American Journal of Epidemiology Advance Access originally published online on January 22, 2007
American Journal of Epidemiology 2007 165(8):874-881; doi:10.1093/aje/kwk075
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
PRACTICE OF EPIDEMIOLOGY |
Evidence of a Healthy Volunteer Effect in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial
1 Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Bethesda, MD
2 Department of Public Health Sciences, University of Toronto, Toronto, Ontario, Canada
3 Office of Disease Prevention, National Institutes of Health, Bethesda, MD
4 School of Public Health, University of Minnesota, Minneapolis, MN
5 Department of Hematology/Oncology, Marshfield Clinic, Marshfield, WI
6 Lombardi Comprehensive Cancer Center, Georgetown University, Washington, DC
7 Department of Medicine, University of Pittsburgh, Pittsburgh, PA
8 Department of Internal Medicine, University of Utah Health Sciences Center, Salt Lake City, UT
9 Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, MD
Correspondence to Dr. Paul Pinsky, Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, 6130 Executive Boulevard, EPN 3064, Bethesda, MD 20892 (e-mail: pinskyp{at}mail.nih.gov).
Received for publication May 17, 2006. Accepted for publication September 25, 2006.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Volunteers for prevention or screening trials are generally healthier and have lower mortality than the general population. The Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) is an ongoing, multicenter, randomized trial that randomized 155,000 men and women aged 55–74 years to a screening or control arm between 1993 and 2001. The authors compared demographics, mortality rates, and cancer incidence and survival rates of PLCO subjects during the early phase of the trial with those of the US population. Incidence and mortality from PLCO cancers (prostate, lung, colorectal, and ovarian) were excluded because they are the subject of the ongoing trial. Standardized mortality ratios for all-cause mortality were 46 for men, 38 for women, and 43 overall (100 = standard). Cause-specific standardized mortality ratios were 56 for cancer, 37 for cardiovascular disease, and 34 for both respiratory and digestive diseases. Standardized mortality ratios for all-cause mortality increased with time on study from 31 at year 1 to 48 at year 7. Adjusting the PLCO population to a standardized demographic distribution would increase the standardized mortality ratio only modestly to 54 for women and 55 for men. Standardized incidence ratios for all cancer were 84 in women and 73 in men, with a large range of standardized incidence ratios observed for specific cancers.
demography; mass screening; mortality; neoplasms; randomized controlled trials; standardized mortality ratio; survival rate
Abbreviations: PLCO, Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; SEER, Surveillance, Epidemiology, and End Results
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Volunteers for prevention or screening trials tend to be healthier than the overall population; this has been denoted the "healthy volunteer effect." Ederer et al. (1) summarized selection effects that may lead to a healthier volunteer population as arising from one or more of the following sources: specific inclusion/exclusion criteria in the study protocol, specific targeting of healthier individuals in recruitment, self-selection of persons with above average education or income or with a healthier lifestyle, and self-exclusion of persons in poor health who otherwise meet study entry criteria. Control arm populations from a number of large prevention or screening trials, including the Multiple Risk Factor Intervention Trial, the Physicians' Health Study, and the Colon Cancer Control Study, were reported to have reduced overall or cause-specific mortality relative to that of the general population (2–4). Such differences between volunteer trial populations and the larger population of interest have given rise to concerns about the generalizability of results from some randomized trials utilizing volunteers (5).
Although the existence of the healthy volunteer effect has been well documented, there is less literature that examines this effect over a wide range of outcomes or that quantitatively analyzes its causes. The purpose of this investigation is to examine this effect in depth by use of the cohort of volunteers enrolled in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO). PLCO is an ongoing, large, multicenter trial designed to test whether screening for prostate, lung, colorectal, and ovarian cancers reduces mortality from these cancers (6). Trial subjects were volunteers recruited largely through population mass mailings at 10 screening centers across the United States. The large size of PLCO, the variety of baseline data collected, and the extensive follow-up of trial subjects make this cohort ideal for examining many aspects of the healthy volunteer effect.
Specifically, we examine overall mortality rates, cause-specific mortality rates for major disease groupings, and incidence and survival rates of major cancers in the PLCO cohort to date, and we compare these with expected rates based on US population statistics. We also examine mortality rates as a function of time on study to assess whether any observed healthy volunteer effect may be changing over time. We do not display incidence or mortality rates for PLCO cancers because PLCO is an ongoing trial and these endpoints are not yet available for release. Because PLCO cancers are excluded, all analyses are performed on screening and control arms combined.
