NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Wilson B, Qureshi N, Little J, et al. Clinical Utility of Cancer Family History Collection in Primary Care. Rockville (MD): Agency for Healthcare Research and Quality (US); 2009 Apr. (Evidence Reports/Technology Assessments, No. 179.)

  • This publication is provided for historical reference only and the information may be out of date.

This publication is provided for historical reference only and the information may be out of date.

Cover of Clinical Utility of Cancer Family History Collection in Primary Care

Clinical Utility of Cancer Family History Collection in Primary Care.

Show details

3Results

Our comprehensive search yielded 10,644 unique citations; from these 9,765 were excluded as they were not an English language publication, not on the cancers of interest, or on topic for any of the four research questions (Figure 3). The remaining 879 citations were screened at full text and of these a total of 12 primary studies and 29 systematic reviews were eligible for inclusion in this review.

Research Q1: Which Risk Stratification Algorithms or Guidelines Delineate Risk Accurately, and in a Clinically Meaningful Way?

General Approach

We reviewed published studies which examined the ability of models (or algorithms, or guidelines) which used family history information to accurately predict individual risk of breast, ovarian, colorectal, or prostate cancer. To be eligible, the model had to include systematic collection of specified family history information, either alone or with other personal or clinical information which would be available for all patients and routinely available to a primary care practitioner. We examined the performance of models in relation to populations not selected for known or suspected high risk of cancer.

Model performance was assessed by predictive accuracy, in terms of calibration and discrimination. “Calibration” is a model's ability to correctly predict the number of observed events (incidence of cancer) in a population and is generally evaluated by its goodness-of-fit to observed events. The ratio of observed to expected cases provides an overall epidemiological assessment of how well a model might perform for a defined population.

“Discrimination” is the assessment of how well a model separates out individuals who will go on to develop different outcomes. Discriminatory accuracy for dichotomous outcomes (e.g., disease/no disease) is best examined through metrics such as sensitivity and specificity, predictive value, likelihood ratio, and the area under the receiver operator characteristic (ROC) curve (or area under the curve (AUC)). The AUC is also referred to as the c-statistic and is defined in the unit square with a range of 0 to 1.0. Since chance alone will follow a perfect diagonal from (0,0) to (1,1), the subsequent AUC will be 0.5 (no apparent discrimination), whereas an AUC of 1.0 indicates perfect discriminatory accuracy.

Discriminatory accuracy is a more relevant evaluation than calibration from the point of view of clinical practice, as it directly indicates how well the model predicts an individual patient's likelihood of developing cancer within a defined time scale. A model that is well calibrated at a population level may not necessarily be highly discriminating when used for individual prediction.

Studies Reviewed

Eight evaluation studies were retained for data abstraction after full text review.714 All except one focused on breast cancer risk assessment. The individual breast cancer models were all conceptually related: the “original” Gail model,15 the modified Gail model,84 the modified Gail model for African American populations,10 the modified Gail model for Italian populations,11, 12 and the model developed from the Contraceptive and Reproductive Experiences (CARE) study.7 The exception was Harvard Cancer Risk Index (HCRI),16 which was developed for calculating the risk of a several cancers, including breast, ovary, colon, and prostate. An eligible validation14 of the HCRI was published only for colon cancer.

The details of the models are summarized in Table 4. All of the models examined used a set of predetermined input variables in addition to family history as the basis for risk assessment. These variables included a range of personal demographic and disease-specific risk factors such as age, ethnicity, reproductive factors, diet, history of clinical investigations (e.g., breast biopsies, colonoscopies), and other risk factors such as body mass index, and alcohol consumption.

Table 4. Details of risk prediction models.

Table 4

Details of risk prediction models.

Outcomes

The details of evaluations of the models are summarized in Table 5. All evaluations were performed as secondary analyses of data derived from observational studies or trials. The sample sizes for the evaluations ranged from 1,450 to 147,916, covering a wide age range of participants. Six of the evaluations710, 13, 14 were conducted in U.S. populations: the Nurses' Health Study (NHS)14, 85, 86 (4 studies), the Women's Health Initiative (WHI)7, 87 (2 studies), the Black Women's Health Study,88 and the Health Professionals Follow-Up Study (HPFS).87 The remaining two evaluations11, 12 were conducted in Italian populations. Followup periods in the validation cohorts ranged from 5 to 10 years.

Table 5. Evaluations of risk prediction models.

Table 5

Evaluations of risk prediction models.

Gail Model

The first published version of the ‘Gail Score’15 used information on a woman's age at menarche, her age at the time of the birth of her first child, the number of her first degree relatives who had had breast cancer, and the number of previous breast biopsies that she had undergone. It was designed for women with no personal history of breast cancer who were being followed by annual screening mammography. It estimates the absolute probability of developing invasive or in situ breast cancer over a defined age interval. The model uses estimates of baseline hazard and attributable risk derived from the Breast Cancer Detection Demonstration Project (BCDDP).89 The authors indicated that its primary application would be to “determine eligibility for entry to breast cancer prevention trials, where an important determinant of sample size is the absolute risk of breast cancer” in the study population.15

We identified a single study8 which evaluated the original Gail model,15 a secondary analysis of data from the NHS.85, 86 The ratio of expected to observed breast cancer cases was 1.33 (95 percent CI 1.28-1.39), which was a significant overestimation. Only modest discriminatory accuracy was demonstrated (Pearson correlation coefficient 0.67).

Modified Gail Model

The modified Gail model incorporated revisions to improve its validity and applicability to the North American population.84 The key revisions were a focus on the absolute risk of invasive breast cancer only (i.e., in situ cancer was excluded): the inclusion of a diagnosis of atypical hyperplasia on biopsy as an additional risk factor; and the substitution of age-specific invasive breast cancer rates from the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute (NCI) for the BCDDP-based data used in the original model.

