• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information

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.)

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

Clinical Utility of Cancer Family History Collection in Primary Care.

Show details

4Discussion

The purpose of this review was to establish the evidence base to answer the question, “In relation to the cancers of interest, would routinely taking and using family history for risk assessment in primary care settings be likely to lead to net health benefits?” Acknowledging the scope of this question, the evidence was assembled across a number of subsidiary questions, addressed individually below.

Throughout the review, the focus was the primary care context. This led to two decisions which underpinned the review's methodology, specifically the eligibility criteria. Firstly, across all questions, the populations studied had to reflect primary care populations. In practice, this meant that populations who already had cancer, a pre-cancerous condition, or who were suspected of carrying a genetic risk, were excluded. Secondly, studies of family history taking as a primary care intervention, i.e., as an intervention in and of itself, were included, but those where family history taking was approached as a specialist activity, and/or embedded within a larger set of clinical activities such as assessment for genetic testing, were excluded.

We also drew a distinction between “taking family history” as a distinct activity practiced by health care providers (of central interest in this review) and “being aware of a positive family history” as an attribute of study participants. A patient's ability to accurately report family history information is a prerequisite for valid family history collection. However, some people may also have a pre-existing perception of an unusually high family prevalence of cancer, leading to this being itself a cause of anxiety, quite separate from any effect of the clinical activity of taking a family history. Thus, evaluations of family history taking as an intervention need to be carefully designed to take account of this complicating factor.

Risk Stratification Algorithms

Many cancer risk stratification algorithms, models and systems exist, and the goal of this review was to identify which of those, based on family history information, performed well in primary care type populations. In approaching this question, we sought to identify evaluations of frameworks devised specifically for primary care, or which might be transferable to primary care even if originally designed for other purposes.

Further, we were interested in evaluations which assessed the ability of a system to correctly predict risk in individuals, not simply on studies of overall associations between family history and disease incidence at a population level. As explained in Chapter 3, to be suitable for implementation in routine clinical settings, a risk prediction system needs to differentiate between the disease risk of individuals, such that it consistently predicts higher risk of cancer in those who are truly destined to go on to develop the disease than in those who will not.

Review question (Q) 1 was therefore tightly focused on evaluations of models predicting individual risk, and this led to the exclusion of a large number of analytical epidemiological reports, descriptive clinical studies, and validation studies of models which did not present data on individual discriminatory accuracy (e.g., Constatino et al., 199984). We were able to identify evaluations of only two distinct approaches, one a family of models based on the Gail model15 (developed for breast cancer risk prediction), and the Harvard Cancer Risk Index16 (designed to predict the risk of a range of cancers, although validation data were available only for CRC). Both included a range of variables in addition to family history.

The Gail model is publicly available on the National Cancer Institute's web site (www.cancer.gov/bcrisktool/). It should be noted that its original purpose was epidemiological - to facilitate the design of clinical trials by permitting sample size calculations to be based on improved assumptions about expected disease outcome rates - rather than for clinical decisionmaking.15 Evaluations of this model indicated excellent performance at the population level, i.e., calibration, for predominantly white U.S. and Italian populations, judged by the ratio of expected to observed cases in the population studied. The model performed less well in more diverse U.S.13 or African American10 populations. The CARE model also showed good calibration.7

However, for all of these models, evaluations indicated generally modest discriminatory ability, with the c-statistic ranging from 0.55–0.59. Many women who would be judged to be high risk by one of these models would not go on to develop breast cancer, and vice versa.

As a contrast, family history-based risk stratification has been found to perform much better as a predictor of coronary artery disease, with a c-statistic of 0.87.173 While family history can provide some useful predictive information on some common health conditions, the situation for breast cancer is less clear-cut. Although there are breast cancer family syndromes, these are associated with only about 5 percent of breast cancer cases. Over ninety percent of the incidence of breast cancer at a population level is not associated with a distinct familial pattern, and many women appear to develop breast cancer ‘out of the blue’. The known general risk factors - e.g., younger age at menarche, and later age at birth of first child - each contribute only a little to overall incidence of the disease (individual relative risks are modest), and they are reasonably prevalent in most populations

Had the review found greater evidence of adequate discriminatory accuracy for some of these tools, this would not guarantee that their use would lead to better health outcomes in practice. A number of other conditions would need to be satisfied, for example evidence that different risk categories are matched with evidence on appropriate risk-specific preventive interventions. We note, for example, that the Gail and related models have been used primarily for assessing eligibility for cancer chemoprevention trials, although their use as more general clinical predictive tools is implied by their wider availability to the professional community and the general population. Stronger evidence is needed on the application of the tools in settings closer to routine clinical practice, and trials are required which directly assess the outcomes from risk assessment combined with risk-appropriate preventive interventions, not just assessments of the technical accuracy of risk prediction. Also, the use of tools needs to take into account standard of practice for the particular clinical context. As a hypothetical example, if guidelines were to recommend colonoscopy and polypectomy for all individuals with a family history of colorectal cancer, and this was widely implemented, and it led to a reduction in absolute colorectal cancer rates in the general population, then evaluations of risk prediction algorithms would need to consider the power implications of using cancer incidence as a primary outcome.