Differences from the expected in mortality or cancer incidence rates in PLCO may arise in part due to demographic differences between the PLCO cohort and the overall US population. To assess this possibility, we compared the demographic characteristics of the PLCO cohort with those from a similarly aged cohort participating in the National Health Interview Survey and utilized a statistical model that quantitatively estimates how much of any observed differences is due to demographic differences.
A more complete understanding of the healthy volunteer effect will be helpful in planning for and interpreting results from trials that use volunteer populations. In addition, by examining various aspects of the healthy volunteer effect, for example, how its magnitude varies over the spectrum of cancers, we may gain further insight into the factors affecting disease prevention and health promotion.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The PLCO was designed to determine the efficacy of screening for prostate, lung, colorectal, and ovarian cancers in reducing cause-specific mortality. Randomization of subjects aged 55–74 years to an intervention or control arm began in November 1993 and was completed in July 2001. The study was carried out at 10 screening centers; institutional review boards at each center approved the study. Informed consent was obtained from all subjects prior to randomization. Subjects with a personal history of one of the four PLCO cancers or who were currently undergoing treatment for any cancer, except basal or squamous cell skin cancer, were excluded from the trial, as were subjects with a recent history of certain screening procedures for the PLCO cancers. Recruitment was performed primarily through mass mailing of invitational brochures and letters of invitation to age-eligible individuals identified from public, commercial, or screening center-specific mailing lists. Further details of the study design have previously been published (6). At randomization, subjects completed a written, self-administered baseline questionnaire, which included questions on demographics, medical history, health-related behaviors, and family and personal history of cancer. Subjects in the screened arm received a prostate-specific antigen test and a digital rectal examination for prostate cancer, posterior-anterior view chest radiograph for lung cancer, flexible sigmoidoscopy for colorectal cancer, and a cancer antigen 125 (CA-125) blood test and transvaginal ultrasound for ovarian cancer; control arm subjects received usual care.
Study subjects were sent annual study update forms, which asked about overall status and cancer diagnoses. Cancers reported from these forms or other sources (i.e., death certificates) were confirmed through medical records; we report here only on confirmed cancers diagnosed by the study cutoff date of December 31, 2003. About 8 percent of reported cancers were unconfirmed; historically in PLCO, about 65 percent of such cancers are eventually verified as cancer. Deaths were ascertained by various means, including reports from next of kin, local searches, and National Death Index searches. Suspected or reported deaths were confirmed with death certificates; only confirmed deaths were utilized for this analysis. Over 99 percent of reported deaths were confirmed.
Person-years of follow-up were computed as the time between randomization and the date of death or study cutoff date, whichever came first. Because randomization occurred over an extended period (1993–2001), the length of follow-up of subjects was variable. Expected deaths and incident cancers were computed by applying sex- and age-specific national mortality rates and Surveillance, Epidemiology, and End Results (SEER) 13 registries' incidence rates for the period 1996–2000 to the person-years of follow-up in PLCO. Standardized mortality and incidence ratios were defined as the ratio of observed to expected deaths or incident cancers (x 100). Cause-specific standardized mortality ratios were computed for major disease groupings by use of the underlying cause of death.
Expected 5-year relative survival rates for specified cancers were calculated utilizing age- and sex-specific SEER registries' 5-year relative survival rates (for the years 1995–2000) (7). SEER registries' relative survival statistics are calculated as the ratio of observed to expected survival, where expected survival is based on age- and sex-specific all-cause mortality rates for the overall SEER registries' population (8). Observed relative survival rates for incident cancers in PLCO were computed similarly as observed over expected survival; however, because overall mortality rates in PLCO differ from those of the general population, expected survival here was based on all-cause mortality rates for the PLCO cohort, stratified by age, sex, and study year.
With regard to statistical methods, the number of deaths or incident cancers in a given time period was assumed to follow a Poisson distribution. To test whether an observed number of events differed significantly from expected, we assumed that the expected number of events was fixed. The likelihood ratio test was used to determine the statistical significance of differences in standardized mortality ratios between men and women.
