourn time on screening. British journal of cancer, 100(7), 1198–204, 2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4).Etzioni R, Cha R, Feuer EJ, Davidov O. Asymptomatic incidence and duration of prostate cancer. AJE 148:775–785, 1998. [DOI] [PubMed] [Google Scholar]
- 5).Tangen CM, Goodman PJ, Till C, Schenk JM, Lucia MS, Thompson IM. Biases in recommendations for and acceptance of prostate biopsy significantly affect assessment of prostate cancer risk factors: Results from two large randomized clinical trials. J Clin Oncolo 34:4338–4344, 2016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6).McNeil JJ, Nelson MR, Woods RL, Lockery JE, Wolfe R, Reid CM, Kirpach B, Shah RC, Ives DG, Storey E, Ryan J, Tonkin AM, Newman AB, Williason JD, Margolis KL, Ernst ME, Abhayarantna WP, Stocks N, Fitzgerald SM, Orchard SG, Treevaks RE, Beilin LJ, Donnan GA, Givvs P, Johnston CI, Radziszewska B, Grimm R, Murray AM. Effect of aspirin on all-cause mortality in the healthy elderly. N Engl J Med 379:1519–1528, 2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7).Soni PD, Hartman HE, Dess RT, Abugharib A, Allen SG, Feng FY, Zeitman AL, Jagsi R, Schipper MJ, Spratt DE. Comparison of population-based observational studies with randomized trials in oncology. J Clin Oncol 37(14):1209–1217, 2019. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8).Barrington WE, Schenk JM, Etzioni R, Arnold KB, Neuhouser ML, Thompson IM, Lucia MS, Kristal AR. Difference in association of obesity with prostate cancer risk between US African American and Non-Hispanic White men in the Selenium and Vitamin E Cancer Prevention Trial (SELECT), JAMA Oncol 1(3):342–349, 2015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9).Lippman SM, Goodman PJ, Klein EA, Parnes HL, Thompson IM Jr; Kristal AR, Santella RM, Probstfield JL, Moinpour CM, Albanes D, Taylor PR, Minasian LM, Hoque A, Thomas SM, Crowley JJ, Gaziano JM, Stanford JL, Cook ED, Fleshner NE, Lieber MM, Walthe PJ Khuri FR, Karp DD; Schwartz GG, Ford LG, Coltman CA Jr. Designing the selenium and vitamin E cancer prevention trial (SELECT). Journal of the National Cancer Institute 97(2):94–102, 2005. [DOI] [PubMed] [Google Scholar]
- 10).Lippman SM, Klein EA, Goodman P, Lucia MS, Thompson IM, Ford LG, Parnes HL, Minasian LM, Gaziano JM, Hartline JA, Parsons JK, Bearden JD, Crawford ED, Goodman GE, Claudio J, Winquist E, Cook ED, Karp DD, Walther PJ, Lieber MM, Kristal AR, Darke A, Arnold K Ganz PA, Santella RM, Albanes D, Taylor PR, Probstfield JL, Jagpal TJ, Crowley JJ, Meyskens FL, Baker LH, Coltman CA. Effect of selenium and vitamin E on risk of prostate cancer and other cancers: The Selenium and Vitamin E Cancer Prevention Trial (SELECT). Journal of the American Medical Association 301(1):39–51, 2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11).Klein EA, Thompson IM Jr., Tangen CM, et al. : Vitamin E and the risk of prostate cancer: The Selenium and Vitamin E Cancer Prevention Trial (SELECT). JAMA 306:1549–1556, 2011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12).Thompson IM, Ankerst D Pauler, Chi C, Goodman PJ, Tangen CM, Lucia MS, Feng Z, Parnes HL, Coltman CA Jr. Assessing prostate cancer risk: results from the Prostate Cancer Prevention Trial. J Natl Cancer Inst 98(8):529–534, 2006. [DOI] [PubMed] [Google Scholar]
- 13).Design of the Prostate Cancer Prevention Trial (PCPT). Feigl P, Blumenstein B, Thompson I, Crowley J, Wolf M, Kramer BS, Coltman CA Jr, Brawley OW, Ford LG. Control Clin Trials 16(3):150–63, 1995. [DOI] [PubMed] [Google Scholar]
