Cancer Risk Assessment Tools in Primary Care: A Systematic Review of Randomized Controlled Trials

Abstract

PURPOSE

We conducted this review to identify published randomized controlled trials (RCTs) of cancer risk assessment tools used in primary care and to determine their impact on clinical utility (clinicians), screening uptake (patients), and psychosocial outcomes (patients).

METHODS

We searched EMBASE, PubMed and the Cochrane databases for RCTs of cancer risk assessment tools in primary care up to May 2014. Only studies set in primary care, with patients eligible for screening, and English-language articles were included.

RESULTS

The review included 11 trials of 7 risk tools. The trials were heterogeneous with respect to type of tool that was used, type(s) of cancer assessed, and outcomes measured. Evidence suggested risk tools improved patient risk perception, knowledge, and screening intentions, but not necessarily screening behavior. Overall, uptake of a tool was greater if initiated by patients, if used by a dedicated clinician, and when combined with decision support. There was no increase in cancer worry. Health promotion messages within the tool had positive effects on behavior change. Trials were limited by low-recruitment uptake, and the heterogeneity of the findings necessitated a narrative review rather than a meta-analysis.

CONCLUSIONS

Risk tools may increase intentions to have cancer screening, but additional interventions at the clinician or health system levels may be needed to increase risk-appropriate cancer screening behavior.

INTRODUCTION

Cancer screening programs have been introduced in many countries for breast,¹ colorectal,² and cervical³ cancer. With the growing recognition of the potential harms from population-based cancer screening programs,⁴ risk-stratified screening is being proposed as a way of reducing harm and focusing on populations at higher risk of cancer. This concept can also be applied to primary preventive measures, especially as the evidence to support chemoprevention for common cancers such as breast and colorectal builds.^5,6 If risk-stratified cancer prevention is to be implemented, it requires risk assessment tools that can be used in primary care to identify those most likely to benefit from tailored prevention.⁷

Cancer risk prediction models, based on epidemiologic data, calculate an individual’s likelihood of developing cancer, identify an individual’s risk of carrying a genetic mutation for a specific cancer (eg, BRCA 1 or BRCA 2), or both.^8,9 Newer risk models are beginning to incorporate genomic profiles and environmental exposures,¹⁰ a trend that is likely to grow with the movement toward precision medicine.¹¹ Risk assessment tools facilitate the translation of these risk models to estimate an individual’s likelihood of developing different cancers by assessing the combination of risk factors including genetic, environmental^12,13 and behavioral¹² risk factors. Examples include the US National Cancer Institute (NCI) colorectal cancer risk tool,¹⁴ which incorporates the risk model developed by Freedman et al¹⁵; the NCI breast cancer risk tool, which applies the Gail breast cancer risk prediction model¹⁶; and MelaPRO for assessing risk of melanoma.¹⁷

Primary care has an important role in the delivery of cancer screening programs and can increase screening uptake.¹⁸ Successful implementation of risk assessment tools into primary care is needed if risk-stratified cancer prevention and the promises of precision medicine are to be achieved.

In this article, we report the first systematic review of randomized controlled trials (RCTs) that have tested cancer risk tools in primary care. The review specifically investigated measures of clinical utility such as clinician referrals and patient cancer screening behaviors, as well as psychosocial outcomes.

METHODS

This systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Table 1)¹⁹ and is registered with Prospero (registration number: CRD42014008892).²⁰

Table 1.

The Systematic Review Question Design According to the PRISMA Guidelines¹⁹

Population	Intervention	Comparison	Outcomes	Study Design
Main concept
Primary care practitioners Primary care patients	Cancer risk assessment tool to determine a primary care patient’s individual risk of cancer	Standard clinical care	Clinicians  Clinical outcomes including appropriate referral behavior  Patterns and accuracy of risk perception  Cancer knowledge  Frequency of use  Acceptability by physicians  Confidence of use by clinicians  Attitudes to the tool Patients  Patient cancer anxiety/worry  Acceptability by patients  Patient behavior including uptake of secondary referral behavior  Adherence to screening recommendations  Intention to undergo screening  Satisfaction with consultation	Randomized controlled trials
Synonyms/search terms
Primary care Primary care clinicians Primary care physicians Family practice General practice GPs Patients	Risk-assessment tool Clinical tool Risk-prediction tool Decision-support tool Risk-assessment model Computer decision-support tool Adult population Cancer Family history [and synonyms for family]	Standard care Usual care	Acceptability Effectiveness Frequency of use Referral data Appropriateness of management Risk accuracy Patient risk perception Psychosocial outcomes Cancer worry Patient behavior	–

GP = general practitioner; PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

We searched the PubMed, EMBASE, and Cochrane databases for English-language articles published up to May 2014, focusing on search terms based on the concepts of “risk assessment tools,” “cancer,” “primary care,” and outcomes such as “cancer worry,” “risk perception,” “clinician confidence,” “referral behavior,” and “screening behavior.” Additional articles were identified through citation tracking and reference checking.