A standardized mortality ratio (or standardized incidence ratio) different from 100 could be partially explained by a different distribution of demographic and other risk factors in PLCO than in the general population. We computed an adjusted standardized mortality ratio (standardized incidence ratio) by calculating how many deaths (cancers) would have been observed if the distribution of risk factors in PLCO matched that of the general population. Specifically, we estimated hazard ratios (for mortality and for cancer incidence) for various risk factors using the PLCO data and a multivariable proportional hazards model. We estimated the baseline hazard by computing observed rates in subjects with no risk factors. The baseline hazard and the estimated hazard ratios then enable estimation of rates for any risk factor combination. Taking a weighted average of the rates, with weights proportional to the prevalence of the risk factor combination in the general population, then gives the adjusted observed rate. Data from the National Health Interview Survey for the years 1997–2000 for the same age group as in PLCO were used to generate the general population risk factor distributions (9). Adjusted standardized mortality ratios (standardized incidence ratios) were computed separately for men and women.
We utilized here two sets of risk factors. The first contained demographic and lifestyle variables as follows: education (high school degree, yes/no), marital status (married, yes/no), cigarette smoking (never, former, current), Black race (yes/no), body mass index (<30, 30–35, >35 kg/m2), and regular physical activity (yes/no). The second, used for the standardized mortality ratio only, contained the first set above and also included reported histories of each of the following conditions: myocardial infarction or coronary heart disease, stroke, emphysema, cancer (except nonmelanoma skin), diabetes, and hypertension.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Table 1 gives the population characteristics of the 78,234 women and 76,704 men enrolled in PLCO. Most were non-Hispanic White and aged less than 65 years. The average length of follow-up for this analysis was 5.4 years; 95 percent had complete follow-up through year 3, 71 percent had complete follow-up through year 5, and 40 percent had complete follow-up through year 7.
|
For comparison, also displayed in table 1 are the characteristics of the population completing the National Health Interview Survey. PLCO subjects had higher levels of education and physical activity, were less likely to be current smokers, and had lower reported histories of cancer, diabetes, and cardiovascular and respiratory diseases than did National Health Interview Survey subjects.
Table 2 displays standardized mortality ratios for selected causes and overall (excluding PLCO cancers). Standardized mortality ratios were 43 for all-cause mortality, 56 for non-PLCO cancers, 37 for cardiovascular deaths, 34 for both digestive and respiratory diseases, 28 for diabetes, and 64 for injuries and poisoning. Among major non-PCLO cancers, the highest standardized mortality ratios were 81 for pancreatic and 89 for brain cancers, and the lowest were 28 for breast, 22 for uterine, and 16 for cervical cancers. For all listed causes except brain cancer, the observed number of deaths was statistically significantly less than expected.
|
Men had statistically significantly higher standardized mortality ratios than did women for all-cause mortality (46 vs. 38), all cancer mortality, respiratory diseases, diabetes, cardiovascular diseases, and non-Hodgkin's lymphoma (table 2). Standardized mortality ratios for all-cause mortality differed little by age; standardized mortality ratios were 43 for men and 36 for women aged 55–64 years at baseline versus 47 for men and 38 for women aged 65–74 years at baseline.
Table 2 also displays standardized incidence ratios for cancer incidence. The standardized incidence ratios for all cancer (excluding PLCO cancers) of 73 for men and 84 for women were significantly less than 100. Standardized incidence ratios for men were 48–62 for four cancers (oral cavity, stomach, bladder, and liver), 100 for leukemia, and 70–92 for all other major cancers. Standardized incidence ratios for bladder, esophageal, kidney, liver, melanoma, oral cavity, and stomach cancers were all significantly below 100 in men. The range of standardized incidence ratios was greater for women than for men, where there were six listed cancers with standardized incidence ratios between 34 and 61 (bladder, cervical, esophageal, liver, oral cavity, and stomach), seven between 70 and 95 (brain, breast, kidney, non-Hodgkin's lymphoma, pancreatic, thyroid, and uterine), and two above 100 (melanoma and leukemia).