- 14).Goodman PJ, Tangen CM, Crowley JJ, et al. : Implementation of the Prostate Cancer Prevention Trial (PCPT). Control Clin Trials 25:203–222, 2004. [DOI] [PubMed] [Google Scholar]
- 15).Thompson IM, Goodman PJ, Tangen CM, et al. : The influence of finasteride on the development of prostate cancer. N Engl J Med 349:215–224, 2003. [DOI] [PubMed] [Google Scholar]
- 16).Redman MW, Tangen CM, Goodman PJ, Lucia MS, Coltman CA Jr, Thompson IM. Finasteride does not increase the risk of high-grade prostate cancer: a bias-adjusted modeling approach. Cancer Prev Res 1(3):174–181, 2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17).Grubb RL III, Black A, Izmirlian G, Hickey TP, Pinsky PF, Mabie JE, Riley TL, Ragard LR, Prorok PC, Berg CD, Crawford ED, Church TR, Andriole GL Jr. Serum Prostate-Specific Antigen Hemodilution Among Obese Men Undergoing Screening in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. Cancer Epidemiol Biomarkers Prev 18:748–751, 2009. [DOI] [PubMed] [Google Scholar]
- 18).Liang Y, Ankerst DP, Sanchez M, Leach RJ, Thompson IM. Body Mass Index Adjusted Prostate-specific antigen and its application for prostate cancer screening. Urology 76:1268.e1–1268.e6, 2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19).Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic Research: Principles and Quantitative Methods. Van Nostrand Reinhold: New York, 1982. [Google Scholar]
- 20).Tangen CM, Koch GG. Nonparametic covariance methods for incidence density analyses of time-to-event data from a randomized clinical trial and their complementary roles to proportional hazards regression. Statist Med 20:2585–2607, 2001. [DOI] [PubMed] [Google Scholar]
- 21).LaVange LM, Keyes LL, Koch GG, Margolis PA. Application of sample survey methods for modelling ratios to incidence densities. Statistics in Medicine 13:343–355, 1994. [DOI] [PubMed] [Google Scholar]
- 22).Arnold KM, Hermos J, Anderson K, Minasian L, Tangen CM, Probstfield J, Cook E. Retention of black and white participants in the Selenium and Vitamin E Cancer Prevention Trial (SWOG-Coordinated Intergroup Study S0000) Cancer Epidemiology, Biomarkers and Prevention 23(12):2895–2905; 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23).Poyet C, Nieboer D, Bhindi B, Kulkarni GS, Wiederkehr C, Wettstein MS, Largo R, Wild P, Sulser T and Hermanns T Prostate cancer risk prediction using the novel versions of the European Randomised Study for Screening of Prostate Cancer (ERSPC) and Prostate Cancer Prevention Trial (PCPT) risk calculators: independent validation and comparison in a contemporary European cohort. BJU Int, 117: 401–408, 2016. [DOI] [PubMed] [Google Scholar]
- 24).Ankerst DP, Boeck A, Freedland SJ et al. W. Evaluating the PCPT risk calculator in ten international biopsy cohorts: Results from the Prostate Biopsy Collaborative Group. World J Urol 30(2): 181–187, 2012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25).Greenland S, Robins JM. Confounding and misclassification. Am J of Epidemiol 122(3):495–506, 1985. [DOI] [PubMed] [Google Scholar]
- 26).Robins JM, Finkelstein DM. Correcting for noncompliance and dependent censoring in an AIDS clinical trial with inverse probability of censoring weighted (IPCW) log-rank tests. Biometrics. 56:779–788, 2000. [DOI] [PubMed] [Google Scholar]
- 27).Hernán MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 11:561–570, 2000. [DOI] [PubMed] [Google Scholar]
- 28).Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 11:550–560, 2000. [DOI] [PubMed] [Google Scholar]