Since the overall purpose of this review was to assess elements of the clinical utility of family history taking, we would ideally have liked to evaluate cancer risk stratification systems based solely on family history. At the time of writing, perhaps the prototypical system of interest is Family HealthwareTM,174 a Web-based tool designed to assess a person's familial risk for coronary heart disease, stroke, diabetes, and colorectal, breast, and ovarian cancer. It also provides users with personalized recommendations for lifestyle changes and screening. At the time of writing, evaluation data for this tool were not available.

In the absence of validation studies of family history-based systems, we extended the review to include tools with a family history component. Despite this, a large number of predictive tools were excluded, either because they did not include any history items, some items would not be available for all patients, or would not routinely be available to a primary care practitioner. Further details of the wider range of risk prediction systems can be found at http://riskfactor.cancer.gov/cancer_risk_prediction/.

Cancer Prevention Interventions

Review (Q2) was an assessment of the overall benefits and harms of available preventive interventions for breast, ovarian, colorectal, and prostate cancers. As noted in Chapter 1, answering this question was an essential step in addressing the overall question of clinical utility of cancer family history taking, but cancer prevention in general was not the primary focus of this report. With a focus on published reviews of evidence, we again applied the criterion of primary care applicability in terms of unselected populations, which would be expected to contain individuals at high, medium, and low risk. Reviews were excluded only where they specifically focused on studies of high risk populations such as those affected by cancer or known pre-cancerous conditions, or people at known or suspected high risk of an inherited genetic disorder. Thus, while all reviews focused on “general” populations, none specified having relatives with cancer (but not suspicion of inherited genetic disorder) as an exclusion criterion. Most did not in fact specify this one way or the other, except one23 which explicitly stated that studies with participants with a positive family history of cancer were eligible. In the end, none of the reviews actually reported results separately for studies of participants with affected relatives, likely reflecting a lack of such primary studies.

We also decided to focus on reviews of intervention studies, noting (for example) how apparently clear-cut evidence from observational epidemiological studies in the areas of beta carotene and lung cancer7173 and hormone replacement therapy and breast cancer175 has been contradicted by subsequent randomized controlled trials.74, 176 Where more than one eligible review was identified, we included the data from the most recent and highest quality review. We found no reviews relating to primary or secondary prevention of ovarian cancer.

With respect to primary prevention of the cancers of interest, we found a striking lack of experimentally-derived evidence. We do not suggest, however, that this lack of evidence of effectiveness means that the interventions examined are ineffective. A number of issues and limitations must be considered.

Firstly, there is the difficulty of translating the exposures examined in many observational studies into implementable interventions in trial settings. Observations on the potential protective effect of specific antioxidants, micronutrients, and vitamins have generally been derived from analyses of dietary habits or supplements containing multiple vitamins and minerals, whereas intervention studies generally investigated the effects of one or two supplemental agents in factorial trials. Moreover, in some instances, notably the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study, the doses of these agents were substantially higher than the exposures obtained through dietary and multivitamin supplements. Hence, few intervention trials have investigated ‘whole diet interventions’. To take calcium as an example, pooled analyses of cohort studies indicate an approximately 15 percent reduction in risk of colorectal neoplasia associated with higher dietary calcium intake,45 but no reviews were available of dietary calcium as an intervention in cancer prevention. Also, chemoprevention interventions may act early in cancer development, meaning that preventive effects may only be observable in long-term studies, not reflected in the followup timing reported in many of the primary trials contained in the reviews.