We identified five studies which evaluated the performance of the modified version of the Gail model, either in its originally published form, or adjusted to take account of different underlying breast cancer incidence patterns in different populations (African American,10 Italian11, 12). The ratios of expected to observed invasive cancer cases ranged from 0.79 to 0.96, mostly not statistically significantly different from unity. Validations, performed in a predominantly white U.S. population9 and two Italian populations,11, 12 suggested reasonably good calibration, but poorer performance in “more diverse”13 and African American10 U.S. populations. These studies all suggested that the standard and modified versions all had very modest discriminatory accuracy (concordance statistic 0.58-0.59).

CARE Model

The CARE Model7 was developed as an African American adaptation of the Gail model, using data from the Women's Contraceptive and Reproductive Experiences (CARE) Study,90 the SEER program, and National Center for Health Statistics. It is a simpler model, in that two variables in the modified Gail model were removed and one was dichotomized.

We identified one study that validated this model, in a subset of African American women in the WHI. The model showed good calibration, with an expected observed ratio of cases not statistically significantly differently from unity. The discriminatory accuracy, expressed as AUC, was 0.555, which suggests only very modest predictive ability at best.

The authors compared the CARE Model with the MGM in classifying eligibility for a chemoprevention trial, and found that it doubled the proportion of women who met inclusion criteria (30.3 percent compared with 14.5 percent).

Harvard Cancer Risk Index

The HCRI was developed as a tool to assist clinicians in counseling patients about cancer risk reduction.16 It addressed prediction of the most common cancers in American men and women (prostate, breast, lung, colon, bladder, endometrium, non-Hodgkin's lymphoma, ovary, kidney, leukemia, cervix, pancreas, skin melanoma, and stomach). The tool was developed using a consensus-based process in which available evidence was used to assign points for different levels of defined risk factors. Risk was stratified into a seven point categorical scale, set relative to the average U.S. population risk for each cancer. Several risk factors contributed to the score for each cancer, family history being included in all except leukemia and cervical cancer.

We identified one validation study,87 in which the HCRI's predictive validity for colon cancer was assessed in two cohorts, the NHS (women) and the HPFS (men). Results pertaining to ovarian cancer were also reported in this publication, but family history items were lacking, therefore these data were excluded from the current review. In addition to family history, eight variables for men and eleven for women were included in the HCRI for colon cancer.

The overall ratios of observed and expected cases were not reported. For women, the ratios by initial risk stratum ranged from 0.58 to 1.79, and for men 0.75 to 2.35. A better fit was observed for women than men. The AUC was 0.67 for women and 0.71 for men, suggesting moderately good discriminatory accuracy.

Conclusion

There were essentially two families of tools on which relevant performance data were available, those based on the Gail model, and the HCRI. None of the identified studies evaluated the performance of a predictive tool or algorithm, designed for use in populations not already pre-selected for higher risk, and using family history information alone. No validation studies of tools designed for risk prediction for ovarian or prostate cancer were found.

Most, but not all models demonstrated good calibration for the populations for which they were developed. However, none of the models identified demonstrated more than moderate ability to correctly discriminate risk at an individual level. The highest concordance presented was 0.71: this is equivalent to the correct classification of future disease (present or absent) in 71 percent of individuals, and incorrect classification of 29 percent.

Research Q2: For Which Behaviors and Clinical Preventive Services is There Evidence of Benefit in Terms of Actual Reduction of Disease Risk, and What Harms, if any, Have Been Identified?

General Approach

For the purposes of this review, we included published quantitative and qualitative reviews of the effectiveness of personal behavioral/lifestyle and clinical interventions that are commonly recommended as part of cancer risk reduction strategies in primary care settings. The list of interventions of interest was developed by the study team in consultation with the partners and the members, and was incorporated into the eligibility criteria for the review. Table 6 lists the interventions of interest.

Table 6. Target interventions for review.

Table 6

Target interventions for review.

Where multiple reviews addressing the same intervention were identified, they were scrutinized to determine the degree of overlap, as well as for quality. We selected the most recent and/or most comprehensive review for reporting, bearing in mind any differences in quality.17

Studies Reviewed

Twenty-nine systematic reviews fulfilled eligibility criteria after full text review. These eligible reviews addressed evidence for four chemoprevention interventions (antioxidant supplementation, calcium supplementation, non-steroidal anti-inflammatory drugs (NSAIDS), in the form of aspirin, but not COX-2 inhibitors and statins) and five screening interventions (breast self-examination (BSE), screening mammography, fecal occult blood testing (FOBT), flexible sigmoidoscopy (FS), and prostate specific antigen (PSA), for three cancers (breast, colorectal, and prostate).

No reviews were identified which examined evidence for magnetic resonance imaging in breast screening, ultrasound ovarian screening, colonoscopy as a stand-alone screening intervention, regular exercise, dietary interventions other than supplements (reviewed under chemoprevention), reduction in alcohol consumption, smoking cessation, seeking health care advice or referral for genetic counseling and/or genetic testing. As mentioned in chapter 2, we did not search for reviews for food, nutrition or physical activity interventions before 2006 as these were well evaluated in the World Cancer Research Fund/ American Institute for Cancer Research Second Expert Report.81

Of the eligible studies, data were extracted from the 10 reviews which represented the most comprehensive, up-to-date, and high quality evidence.1827 Twelve additional reviews reported overlapping data2839 but these data were not extracted. Five reviews4044 did not identify primary intervention studies despite being designed to do so; and two reviews45, 46 identified apparently relevant intervention studies but did not report usable data. The studies, which were not reported, on are listed in Table 1 in Appendix C.