Eligibility Criteria

Studies were ineligible if they involved tools that did not estimate cancer risk, assessed prognostic tools for patients with an existing cancer diagnosis, were not implemented in a primary care setting, or did not evaluate the tool using RCTs (Figure 1).

Figure 1.

Article selection for inclusion.

The populations studied included primary care clinicians (general practitioners, family physicians, and community medicine clinicians) and patients of primary care clinicians, as long as they were adults without an existing (known) cancer diagnosis (Table 1).

Study Appraisal and Synthesis Methods

The primary author (J.W.) assessed all citation abstracts, which were reviewed by a second author (M.P.). Two other researchers (P.P.C.C., S.L.) assessed full-text articles. Data were extracted and studies were critically appraised for bias by 3 reviewers (J.W., P.P.C.C., S.L.) using the Cochrane Collaboration’s tool for assessing risk of bias²¹ (Table 2).^22–31

Table 2.

Characteristics of Trials of Cancer Risk Assessment Tools in Primary Care (N = 11)

Author, Year, Risk Tool, Setting	Disease(s)	Sample	Study Design	Intervention(s)	Overall Risk of Biasa
Schroy et al22 2011 Your Disease Risk United States	CRC	665 patients (223 combined intervention; 212 decision aid alone; 231 control) 50 clinicians (47 general internists; 3 nurse practitioners) 2 clinics	RCT (3 groups) Patients randomized before routine visit with primary care clinician	Control: usual care and generic lifestyle change advice for disease prevention Intervention 1: decision aid for CRC screening Intervention 2: decision aid for CRC screening plus CRC personalized risk assessment	Low/unclear
Schroy et al23 2012 Your Disease Risk United States	CRC	825 patients (280 combined intervention; 269 decision aid alone; 276 control) 61 clinicians (47 general internists; 11 family physicians; 3 nurse practitioners) 2 clinics	RCT (3 groups) Patients randomized before routine visit with primary care clinician	Control: usual care and generic lifestyle change advice for disease prevention Intervention 1: decision aid for CRC screening Intervention 2: decision aid for CRC screening plus CRC personalized risk assessment	Low/unclear
Rubinstein et al24 2011 Family Healthware Impact Trial (1) United States	CRC, BC, and OC,b heart disease, stroke, and diabetes	3,283 patients (2,077 intervention; 1,206 control) 41 clinics (23 intervention; 18 control)	Cluster RCT Cluster randomization at clinic level	Control: standard print messages about screening and lifestyle choices recommended for general health Intervention: patient self-completed risk assessment using the Family Healthware risk assessment tool; personalized risk prevention messages tailored to familial risk	Unclear
Ruffin et al25 2011 Family Healthware Impact Trial (2) United States	CRC, BC, and OC,b heart disease, stroke, and diabetes	3,344 patients (2,105 intervention; 1,239 control) 41 clinics (23 intervention; 18 control)	Cluster RCT Cluster randomization at clinic level	Control: standard print messages about screening and lifestyle choices recommended for general health Intervention: patient self-completed risk assessment using the Family Healthware risk assessment tool; personalized risk prevention messages tailored to familial risk	Unclear
Wang et al26 2012 Family Healthware Impact Trial (3) United States	CRC, BC, and OC,b heart disease, stroke, and diabetes	3,344 patients (2,105 intervention; 1,239 control) 41 clinics (23 intervention; 18 control)	Cluster RCT Cluster randomization at clinic level	Control: standard print messages about screening and lifestyle choices recommended for general health Intervention: patient self-completed risk assessment using the Family Healthware risk assessment tool; personalized risk prevention messages tailored to familial risk	Unclear
Emery et al27 2007 GRAIDS Trial England	CRC, BC, and OCb	240 patients received GRAIDS intervention; 84 referred to cancer genetics clinic from control practices 45 clinics (23 intervention; 22 control)	Cluster RCT Cluster randomization at clinic level	Control: 45-minute presentation to all GPs in practice on cancer genetics and copy of referral guidelines for cancer genetics clinic Intervention: 45-minute presentation on cancer genetics to all GPs in practice and copy of referral guidelines for cancer genetics clinic; 1–2 “lead clinicians” per practice trained to use web-based GRAIDS risk assessment tool for OC, CRC, and BC	Low
Campbell et al28 1997 Health risk survey Australia	Cervical cancer	679 female patients (354 intervention; 325 control) 2 clinics	RCT Randomization at patient level	Control: patient self-completed health risk survey Intervention: patient self-completed health risk survey and was given summary including eligibility for cervical screening and date of last Pap test	Low/unclear
Wilson et al29 2006 Risk assessment checklist Scotland	BC	346 clinicians (230 intervention; 116 control) 86 clinics (57 intervention; 29 control)	Cluster RCT (2:1) Randomization at clinic level	Control: standard Scottish guidelines to assess risk for referral to cancer genetics sent to GPs Intervention: multifaceted intervention including risk assessment checklist for CRC, BC, and OC; information about cancer genetics; patient information booklets; web links cancer/genetics; e-mail link to cancer genetics services; referral letter proforma; education sessions about cancer genetics	Low
Emmons et al30 2004 Harvard Colorectal Cancer Risk Assessment Tool United States	CRC	353 patients (134 absolute risk only; 146 absolute plus relative risk; 73 control) 2 clinics	RCT Randomization at patient level	All participants used the Harvard Colorectal Cancer Risk Assessment Tool Control: patients received passive risk communication without risk presentation Intervention: patient risk tool providing 4 different combinations of presentations of risk: (1) absolute and relative risk, (2) absolute risk only, (3) absolute and relative risk with the ability to manipulate the risk input to change the output, and (4) same as for (3) but absolute risk only	Low
Weinstein et al31 2004 Harvard Colorectal Cancer Risk Assessment Tool United States	CRC	353 patients (134 absolute risk only; 146 absolute plus relative risk; 73 control) 2 clinics	RCT Randomization at patient level	All participants used the Harvard Colorectal Cancer Risk Assessment Tool Control: patients received passive risk communication without risk presentation Intervention: patient risk tool providing 4 different combinations of presentations of risk: (1) absolute and relative risk, (2) absolute risk only, (3) absolute and relative risk with the ability to manipulate the risk input to change the output, and (4) same as for (3) but absolute risk only	Low
Holloway et al22 2003 Risk assessment scale Wales	Cervical cancer	1,890 female patients (772 intervention; 1,118 control) 29 clinics (15 intervention; 14 control)	RCT Randomization at clinic level	Control: no risk assessment Intervention: practice nurse risk communication package including a paper-based risk assessment scale based on level of education, current smoking status, number of years of oral contraceptive use, and number of sexual partners ever33	Low