Table 3 examines standardized mortality ratios (and standardized incidence ratios) as a function of time since randomization. The standardized mortality ratio for all-cause mortality increased sharply from the first to the second year after randomization and then increased more gradually over the next 5 years. Nevertheless, the standardized mortality ratio remained substantially below 100 throughout the 7-year period of observation, peaking at 48 at year 7. On average, the standardized mortality ratio for all-cause mortality increased by 2.4 per year. The change in standardized mortality ratios over time differed by cause of death. Deaths due to non-PLCO cancer increased at the greatest rate (average change in standardized mortality ratio per year = 7.1). Standardized mortality ratios for respiratory and digestive diseases also increased significantly, by 2.9 and 4.1 per year, respectively. Standardized mortality ratios for cardiovascular and injury deaths were essentially constant over time.
|
Table 4 examines survival rates for incident cancer cases. Observed 5-year relative survival rates were generally slightly higher than expected. Non-Hodgkin's lymphoma (68 percent observed vs. 57 percent expected), kidney cancer (73 percent vs. 62 percent), and breast cancer (96 percent vs. 89 percent) all had confidence intervals for observed rates that did not overlap the expected rate.
|
Table 5 displays adjusted standardized mortality ratios. We found that each of the demographic variables used to calculate the adjusted standardized mortality ratios (i.e., smoking, education, Black race, physical activity, body mass index, and marital status) was associated with mortality in both men and women. Adjusting for these demographic/behavioral variables increased the standardized mortality ratio from 46 to 55 in men and from 38 to 54 in women. Further adjusting for disease history increased the standardized mortality ratio to 63 in men and 62 in women. With respect to cancer incidence, adjusting for demographic/behavioral variables increased the standardized incidence ratio in men from 73 to 80; in women, the adjusted standardized incidence ratio of 82 was slightly less than the unadjusted standardized incidence ratio of 84.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
PLCO subjects are better educated than the general population, more physically active, more likely to be married, and less likely to be current smokers. This is consistent with results from other large North American screening studies, such as the Canadian National Breast Screening Study, where it was found that the volunteers were better educated, slightly less likely to smoke, and of higher socioeconomic status than the general population (10). The men and women who are enrolled in PLCO also have considerably lower than expected all-cause mortality, with standardized mortality ratios of 46 for men and 38 for women. Within the PLCO cohort, ever smoking, low education, Black race, lack of physical activity, and not being currently married were all associated with increased mortality rates in both men and women, with relative risks or hazard ratios similar to those reported in previous studies (11). The favorable distribution in PLCO of these standard demographic and behavioral risk factors that predict for mortality, however, is not the primary explanation for the low standardized mortality ratios observed in the PLCO cohort. Standardizing the distribution of these factors to match that of the US population only modestly increased the standardized mortality ratio, to about 55 for both men and women. Further adjusting for a history of various diseases (stroke, coronary heart disease/myocardial infarction, emphysema, diabetes, cancer, and hypertension) provided little additional increase, with standardized mortality ratios still only around 63. Interestingly, the adjustment for risk factor distribution made the difference in standardized mortality ratios between men and women disappear. There are other factors not included in these adjustments, such as alcohol consumption, diet, and access to medical care, that could also have a favorable distribution in the PLCO cohort and may also be playing a role.
The remaining difference in standardized mortality ratio probably reflects the fact that subjects with certain chronic diseases or conditions that greatly predispose to mortality over the next 5–10 years were unlikely to volunteer for the PLCO. However, once enrolled, a certain proportion of subjects will develop such chronic diseases; thus, one would expect standardized mortality ratios to increase over time since enrollment. In PLCO, standardized mortality ratios for all-cause mortality increased from 31 at year 1 to 48 at year 7. The Colon Cancer Control Study found a similar result, with the standardized mortality ratio increasing from 56 at year 1 to 72 at years 8–10 (4). Linsted et al. (12) found that the all-cause mortality ratio for nonresponders versus responders in the Adventist Health Study decreased from 2.5 at year 1 to 1.4 at year 6. In our study, the standardized mortality ratio for non-PLCO cancer increased the most per year on study, followed by respiratory disease; no significant increase was seen for cardiovascular mortality.
The standardized mortality ratios and standardized incidence ratios reported here were computed by use of both arms of the PLCO combined. Because both incidence of and mortality from PLCO cancers were excluded in the analyses presented here, one would expect that the two arms would be comparable in terms of mortality and cancer incidence, and this was indeed the case. The standardized mortality ratios for all-cause mortality (excluding deaths from PLCO cancers) were 42.5 for the control arm versus 42.6 for the intervention arm, whereas the standardized incidence ratios for all non-PLCO cancer incidences were 78.6 for the control arm versus 79.1 for the intervention arm.