A second issue is the baseline population risk, and the power of meta-analyses to detect an intervention effect. Taking calcium as an example again, we note that reviews which have conducted pooled analyses of studies conducted in higher risk populations (e.g., patients with adenomatous polyps, therefore excluded from the present review) have demonstrated an apparent 20 percent lower recurrence rate amongst those randomized to calcium supplementation than those on placebo.177 It is perhaps worth noting that evidence of possible preventive effects of vitamin E19 and of calcium26 was noted in relation to prostate cancer; although the latter review met our (Q2) eligibility criteria in that the primary studies had recruited participants not at high risk for prostate cancer, the data came from a secondary analysis of a study with participants at elevated risk of colorectal cancer (the prostate outcomes being a secondary analysis).47 There is a current stream of thinking suggesting that there may be some common etiological pathways for prostate and colorectal cancer.178, 179 This might mean that these apparently unselected participants were de facto at elevated risk of prostate cancer because they were at elevated risk of colorectal cancer.

The focus on reviews also meant that we did not include large primary studies of interventions which have not yet been the subject of reviews. The report finds itself in the anomalous situation where some included reviews were based on analyses of single trials (and therefore, effectively present single trial data), but where interventions evaluated in large single trials were excluded on the basis of ineligible design (i.e., not a review). It was not possible, at the outset, to predict the results of the literature search for review (Q2) and to anticipate this situation. Extending (Q2) to include primary studies would have been a significant undertaking and was beyond the resources available for this review. We acknowledge that this has probably resulted in the exclusion of data on cancer prevention interventions which suggest protective effects. For example, one study180 was legitimately excluded, which reported a large combined analysis of two selected RCTs along with a systematic review restricted to observational studies. This study was not eligible because the systematic review element excluded intervention studies, and the two RCTs reported represented selective reporting of all relevant trial evidence. The two RCTs were designed primarily to assess the effects of acetylsalicylic acid (ASA) on non-cancer outcomes181, 182 but had extended followup to examine the effects on cancer.180 The pooled analysis of the two trials (ASA assigned at doses of 300mg, 500mg or 1200mg per day) suggested a protective effect on colorectal cancer incidence at ten years (pooled relative risk 0.74 (95 percent CI 0.56–0.97)). The magnitude of effect was consistent with observational studies, and no effect on the incidence of other types of cancer was observed.

Despite extensive investment in cancer research over the last four decades, underlying mechanistic pathways for individual cancers still remain to be fully elaborated. Thus, the ‘risk factor’ approach of many analytical observational studies is an insufficient basis for drawing definitive conclusions on biological mechanisms of cancer causation and prevention. Emphasis is now being placed on large-scale prospective cohort studies with associated biobanks that hold promise of improved exposure assessment, and that will enable joint effects of genes and exposures to be investigated with adequate statistical power.183

The evidence regarding screening for the cancers of interest provided greater clarity than that for primary prevention. As for secondary prevention, we found no reviews of screening or surveillance interventions relating to ovarian cancer. Regarding breast cancer, there appears to be clear evidence, on the basis of a pooled analysis of three large population-based trials, that teaching women breast self-examination and encouraging them to carry it out regularly has no measurable effect on breast cancer mortality. In some countries, the emphasis has shifted from exhorting women to examine their breasts monthly to becoming ‘breast aware’ - being familiar with what is ‘usual’ over a monthly cycle and therefore being able to identify and act on what is not ‘normal’.184

A large number of reviews of population-based mammography screening have been conducted, and the evidence seems to suggest that it reduces breast cancer mortality by about 15 percent. The evidence is most clear cut for women aged 50 and over, the impact being limited in younger women by greater breast density and lower baseline risk. There is debate about the validity of using cancer-specific mortality as the primary outcome for evaluations of mammography screening; arguments revolve around the possibility of bias in the coding of death between screened and control populations, and the notable lack of impact of screening on all cause mortality. Where trials have demonstrated a reduction in mortality, there is no suggestion that annual screening offers benefits over biennial screening in women aged 50 or older. This also holds for younger age groups, although the lower sensitivity of the test and evidence of the more rapid growth of tumors in younger women have led some experts to suggest more frequent screening to maximize sensitivity.185, 186

Regarding colorectal cancer, the evidence appears to indicate that fecal occult blood testing (FOBT) and followup with colonoscopy reduces colorectal cancer mortality. The very limited evidence (one trial) suggests that flexible sigmoidoscopy does not improve this outcome, and there is no consensus on the optimum screening interval. Observers have also noted the lack of impact of screening on overall mortality, and some analyses have indicated an increase in non-colorectal cancer mortality. Although a 2008 U.S. Preventive Services Task Force (USPSTF) report187 contains a Grade A recommendation which includes colonoscopy as a screening intervention for colorectal cancer in individuals aged 50–75 years, the background evidence report188 indicates that no trials of colonoscopy as a stand-alone screening intervention were identified.