BC = breast cancer; CRC = colorectal cancer; GP = general practitioner; GRAIDS = Genetic Risk Assessment on the Internet with Decision Support; OC = ovarian cancer; Pap = Papanicolaou; RCT = randomized controlled trial.

^aBias assessed using the Cochrane Collaboration risk of bias based on: (1) sequence generation; (2) allocation concealment; (3) blinding of participants, personnel, and outcome assessors; (4) assessment of incomplete outcome data; (5) selective outcome reporting; (6) “other” sources of bias not listed. Low risk of bias = low risk of bias across all domains. Unclear risk of bias = unclear risk of bias for 1 or more key domains. High risk of bias = high risk of bias for 1 or more domains.

^bThese trials assessed patients’ risk for BRCA mutation rather than specifically discussing ovarian cancer screening.

The heterogeneity of interventions and outcomes precluded any meta-analysis of the data. The review provides a narrative synthesis of the data.

RESULTS

Study Selection

Our database searches identified 989 studies. After title and abstract review, and removal of 37 duplicates, 210 full-text articles were assessed for eligibility. Eleven articles reporting trials of 7 risk tools were included (Table 1 and Figure 1).

Study Characteristics

The review included trials of risk assessment tools that were either completed by clinicians with patients or self-completed by patients. Risk tools included web-based risk tools, paper-based risk checklists, and multifaceted interventions involving patient resources, clinician education, or both. The studies included a wide range of outcomes and cancers (Table 2).

The trials varied in their design, including the unit of randomization (clinic or patient) and the population testing the intervention (clinician or patient). Of the 11 studies, 3 randomized by clinic and trialed a clinician-targeted intervention,^27,29,32 3 randomized by clinic and trialed a patient-targeted intervention,^24–26 and 5 randomized by patient and trialed a patient-targeted intervention.^{22,23,28,30,31} One study randomized patients by clinic days to reduce potential contamination.²⁸ We examined the unit of randomization as a possible source of heterogeneity of the results and found no clear trends (Table 3).