The standardized mortality ratio observed here is lower than that reported in other disease prevention or screening studies. The Colon Cancer Control Study had a standardized mortality ratio of 69 over 10 years of follow-up (1). In the Multiple Risk Factor Intervention Trial, the 6-year control group mortality from coronary heart disease was 67 percent of expected (3). In the Rotterdam section of the European Randomised Study of Screening for Prostate Cancer, the standardized mortality ratio for all-cause mortality was 91 (13). One reason for the relatively high standardized mortality ratio in the Rotterdam Study may be the high participation rate (proportion invited who were randomized) in that study, 49 percent. In US studies using healthy volunteers, participation rates are often difficult to calculate because of the different methods of recruitment than in Europe; however, most studies, including PLCO, would be estimated to have substantially lower participation rates.
The standardized incidence ratio for overall cancer incidence was considerably higher than the standardized mortality ratio but was still statistically significantly below 100. Adjusting for the risk factor distribution had less of an effect on the standardized incidence ratio than on the standardized mortality ratio and actually made the standardized incidence ratio in women decrease slightly. This is because the magnitude of effect of these demographic and behavioral risk factors was generally lower for cancer incidence than for mortality, and the direction of effect even changed for some factors. Low education and Black race were associated with higher mortality but lower cancer incidence in women in PLCO. This mirrors national statistics, which show that Black women have a lower cancer incidence than White women but higher all-cause mortality. The incidence of the most common noncutaneous cancer in women, breast cancer, has also been shown to be increased in women of higher socioeconomic status (14). As with the standardized mortality ratio, the difference in standardized incidence ratios between men and women disappeared when the adjusted standardized incidence ratios were used. One important factor that we did not control for is diet. Although diet is thought to be an important factor in the incidence of many cancers, it is such a complex factor and its effect is so difficult to quantify that we did not attempt to control for it. This could in part explain the finding of the adjusted standardized incidence ratios still being below 100.
In men, and especially in women, there was a wide range of standardized incidence ratios for specific cancer types, ranging from 34 for cervical cancer in women to 117 for leukemia in women. This may reflect the fact that the differential demographic and risk factor profile in PLCO has a greater effect on some cancers than on others. Cervical cancer incidence is generally lower in populations with high socioeconomic or educational status, partly because such women are more likely to have had earlier cervical screening that can prevent future cancers by removing preneoplastic lesions (15). On the other hand, as mentioned above, breast cancer incidence is positively correlated with socioeconomic status, and the standardized incidence ratio of 92 for breast cancer was close to 100. The lower than expected incidence in PLCO of several smoking-related cancers (e.g., esophagus, bladder, kidney, pancreas, and oral cavity) may reflect the fact that the PLCO cohort has lower than average rates of current smoking.
The power of a randomized screening or prevention trial depends on the event rate in the control group and on the hypothesized magnitude of effect. If a trial is planned for the general population, and if control rates are estimated from population data, then adjustments should be made on the basis of a likely healthy volunteer effect. The data presented here can be used to help estimate the magnitude of that effect. The original power calculations for PLCO did assume some magnitude of a healthy volunteer effect, and thus the results reported here do not imply that the power of the PLCO is compromised (6).
If a healthy volunteer effect is evident in a trial population, what are the implications in terms of the generalizability of trial results? All other things being equal, the percentage reduction in disease incidence or mortality in the intervention as compared with the control arm is independent of the baseline disease rate. For example, diluting each arm with an equal number of subjects who will not get disease does not affect the percentage reduction. It is possible, though, that the effectiveness of an intervention may differ in the trial as compared with the general population because of various factors, including different patterns of compliance with screening and different patterns of survival between the trial and general populations. However, a randomized trial retains its internal validity even in the presence of a healthy volunteer effect. Further, in North America and other places, choosing to be screened is discretionary, so that the population likely to undergo screening when it is disseminated to the public also will display a healthy volunteer effect, though one that is probably less pronounced than among a population volunteering for a trial.