We found no evidence of an effect of prostate specific antigen (PSA)-based screening strategies on prostate cancer mortality. In a recent USPSTF review, two RCTs were identified that did not show a mortality benefit from PSA screening independently or in a meta-analysis; important flaws in design and analysis were noted.189 The USPSTF review also identified one cross-sectional and two prospective cohort studies of possible psychological effects of PSA screening results. These suggested that false-positive PSA screening results cause psychological adverse effects for up to 1 year after the test.

Although we sought reviews of ultrasound screening for ovarian cancer as an intervention, we did not consider reviews of other interventions such as CA-125 screening. In retrospect, it would have been legitimate to include this since the report also reviewed PSA-based screening for prostate cancer, which is an analogous serum-based cancer screening test. We note that the USPSTF positively recommended against screening general using CA-125 in a 2004 report,190 suggesting that the possibility of earlier detection would lead to small effects, at best, on mortality (because of the low prevalence of ovarian cancer therefore low positive predictive value), and fair evidence of important harms because of the invasive nature of diagnostic investigations.

We also did not review genetic testing as an intervention, considering this not to be in itself a preventive intervention, but we did include the broader interventions of referral for genetic counseling and/or testing as interventions of interest applied to populations not considered high genetic risk. However, no reviews on these interventions were identified.

Family History-Based Risk Assessment and Individual Preventive Behaviors

It is postulated that knowing one's genetic risk of disease, whether through a genetic test or family history, can provide a motivation to comply with advice on preventive interventions191 and some descriptive studies suggest that people who have a family history are overrepresented in studies of screening adherence.192198 Noting concerns discussed in Chapter 1 about distinguishing between the behavior of people motivated by pre-existing perceptions of elevated disease risk because of living with a “family disease” from the effects of family history-based clinical strategy, we focused on intervention studies where the effects of confounding would be less evident. We found only three relevant primary studies, all of them relating to breast cancer prevention. One of them sampled women known to have a first degree relative with breast cancer50 (although not, by definition, formally identified as being at elevated genetic risk themselves), one was health organization-based,49 and the third likely comprised participants who were more than typically interested and generally compliant with some screening recommendations.48 The only study which in any way replicated a primary care consultation involving personal interaction between a health care professional and an individual patient was the third of these studies, which was in the setting of a community pharmacy,48 but was limited by its uncontrolled design. The participants appeared to have had an atypically high average adherence rate with screening mammography recommendations at baseline, which could have resulted in apparent lack of effect because there was little room for improvement (a possible “ceiling effect”). The overall evidence was therefore equivocal, neither confirming nor undermining the hypothesis that systematic feedback of risk motivates compliance.

Direct Harms or Risks From Family History-Based Risk Assessment

All health care interventions should be assessed for evidence of harm as well as benefit. We identified only one study relevant to family history taking that was conducted in an unselected primary care population. The respondents who were not found to be at elevated cancer risk had no evidence of adverse psychological outcomes, and in fact there was some indication that the assessment was beneficial in that it promoted more realistic personal risk perceptions. In contrast, participants whose initial assessments indicated potential elevated risk had higher baseline anxiety levels than those whose initial assessments indicated population risk, irrespective of their final risk assessment. In other words, both “true positives” and “false positives” had higher average pre-test anxiety levels which might suggest that perception of family history-associated cancer risk (whether confirmed or not) rather than collection of family history information might be associated with higher levels of anxiety.

This single study is insufficient to conclude that family history taking as a deliberate clinical strategy is, in itself, likely to be harmful in terms of emotional impact, but it is consistent with findings from studies of genetic testing. It suggests that assessments of psychological status might be appropriate before embarking on family history-based risk assessments in order to identify those individuals who might be most at risk of ongoing anxiety or cancer-related worry, and who might therefore warrant extra support or counseling, irrespective of their actual assessed disease risk.

Limitations

The eligible studies within this systematic review were limited to primary studies and systematic reviews in the English language. We restricted the search of systematic reviews to 2003 forward, to ensure that only relatively recent reviews were selected. The review was also limited to studies in adults; therefore no conclusions can be drawn with respect to children or young people specifically.