The findings of the reviews which were included and reported are presented in Tables 7, 8, and 9. Some reviews addressed evidence related to prevention of more than one cancer type; in this situation, the results are presented separately by cancer type.

Table 7. Breast cancer preventive interventions.

Table 7

Breast cancer preventive interventions.

Table 8. Colorectal cancer preventive interventions.

Table 8

Colorectal cancer preventive interventions.

Table 9. Prostate cancer preventive interventions.

Table 9

Prostate cancer preventive interventions.

Breast Cancer

Five reviews on breast cancer prevention1822 are presented in Table 7 which synthesizes evidence in relation to chemoprevention (antioxidants and statins), screening (BSE), and screening mammography.

Chemoprevention. ‘Chemoprevention’ refers to the use of chemical compounds to arrest or reverse the earliest stages of carcinogenesis or development of pre-cancerous lesions. We apply it very broadly in this review to include the use of both recognized pharmaceutical agents (drugs) and supplements of naturally occurring elements and compounds administered in doses above those naturally encountered in typical diets, and/or administered in tablets or capsule.

Antioxidants. Antioxidants are molecules which inhibit or prevent damage to cells caused by oxidizing agents such as oxygen free radicals. It is suggested that such damage (oxidative stress) is an important early step in the development of many diseases, including cancer. It is hypothesized that antioxidant compounds could act as cancer prevention agents by preventing or inhibiting DNA damage, the earliest stage of carcinogenesis. Many observational epidemiological studies have found an inverse relationship between consumption of foods with high antioxidant content, such as fruits and vegetables, and incidence of cancer.7173

Antioxidants are a diverse group of compounds and different reviews have assessed them collectively and individually. Among the many antioxidants associated with food, the most commonly studied are vitamins (A, C, and E), their precursor molecules (e.g., alpha-tocopherol and ß-carotene), and minerals (e.g., selenium and zinc).

We report the findings of two reviews that examined antioxidants and breast cancer incidence. The first review18 evaluated supplements which contained any combination of ß-carotene, vitamin C, vitamin E, selenium, zinc, and other antioxidants, and the second review19 focused on vitamin E in any combination. Both of these reviews also evaluated outcomes related to colorectal and prostate cancer, the results for which are presented in the relevant sections below.

The first review18 included only placebo-controlled trials where antioxidants were given as supplements where the ingredients were fully disclosed, and which had followed up participants for at least one year. Their breast cancer analysis was based on 1,753 cancers in a total of 88,392 participants enrolled in the primary trials. No evidence was found of a protective effect of any combination of antioxidants against breast cancer (RR 1.00, 95 percent CI 0.90–1.09).

The second review19 examined vitamin E alone or in any combination. It included randomized controlled trials (RCTs) where vitamin E, in tablet or capsule form with or without other components, was evaluated against a control group receiving placebo or no intervention. They assessed the methodological quality of all included trials as high. They identified three trials, involving 62,158 participants, in which breast cancer incidence was reported as an outcome. They concluded that there was no evidence of a protective effect of vitamin E supplementation against breast cancer (RR 0.99, 95 percent CI 0.90–1.10).

Statins. Statins (hepatic 3-hydroxy-3-methylglutaryl coenzyme A [HMG-CoA] reductase inhibitors) are a class of lipid-lowering drugs that are widely prescribed for people at risk of cardiovascular disease. Interest in whether use of statins is associated with cancer risk was prompted by safety monitoring findings from cardiovascular prevention trials.91 Some studies suggest that they may also have a protective effect against cancer,9298 while others suggest the opposite.99101 Setoguchi et al., (2007)102 observed that longterm statin users tend to be healthier overall than non-users, and suggested that this might explain the positive associations.

Table 7 summarizes findings from the most comprehensive recent review20 that examined the evidence for statins in reducing the risk of a range of cancers, including breast, colorectal, and prostate (data for the latter two presented below). This review focused on placebo-controlled trials of any statin involving any population except participants being treated for particular high risk indications (e.g., familial hypercholesterolemia). The overall analysis indicated that statins appear to confer neither a protective nor a harmful effect on breast cancer incidence (RR 1.01, 95 percent CI 0.79–1.30). The findings were unaltered in sub-group analyses comparing lipophilic and lipophobic statins, and low, medium, and high potency statins.

Screening. Breast self-examination (BSE). It has been believed for many years that the practice of regular BSE allows women to detect breast tumors at an early stage, and thus to seek early treatment and improve their chances of cure. It is also considered inexpensive, non-invasive, and can be done in private.103, 104 Observational evidence suggests that women diagnosed with breast cancer who have practiced BSE are more likely to have found the tumor themselves, to have smaller tumors (on average) at the time of diagnosis, and to have benefited from longer survival.105 Critics of such analyses point to the possibility of lead-time bias, and the need to examine mortality rates as a more valid method of examining outcomes.

We report the findings of one systematic review21 which identified primary reports of three large randomized controlled trials, in two of which the intervention was teaching BSE in general populations and, in the third, clinical breast examination followed by teaching BSE. The latter was discontinued after the first screening round because of poor compliance, so data were available for only two trials. A total of 587 breast cancer deaths were observed in a total group of participants of 388,535 across the two trials. The relative risk of breast cancer mortality was 1.05 (95 percent CI 0.90–1.14). There was no statistically significant effect on the total number of cancers identified. There was a non-significant trend towards the detection of smaller tumors (carcinoma in situ RR 1.32, 95 percent CI 0.82–2.14, tumors ≤2cm RR 1.13, 95 percent CI 0.99–1.28). There was a statistically significant increase in total number of breast biopsies and biopsies with benign pathology (RR 1.53, 95 percent CI 1.47–1.60, and 1.88, 95 percent CI 1.77–1.99, respectively). These two large, longterm, population-based trials provide robust evidence that teaching women to perform regular BSE does not translate into a lower mortality rate from breast cancer, and is associated with a higher rate of invasive investigation.