The cohort of volunteers for PLCO has markedly lower than expected observed mortality through an average over 5 years of follow-up, and only a minority of the mortality reduction can be attributed to differences in demographic factors between the cohort and the general population. The overall incidence of cancer was somewhat lower than expected; however, there was a wide range in expected over observed ratios across the various cancer types.
|
ACKNOWLEDGMENTS |
|---|
Conflict of interest: none declared.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
- Ederer F, Church TR, Mandel J. (1993) Sample sizes for prevention trials have been too small. Am J Epidemiol 137:787–96.
[Abstract/Free Full Text] - Multiple Risk Factor Intervention Trial: risk factor changes and mortality results. (1982) Multiple Risk Factor Intervention Trial Research Group. JAMA 248:1465–77.
[Abstract/Free Full Text] - Final report on the aspirin component of the ongoing Physicians' Health Study. (1989) Steering Committee of the Physicians' Health Study Research Group. N Engl J Med 321:129–35.[Abstract]
- Church TR, Ederer F, Mandel JS, et al. (1993) Estimating the duration of ongoing prevention trials. Am J Epidemiol 137:797–810.
[Abstract/Free Full Text] - Paci E and Alexander FE. (1997) Study design of randomized controlled clinical trials of breast cancer screening. J Natl Cancer Inst Monogr 22:21–5.
[Abstract/Free Full Text] - Prorok PC, Andriole GL, Bresalier RS, et al. (2000) Design of the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. Control Clin Trials 21:suppl, 273S–309S.[CrossRef][Web of Science][Medline]
- SEER cancer statistics review, 1975–2001. (2005) (National Cancer Institute, Bethesda, MD) (http://seer.cancer.gov/csr).
- Ries LA, Kosary CL, Hankey BF, et al. (1997) SEER cancer statistics review, 1973–1994. (National Cancer Institute, Bethesda, MD) (NIH publication no. 97-2789).
- National Health Interview Survey. (2005) (National Center for Health Statistics, Hyattsville, MD) (http://www.cdc.gov/nchs/about/major/nhis/quest_data_related_1997_forward.htm).
- Miller AB, Baines CJ, To T, et al. (1993) Canadian National Breast Screening Study: 2. Breast cancer detection and death rates among women aged 50 to 59 years. CMAJ 1992;147:1477–88. Erratum in: CMAJ 148:718.
- Sorlie PD, Backlund E, Keller JB. (1995) US mortality by economic, demographic, and social characteristics: the National Longitudinal Mortality Study. Am J Public Health 85:949–56.
[Abstract/Free Full Text] - Lindsted KD, Fraser GE, Steinkohl M, et al. (1996) Healthy volunteer effect in a cohort study: temporal resolution in the Adventist Health Study. J Clin Epidemiol 49:783–90.[CrossRef][Web of Science][Medline]
- Otto SJ, Schroder FH, de Koning HJ. (2004) Low all-cause mortality in the volunteer-based Rotterdam section of the European Randomised Study of Screening for Prostate Cancer: self-selection bias? J Med Screen 11:89–92.[CrossRef][Web of Science][Medline]
- Yost K, Perkins C, Cohen R, et al. (2001) Socioeconomic status and breast cancer incidence in California in different race/ethnic groups. Cancer Causes Control 12:703–11.[CrossRef][Web of Science][Medline]
- Liu T, Wang X, Waterbor JW, et al. (1998) Relationship between socioeconomic status and race-specific cervical cancer incidence in the United States, 1973–1992. J Health Care Poor Underserved 9:420–32.[Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. M. Croswell, B. S. Kramer, A. R. Kreimer, P. C. Prorok, J.-L. Xu, S. G. Baker, R. Fagerstrom, T. L. Riley, J. D. Clapp, C. D. Berg, et al. Cumulative Incidence of False-Positive Results in Repeated, Multimodal Cancer Screening Ann. Fam. Med, May 1, 2009; 7(3): 212 - 222. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Blom, L. Yin, A. Liden, A. Dolk, B. Jeppsson, L. Pahlman, L. Holmberg, and O. Nyren A 9-Year Follow-up Study of Participants and Nonparticipants in Sigmoidoscopy Screening: Importance of Self-Selection Cancer Epidemiol. Biomarkers Prev., May 1, 2008; 17(5): 1163 - 1168. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z. Kabir Selection Bias, Confounding, or Information Bias? Hypertension, July 1, 2007; 50(1): e9 - e9. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||