The effectiveness of family history-based tools and interventions are dependent on the accuracy of reporting of family history, and it is impossible to comment on this aspect of the topic. We did not restrict studies according to the manner in which cancer family history was collected and considerable variation in the methods used was observed. Almost universally, studies depended on self-report methods and are therefore dependent on the individual respondents' knowledge of their history. This represents a limitation on family history taking in practice rather than a limitation specifically of the review, and was explored in a previously published review.5

In examining the effects of family history taking on behavior (Q3), the eligibility criteria specified the intervention as feeding back family history-based risk alone, or with risk advice. We did not examine taking family history itself as an intervention without some element of feedback to a patient. The studies identified also evaluated interventions which were not terribly reflective of day-to-day primary care practice. It is therefore impossible to comment on whether the capture of family history information might lead a practitioner to consider different preventive strategies, or the incorporation of family history information into the broader knowledge of a patient might lead to changes in the nature or emphasis of preventive advice which is offered. The emphasis on very specific clinical behavioral outcomes also does not allow for exploration of other effects on the part of patients, e.g., seeking out more extensive information from family members as a result of having been asked “the first” set of questions on family history.

In regard to cancer prevention interventions, we were able to provide only an overview of classes of interventions and could not examine differences in their individual implementation (e.g., different doses). For reviews that assessed the same intervention, we selected the single best review based on several factors including number of studies and year of publications and methodological quality of the review.17 We did not contact authors for additional clarification for QUOROM requirements. Neither did we re-abstract or re-analyze original randomized trials eligible within the systematic reviews. In addition, as discussed above, the emphasis on reviews inevitably resulted in the exclusion of RCTs which had not been incorporated into reviews, including large studies such as the Women's Health Initiative trial of a dietary modification program on the incidence of colorectal cancer in post-menopausal women.199 We believe that, where interventions have been evaluated in primary trials which have then been included in published reviews, we have captured and reported the effectiveness evidence objectively; however, we believe that relevant effectiveness evidence for some interventions, based on well-designed trials, has been missed wholesale because primary intervention studies were not eligible for (Q2).

Conclusions

1.

The evidence for the predictive accuracy of algorithms in primary care populations was very limited. Although many tools were identified that incorporated some family history information, no evaluations of solely family history-based tools were identified. The tools on which it was possible to comment related mainly to breast cancer.

Recommendations for future research:

  • The actual performance of tools based only on family history should be formally examined in primary prospective studies, and/or in secondary analysis of large cohort studies.
  • The performance of individual family history components of different risk stratification models which use a wider range of factors (including those examined in this report) should be examined separately from the non-family history components, in order to determine whether they provide sufficient predictive power in the absence of the non-family history factors.
  • For clinical relevance, the focus of validation should be discriminatory accuracy at the individual patient level.
  • More definitive evaluation should examine the effect on health outcomes when risk stratification systems are used in combination with preventive interventions, in actual practice settings. This cannot be done with secondary analyses of observational data and requires well-designed intervention studies.

2.

The evidence establishing the efficacy of interventions for primary and secondary prevention based on systematic reviews of randomized or controlled clinical studies in unselected populations is very limited. Interventions for which there were reviews include chemoprevention (antioxidants, calcium, NSAIDS, and statins) and screening interventions (BSE, mammography, FBOT, flexible sigmoidoscopy, and PSA) for breast, colorectal and prostate cancers. No reviews were found for ovarian cancer. It is likely that this review excluded effectiveness data available from RCTs of interventions which have not yet been the subject of systematic reviews.

Recommendations for future research:

  • The large amount of evidence on potential primary cancer preventive interventions obtained from observational studies of cancer risk factors should continue to be further evaluated in well-designed randomized controlled trials.
  • Further systematic reviews should be conducted to examine the full range of potentially preventive interventions

3.

Within primary care populations, there is very limited evidence to support or refute the effect on risk-reducing behavior changes (e.g., lifestyle changes or uptake of recommended clinical interventions) of taking a family history and using it to personalize risk of breast, ovarian, colorectal, or prostate cancer.

Recommendations for future research:

  • Well-designed trials are required that compare family history-based, personalized risk advice with standard of care on risk reducing behaviors in populations at different risk levels (including population risk).

4.

In primary care populations, there is very limited information to evaluate direct harm incurred from the routine practice of taking family history and using it to personalize risk information.

Recommendations for future research:

  • Trials of family history taking as an intervention should include capture of data to examine the full range of potential impacts on individuals. Baseline data on psychological status should be captured so that this can formally be adjusted for in outcome analyses.

5.

Research on the use of family history tools, risk stratification systems, and family history-based personalized prevention advice should take into account evidence on the factors likely to promote their effective use in practice, such as the educational needs of primary care practitioners and issues which act as barriers or constraints to their implementation in practice

PubReader format: click here to try

Views

  • PubReader
  • Print View
  • Cite this Page

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...