Screening mammography . Mammography as an investigative technology for suspected breast pathology has been available for several decades, and has been gradually introduced, and evaluated, as a potential screening strategy for breast cancer. The rationale for mammography is that screening may detect tumors at a stage before they are palpable through self- or clinical examination, and that these smaller tumors are less likely to have become locally invasive or metastasized. Screening mammography aims to detect early malignant tumors and, if effective, would be expected to reduce breast cancer, and overall, mortality, but not breast cancer incidence.

Screening mammography has been the focus of a relatively large number of controlled trials in a range of countries, and between 1992 and 2002, 22 systematic reviews of screening mammography were published.106127 This level of scrutiny reflects ongoing controversies about the quality of the primary trials, and the possibility for harm which some experts consider inadequately examined and appreciated.106, 128

We summarize the findings of the most recent comprehensive review22 that attempts to address these issues through analyzing the outcome of all-cause mortality as well as breast cancer mortality, by sensitivity analyses according to adequacy of allocation procedures in primary trials, and by assessing rates of different treatment modalities in screened and control groups. The data are presented in Table 7.

This review examined data from 10 completed RCTs involving about half a million women. Of these, the reviewers considered only three trials to have been adequately randomized, and conducted a meta-analysis for these three separately. The trials covered a wide range of age groups, from 39 to 74 years, although most studies focused on women within the 40–59 age group. The reported screening intervals ranged from annual to about 2 years. In some studies, screening was accompanied by clinical breast examination, physical examination, or encouragement to perform monthly BSE. In most trials, the control intervention was usual care.

The overall analysis suggested that screening mammography is associated with reduced breast cancer mortality at 13 years (RR 0.80, 95 percent CI 0.73–0.88), but the association is more marked for the trials considered ‘sub-optimally randomized’ (RR 0.75, 95 percent CI 0.67–0.83) than for ‘adequately randomized trials (RR 0.93, 95 percent CI 0.80–1.09). When all cause mortality outcomes are considered (which would incorporate serious harms caused by over-treatment), the pooled estimates are very similar for trials considered ‘adequately randomized’ (RR 1.00, 95 percent CI 0.96–1.04) and ‘sub-optimally randomized’ (RR 0.99, 95 percent CI 0.97–1.01).

Colorectal Cancer

Six reviews are presented in Table 8 which synthesizes evidence in relation to colorectal cancer (CRC) prevention, four relating to chemoprevention (antioxidants, NSAIDS, and statins) and two relating to screening (FOBT and FS).

Chemoprevention. The rationale for chemoprevention is described in previous section on breast cancer.

Antioxidants. The background to antioxidants is described in previous section on breast cancer.

We report the findings of two reviews which examined antioxidants and CRC incidence. The first18 evaluated supplements which contained any combination of β-carotene, vitamin C, vitamin E, selenium, zinc, and other antioxidants, and the second19 focused on vitamin E in any combination. Further details of these reviews are included in the section on breast cancer.

The first review18 conducted an analysis of 1,523 cancers in a total of 178,086 participants enrolled in the primary trials. No evidence was found of a protective effect of any combination of antioxidants against CRC incidence (RR 1.00, 95 percent CI 0.90–1.10).

The second review19 identified four trials focusing on CRC. For vitamin E given in any combination (91,099 total participants), they found no evidence of an association with CRC incidence (RR 0.95, 95 percent CI 0.81–1.12). Similarly, there was no association when they examined data from two trials (24,114 participants) which evaluated vitamin E alone (RR 1.05, 95 percent CI 0.79–1.39).

Together, these reviews provide no evidence of a positive or negative association between antioxidant or vitamin E supplementation, and CRC risk.

NSAIDS. NSAIDs are a group of compounds which have an anti-inflammatory effect generally due to their action as non-selective inhibitors of the enzyme cyclooxygenase. Their potential protective effect against CRC has been observed in a number of case-control and cohort studies;129136 there is also evidence of their effectiveness in preventing recurrence of colorectal adenomatous polyps.137, 138

We identified two systematic reviews that examined one specific NSAID only, ASA, in CRC prevention. We report the findings of the single review that conducted a formal meta-analysis.23 This review identified two primary studies that evaluated CRC incidence (one at 5 years, the other at 10 years) in individuals at average risk (including no personal history of adenomas); one of these trials also reported CRC mortality and the other examined the incidence of colorectal adenomas. The intervention in both of these studies was ASA at doses recommended for cardiovascular protection (i.e., 325 mg every other day or 100 mg/day). Apparently no primary studies of standard doses (i.e., ≥325 mg/day) have examined cancer outcomes in average risk individuals. The combined number of participants in these two studies was 61,947. No statistically significant association was identified between low dose ASA and CRC incidence (pooled RR 1.02, 95 percent CI 0.84–1.25), CRC mortality (RR not reported) or adenoma incidence at 5 years (RR 0.86, 95 percent CI 0.68–1.10). These results provide no direct information on the effectiveness of either higher doses of ASA, or other NSAIDS, on colorectal neoplasia.

Statins. The background to statins is described in previous section on breast cancer. A comprehensive systematic review20 examined the evidence for statins in reducing the risk of CRC as well as breast (reported above) and prostate cancer (reported below). Data was synthesized for nine trials which examined CRC incidence, and one which evaluated CRC mortality, as outcomes. The overall analysis indicated that statins appeared to confer neither a protective nor a harmful effect on CRC incidence (pooled RR 1.02, 95 percent CI 0.89–1.16) or mortality (RR 0.33, 95 percent CI 0.07–1.63). The findings were unaltered in sub-group analyses comparing lipophilic and lipophobic statins, and low, medium and high potency statins.

Screening. Evidence from a wide range of studies139141 suggests that CRC results from complex interactions between genetic and environmental factors, and that most cancers evolve from small adenomas over a period of years.140, 141 The possibility of preventing CRC cases and deaths by early intervention to remove colorectal adenomas and/or early stage cancers has led to a large number of studies of a range of screening strategies, specifically involving FOBT, colonoscopy and sigmoidoscopy.142 There is no consensus on the optimum combination of these modalities, or on the ideal screening interval.

FOBT. The most widely used approach to FOBT is the stool guaiac test, in which the presence of heme (from haemoglobin) is indicated by a color change when hydrogen peroxide is added. This is a result of the oxidation of guaiac by the peroxide. A newer class of occult blood tests (immunochemical) rely on the detection of globin rather than heme, and it is suggested that these are more sensitive and specific than guaiac tests.143

A number of reviews examined FOBT-based screening strategies, of which one is reported here.24 This review examined RCTs only of FOBT (guaiac or immunochemical) in which at least two rounds of screening were compared with no-screening controls. This review analyzed data from four trials involving a total of 329,642 participants. A statistically significant effect of screening on CRC mortality was observed (pooled RR 0.84, 95 percent CI 0.78–0.90), which remained when the three trials with biennial screening were examined separately. No effect on all-cause mortality was observed (pooled RR 1.00, 95 percent CI 0.99–1.01), but a statistically borderline association with non-CRC mortality was noted (pooled RR 1.01, 95 percent CI 1.00–1.03). The interpretation of these combined results is difficult. It is argued that effective CRC screening would have little impact on all-cause mortality because CRC makes only a small contribution to overall mortality in these populations (and screening trials are therefore inadequately powered to detect such an effect). There is also an argument that biased attribution of cause of death between screened and control groups can lead to an overestimate of the true effect of screening on mortality, therefore an assessment of all-cause mortality would provide a more valid assessment of effectiveness.144

We do not present data from a further systematic review36 which focused on guaiac-based biennial FOBT screening alone; it included a non-randomized controlled trial excluded from the review reported above24 and excluded one of the latter's included trials. The pooled results are consistent with those reported above and in Table 8.

Flexible sigmoidoscopy. Flexible sigmoidoscopy (FS) is an endocopic technique which allows visualization of the colon and rectum distal to the splenic flexure. FS has a very low complication rate.145, 146 The majority of CRCs arise in the distal colon, thus are theoretically detectable with FS, and detection of distal adenomas is an indication for full colonoscopy.147 It is suggested that a combined strategy such as this can detect about 80 percent of CRC cases in men and about 50 percent of those in women, without recourse to more invasive colonoscopy as a primary screening modality.148150

We identified one review25 which examined the evidence for FS in CRC screening. One RCT was identified which compared FS in addition to FOBT against FOBT only (positive screens being followed up by colonoscopy). This review also considered FOBT alone as a screening strategy, but the data were not extracted as they are similar to the review reported above.24 This trial involved 10,978 participants and showed no statistically significant effect of the combined FS plus FOBT strategy on CRC mortality or incidence (RR 0.78, 95 percent CI 0.36–1.73 and 1.37, 95 percent CI 0.88–2.15, respectively). No formal intervention studies comparing FS alone with either no screening, or FOBT only, have been identified.

Prostate Cancer

Five reviews1820, 26, 27 are presented in Table 9 which synthesizes evidence in relation to prostate cancer prevention, four relating to chemoprevention (antioxidants, calcium, and statins) and one relating to screening (PSA, digital rectal examination (DRE), and transurethral ultrasound (TRUS) biopsy).

Chemoprevention. The rationale for chemoprevention is described in previous section on breast cancer.

Antioxidants. The background to antioxidants is described in previous section on breast cancer.

We report the findings of two reviews which examined antioxidants and prostate cancer incidence. The first18 evaluated supplements which contained any combination of β-carotene, vitamin C, vitamin E, selenium, zinc, and other antioxidants, and the second19 focused on vitamin E in any combination. Further details of these reviews are included in the section on breast cancer.

The first review18 conducted an analysis of 2,143 new cancers in a total of 55,709 participants enrolled in the primary trials. Overall, the pooled analysis indicated a small, non-statistically significant decrease in prostate cancer incidence when all antioxidants were considered together (pooled RR 0.87, 95 percent CI 0.74–1.02). Noting high heterogeneity, a sensitivity analysis suggested that vitamin E in particular was associated with a reduced risk (pooled RR 0.82, 95 percent CI 0.67–0.99), based on three trials. The authors of this review also noted that “one trial report a decreased incidence of prostate cancer with selenium,” although they did not provide relative risk data. Reporting on the same trial,151 another review26 (included below) estimated that, for a daily dose of 200μg of selenium (compared with placebo), the relative risk for prostate cancer incidence was 0.37 (95 percent CI 0.20–0.70).

The second review19 evaluated the effect of vitamin E alone or in combination, and performed a meta-analysis on three primary trials. For vitamin E given in any combination (71,759 total participants), the pooled estimate for effect on prostate cancer incidence also indicated slightly reduced risk (pooled RR 0.85, 95 percent CI 0.74–0.96). Two trials examined vitamin E alone, with no evidence of an association (pooled RR 0.86, 95 percent CI 0.70, 1.06). Similarly, there was no association when they examined data from two trials (24,114 participants) which evaluated vitamin E alone (RR 1.05, 95 percent CI 0.79–1.39).

Together, these reviews suggest no evidence of an effect of combined antioxidant supplementation on prostate cancer risk, but potentially a small protective effect of vitamin E supplementation.

Calcium. The association between calcium intake and prostate cancer risk has been examined both from risk increasing and risk reducing perspectives. Observational studies have indicated a positive association between calcium intake and prostate cancer, the suggested mechanism being that calcium lowers circulating vitamin D concentrations, and this in turn alters prostate cell proliferation.152155 In contrast, some studies have suggested that dietary calcium decreases prostate cancer risk.156162

We identified one review which analyzed intervention study data relating to calcium and prostate cancer risk.26 This review identified a single primary trial of calcium supplementation, designed to evaluate calcium as a protective agent against recurrence of colorectal adenomas, where prostate cancer incidence was evaluated as a secondary outcome.47 Supplements equivalent to 1,200 mg elemental calcium daily were given for 4 years and participants followed up for another 7 years. This trial observed 70 prostate cancers in 672 men with an RR for prostate cancer incidence of 0.83 (95 percent CI 0.52–1.32). Detailed review of the results suggested a statistically significantly lower incidence of prostate cancer in the supplement group until 2 years after supplementation was discontinued (RR 0.52, 95 percent CI 0.28–0.98), at which point the risk in both groups converged. It is suggested that calcium may have a slight protective effect, which is maintained only by ongoing supplementation. This finding from a single trial is insufficient to recommend calcium supplementation specifically for the purpose of prostate cancer prevention, particularly given the contradictory findings of previous observational studies.

Statins. The background to statins is described above in previous section on breast cancer.

We present data from the comprehensive systematic review described in previous sections,20 which examined the effectiveness of statins in reducing the risk of several cancers. Four trials examined prostate cancer incidence, and one evaluated prostate cancer mortality. The overall analysis indicated that statins appeared to neither increase nor decrease risk of prostate cancer incidence or mortality (pooled RR 1.00, 95 percent CI 0.85–1.17 and RR 0.99, 95 percent CI 0.14–7.01 respectively). No data were presented with regard to lipophilic versus lipophobic, or low, medium and high potency statins.

Screening. Prostate-specific antigen-based strategies (PSA). Strategies for screening for early prostate cancer have revolved around the combined use of PSA with or without DRE of the prostate, followed by needle biopsy guided transrectal ultrasound (TRUS). Digital rectal examination has limited sensitivity because it is not possible to palpate the entire prostate gland, while PSA testing produces high rates of false positive and false negative results.163 In addition, although reductions in prostate cancer mortality have been demonstrated with early treatment,164, 165 there remains considerable concern about lead and length time bias, the overtreatment of men who have indolent disease (tumors which were never destined to be fatal),166 and harms associated with treatment.167170

We identified a single review27 which examined the effectiveness of population-based screening in preventing death from prostate cancer. This study identified two trials of screening strategies that combined PSA testing, DRE and TRUS biopsy for diagnostic investigation, one of which used an annual, the other a 3 times yearly, screening cycle. The total number of participants randomized was 55,512, and 345 prostate cancer deaths were observed over followup periods of at least 11 years. No statistically significant impact of screening on prostate cancer mortality was found (pooled RR 1.01, 95 percent CI 0.80–1.29). No data were presented in relation to all-cause mortality. The authors of the review considered that both trials had a high risk of bias. The overall findings indicate that PSA-based screening cannot be considered to be an effective secondary prevention intervention in prostate cancer.

Quality Assessment of Studies

Standardized quality assessment checklists using a modified scoring version of the Oxman and Guyatt criteria82 were employed for all systematic reviews. The range of scores was 11–17 out of a possible 18. The major area of weakness was failure to describe adequate control of bias in the selection of studies for review.1820, 22, 23, 27 Other issues encountered in a minority of reviews were incomplete description of search methods,23, 27 and failure to describe criteria for assessing the validity of primary studies,20, 25, 26 or to cite the validity assessments of included studies.20, 26

Overall, the potential for bias in these reviews appears quite low. It is impossible to say whether failing to adequately describe search strategies, methods for controlling selection bias, or assessing validity of studies reflects methodological shortcomings or only failure to report these in published articles.

Conclusion

We were able to locate relevant systematic reviews relating to prevention, in average risk populations, of breast, colorectal, and prostate cancers, but not ovarian cancer. For all three cancers, the core primary prevention strategy for which reviews could be identified was chemoprevention. For breast and CRC, no evidence of an effect on cancer incidence of antioxidant supplements in general, vitamin E supplements in particular, or statins. For CRC, data on NSAIDS were available, with no evidence of an effect on cancer incidence. For prostate cancer, equivocal evidence was found of a possible protective effect of vitamin E supplements, selenium, and calcium supplements.

Screening strategies were also examined. For breast cancer, a review of three large population-based trials, confirm that BSE is not an effective strategy for reducing mortality from breast cancer and may increase morbidity through unnecessary investigations. Screening mammography has been evaluated in a large number of meta-analyses, which indicate that population-based screening appears to consistently reduce breast cancer mortality by about 15 percent, although there is still an open debate on whether all-cause mortality is a more valid measure of benefit of this intervention. There is concern that screening leads to higher rates of investigation and over treatment which undermine overall benefits. The analyses are based on studies with participants with a wide range of age ranges, which makes it difficult to discern the extent to which the profile of benefits and risks change according to target age. Both the technical performance of the screening test, in terms of sensitivity and specificity, and the prior risk of breast cancer, vary according to age, therefore the predictive value is not constant across all age groups. Also, the level of any risks associated with overtreatment will depend on local protocols for diagnostic investigation and treatment.

For CRC, FOBT-based screening strategies (which generally include diagnostic investigation using colonoscopy) appear to be associated with a decrease in CRC mortality, and the limited evidence available suggests that adding FS does not improve their effectiveness. As with breast screening, it is argued that all-cause mortality may provide a more valid assessment of screening effectiveness; however, since the proportion of overall mortality attributable to CRC is low, screening studies are generally underpowered to detect an effect.

With respect to prostate cancer, we found no evidence that PSA-based screening strategies are effective in reducing mortality.

Research Q3: For Those Interventions Identified as Being Based on Reasonable Evidence, What is the Evidence That Providing Information on Risk Status Results in Behavior Change or Increased Uptake of Services on the Part of Individual Patients?

General Approach

We reviewed published intervention studies (RCTs, controlled trials, and before-after studies) that examined the impact of systematic collection of family history information on one or more risk-reduction behaviors for breast, ovarian, colorectal, or prostate cancer. To be included, the intervention had to comprise systematic collection of individual family history information, and also communication of personal risk of one or more of the cancers of interest. This could be accompanied by individualized advice on specific risk reduction behaviors, although this was not necessary for a study to be included. The target behaviors of interest were both personal/lifestyle, and adherence to recommended clinical preventive interventions such as screening; the interventions were those considered standard of care when the primary study was conducted.

Studies Reviewed

Three studies4850 were retained for data abstraction after full text review. These studies all focused on breast cancer risk assessment and mammography screening, with or without other behaviors, as the target intervention (Table 10). Two of the studies were RCTs49, 50 and the other48 was an uncontrolled before-after study.

Table 10. Details of family history and personalized risk interventions.

Table 10

Details of family history and personalized risk interventions.

The sample sizes ranged from 188 to 2,076, and participants were recruited from a health maintenance organization (HMO),49 community pharmacies/health promotion events,48 and as first degree relatives (1DR) of breast cancer patients.50 Family history information was captured by computer-assisted telephone interview,50 postal questionnaire,49 and interviewer-administered questionnaire.48 In all three studies, information relating to personal medical history was collected as well as family history information. The 1DR and community pharmacy studies48, 50 specified use of the Gail model for risk calculation, which requires information on the number of 1DRs with breast cancer; the HMO study49 used a risk stratification algorithm developed in-house, which included information on first and second degree relatives with breast cancer.171

The HMO study49 had four groups, with two levels of collection of family history information (collected or not collected) and three levels of risk information included in screening invitation letters (no reinforcement, general risk messages, personalized risk messages) in the following combinations:

  • no collection of individual risk information plus generic invitation for mammography
  • no collection of individual risk information plus general risk messages embedded in invitation for mammography
  • collection of individual risk information plus general risk messages embedded in invitation for mammography
  • collection of individual risk information plus personalized risk messages embedded in invitation for mammography.

Data on the outcome of mammography uptake by 12 months were captured using the HMO's databases.

The 1DR study50 had two groups, both of which had individual family history information collected, and personalized information on risk of breast cancer fed back. The two groups had different levels of reinforcement of the importance of, and reminders to undertake, the target behavior of screening mammography. Data on the outcome of uptake of mammography was assessed by self-report at 12 months, captured by mail or telephone survey.

The community pharmacy study48 had one group of participants, all of whom had individual family history information collected and personalized information on risk of breast cancer fed back, along with reminders of recommended screening behaviors. Data on the outcomes of self-reported uptake of mammography, and adherence to BSE and clinical breast examination (CBE) recommendations, was captured at 6 months using telephone survey.

It should be noted that the two controlled studies (1DR and HMO49, 50) both examined different levels of personalization of risk information, but only the HMO study49 examined the capture of family history information as an intervention in itself. The community pharmacy study48 did not have a control group without family history capture.

Outcomes

The findings of the evaluation studies of the interventions described above are summarized in Table 11. The 1DR study50 showed a borderline statistically significant difference (P=0.05) in the change in mammography uptake (about 8 percent) between intervention and control groups. The other two studies were null. The community pharmacy study48 was the only one to examine other behaviors, and showed a statistically significant increase in self-reported BSE, but not CBE. The HMO study appeared to be adequately statistically powered; this was also likely the case with the 1DR study although sample size considerations were not discussed. Although the community pharmacy study indicated an a priori sample size calculation, their assumptions about baseline adherence rates may have been erroneous, as they were around 70–80 percent for CBE and mammography, much higher than published general population figures and suggestive of a possible ceiling effect.

Table 11. Evaluations of family history and personalized risk.

Table 11

Evaluations of family history and personalized risk.

All articles reported age-specific analyses, which generally did not show meaningful differences in effectiveness of the interventions. The HMO study49 also analyzed outcomes by breast cancer family history status (positive or negative) in the two groups that had been sent a risk questionnaire. While mammography uptake was similar between those receiving general risk and personalized risk invitations and who had a negative family history (41.4 and 39.7 percent, respectively), uptake was higher in women who had a positive family history and received a personalized invitation (66.7 percent) than women with a positive family history who received a generalized risk invitation (42.9 percent) (P=0.005). The results of the community pharmacy study48 were unchanged when the analyses were limited to the high risk participants only.

The studies vary in the extent to which their participants were representative of the general population. The HMO study was designed to be completely representative of its own patient population, which was described in previous publication as predominantly white, slightly better educated, and having a slightly different income distribution than Washington state as a whole.171 The proportion of participants with a positive first or second degree family history of breast cancer was about 20 percent, which is consistent with the North American female population. In contrast, the 1DR study was confined to women who had at least one first degree relative with breast cancer. The community pharmacy study drew participants from women attending pharmacies and heart health events, and no specific data are presented regarding representativeness. Their analyses indicate that 21 of 140 (15 percent) participants were assigned to the high risk category (≥1.7 percent risk of breast cancer in 5 years), which appears higher than would be expected for an unselected female population in this age group. Also, the high baseline rates of CBE and mammography compared with published figures for the general population may indicate that this study has limited external validity.

Quality Assessment of Studies

Standardized quality assessment checklists were employed on the two studies that used a randomized trial design.49, 50 The modified Jadad scores were 4 out of 8 for both studies.83 The main problem areas for both studies were failure to report measures to achieve blinding, and neither explicitly described randomization procedures or measures to conceal allocation. Both studies implemented the intervention through mail-outs to participants, asking them to take a specific action (schedule a mammography), and the possibility for contamination was probably rather low, particularly for the 1DR study. The potential for bias in these studies therefore appears quite low.

The third study48 was described in the report's abstract as a ‘randomized, paired, pre-post study’, which is misleading. In our assessment, it was an uncontrolled pre-post study in which before-after outcomes for individual participants were analyzed as paired data. No control group was used and therefore no random allocation was possible. The potential for bias in this study is high, given that no assessment could be made of the influence of external factors, or placebo or Hawthorne effects.

Conclusion

Taken together, the three studies identified neither support nor refute the hypothesis that taking family history and using it to personalize risk of breast, ovarian, colorectal, or prostate cancer promotes lifestyle changes which reduce cancer risk, and/or greater adherence with preventive clinical interventions. All three studies focused on breast cancer, the interventions were heterogeneous, some including components beyond family history taking and personalization of risk. The interventions generally did not resemble the routine, personal interaction which might occur between a primary care professional and an individual patient, and the methodological rigor of the evaluations was variable.

The HMO study49 was embedded in a routine screening invitation system, and therefore resembled regular clinical practice, albeit in a non-personal way. The evaluation was well-designed and had a large sample size. This study (the oldest of the three reviewed) provides no evidence that personalization of risk information would be an effective overall strategy, but suggests that it might be worth exploring as a way of promoting risk reduction behavior in high risk sub-groups.

The 1DR study50 provided evidence of a possible modest effect on mammography uptake of personalizing risk information for a group already likely to have higher than average personal risk perceptions. The intervention had several components, and it is impossible to separate out the individual effects of the family history collection, the personalization of risk information, and the materials designed to reinforce the importance of mammography. Even this thoughtfully designed intervention produced only a small increase in screening behavior. The transferability of the approach to a clinical setting, and the absolute size of benefits that would be achieved, is unclear.

The community pharmacy study48 showed some evidence that personalized risk information could promote cancer prevention behavior, but the lack of a control group, the questionable validity of the outcome measurement, and the likely selection bias of participants all make it impossible to judge the wider applicability of its findings.

Research Q4: What are the Harms or Risks to Individual Patients That may Result From the Collection of Family History Information in Itself, and/or the Provision of Family History-Based Risk Information?

General Approach

We reviewed published intervention studies, (RCTs, controlled trials, and before-after studies) which assessed negative impacts of systematically collecting family history information and providing patients with risk information for the cancers of interest based on their family history. The focus was on systematic collection of individual family history information, and communication of personal risk of one or more of the cancers of interest, in populations considered representative of primary care populations. Specialist genetic counseling (with or without genetic testing) for patients selected because of possible genetic risk was excluded from this definition.

The outcomes of interest were impaired quality of life, negative psychological effects (cognitive, affective, or behavioral), social impacts (e.g., negative impact on family well-being, discrimination, stigmatization), which could be directly attributed to this intervention, not subsequent clinical or other preventive activities.

Studies Reviewed

One study was identified which met all eligibility criteria.51 This was an uncontrolled before-after study designed to evaluate the psychological impact of providing family history information and receiving a personalized risk assessment. Patients aged 35–65 registered with a single family doctor's office were invited to complete a postal cancer family history questionnaire, and were provided with individual feedback on their genetic risk of CRC and breast cancer (where appropriate). General anxiety and cancer worry was assessed at baseline and 4 to 6 weeks after risk information feedback using the Spielberger State-Trait Inventory (STAI) and a multidimensional cancer worry scale, respectively. Details of the study are summarized in Table 12.

Table 12. Harm/risk of family history collection.

Table 12

Harm/risk of family history collection.

Outcomes

This study analyzed participants in two groups. Firstly, ‘lower risk’ (those at no more than slightly elevated risk) participants, for whom no followup was necessary, were given feedback by letter only. No statistical difference was observed in anxiety and most other cancer worry measures following the intervention, with the exception of a small reduction in participants' perception of their own risk (P<0.01).

Of the remaining participants, most were interviewed to clarify details of the family history, which led to further designation into ‘higher risk’ and ‘false positive’ groups (the latter comprising patients deemed not actually to be at high risk after further enquiry). For both ‘higher risk’ and ‘false positive’ groups, no difference between baseline and followup responses to general anxiety and cancer worries scales was observed. However, both of these groups showed higher baseline cancer risk perception scores compared to the lower risk group (P<0.001 for ‘higher risk’ group and P=0.003 for ‘false positive’ group).

Overall, the findings suggested no association between the exercise of capturing family history information and feeding back personalized risk, and anxiety or cancer worry, in patients who are close to average risk. In fact, it is possible that the intervention may have led to more realistic (lower) perceptions of personal risk. In contrast, the higher anxiety and cancer worry outcomes in the ‘true’ and ‘false positive’ high risk groups may reflect their baseline (pre-intervention) status rather than an effect of family history taking.

Conclusion

The evidence base for addressing (Q4) is limited to a single study. It suggested that structured family history collection and feedback of breast cancer risk information had no deleterious psychological effects on any of the risk groups, and in women, who were not at high risk, may have led to appropriate reductions in perceived risk. Higher average baseline levels of anxiety and cancer worry in the groups who went on to have further assessment may reflect pre-existing concerns about a positive family history and need to be taken into account when family history interventions are evaluated.