Table 3.

Results of Trials of Cancer Risk Assessment Tools in Primary Care

Outcomes Evaluated	Author, Year	Randomization Unit	Results
Patients
Risk perception	Wang et al26 2012	Clinic	In patients who underestimated their CRC risk, the intervention increased accuracy of risk perception (intervention 17% vs control 10%, P = .05). There was no increase in accuracy of risk perception between groups in women who underestimated their risk for BC (intervention 18% vs control 14%, P = .4) or OC (intervention 8% vs control 13%, P = .4).
	Emery et al27 2007	Clinic	There was no difference in mean risk perception between patients referred from intervention vs control practices. Nonsignificant trend seen toward more accurate risk perception at the point of referral in intervention patients, with fewer overestimating their risk of cancer (OR = 1.50; 95% CI, 0.62–3.67; P = .36).
	Holloway et al32 2003	Clinic	There was no change in risk perception of cervical cancer between groups (OR = 1.07; 95% CI, 0.85–1.35).
	Emmons et al30 Weinstein et al31 2004	Patient	Accuracy of risk perception increased if risk was presented as combined relative and absolute risks or as absolute risk only vs control (for both people who overestimated and who underestimated their isk preintervention).
Screening intentiona	Holloway et al32 2003	Clinic	Women at intervention clinics were more likely to intend to reduce their screening interval for cervical screening in line with national guidelines (intervention 44% vs control 61%; OR = 0.51; 95% CI, 0.41–0.64; P <.001).
	Schroy et al22 2011	Patient	Mean intention scores to schedule a CRC screening test were higher for both intervention groups vs the control group: intervention group 1: DA (mean = 4.4; SD = 1.0); intervention group 2: DA+YDR (mean = 4.3; SD = 1.0); control group (mean = 3.9; SD = 1.4) (P <.001). Mean intention scores to complete a CRC screening test were higher for both intervention groups vs the control: intervention group 1: DA (mean = 4.3; SD = 1.0); intervention group 2: DA+YDR (mean = 4.3; SD = 1.0); control group (mean = 3.9; SD = 1.3) (P <.001).
	Schroy et al23 2012	Patient	Booking a screening test:  DA group was more likely to book a CRC screening test than control group at 1 month (69.1% vs 60.5%, P <.035); 3 months (71.8% vs 62.3%, P = .019); 6 months (77.0% vs 65.2%, P = .002); and 12 months (80.7% vs 71.4%, P = .011).  DA group was more likely than DA+YDR group to book a CRC screening test at 1 month (69.1% vs 60.4%, P <.031); 6 months (77.0% vs 67.1%, P <.010); and 12 months (80.7% vs 73.6%, P = .048).
Screening adherenceb	Rubinstein et al24 2011	Clinic	CRC screening increased in both groups over time: intervention, from 76% to 84%, and control, from 77% to 84% (P = .95). BC screening increased in both groups over time: intervention, from 73% to 82%, and control, from 78% to 85% (P = .82). No difference between intervention and control groups in screening adherence for CRC, BC, or OC (P >.09) after 6 months.
	Holloway et al32 2003	Clinic	No difference in actual cervical screening intervals and consistency with guidelines between groups at 5 years: intervention 5%, control 7% (OR = 0.61; 95% CI, 0.36–1.03; P = .063).
	Schroy et al23 2012	Patient	Completing a CRC screening test:  DA group was more likely than control group to complete test (43.1% vs 4.8%, P = .046)
	Campbell et al28 1997	Patient	No difference in cervical screening in women identified as being “underscreened” (P >.05).
Behavior change	Ruffin et al25 2011	Clinic	Intervention group was more likely than control group to increase daily fruit and vegetable intake from ≤5 servings to ≥5 servings (OR = 1.29; 95% CI, 1.05–1.58) and to increase physical activity to 5–6 times/week for ≥30 minutes per day (OR = 1.47; 95% CI, 1.08–1.98).
Anxiety/worry	Emery et al27 2007	Clinic	Cancer worry was lower in patients referred from intervention practices vs from control practices: mean difference = −1.44 (95% CI, −2.64 to 0.23; P = .02).
	Holloway et al32 2003	Clinic	Women at intervention practices were less likely to be “fearful of cervical cancer” (OR = 0.66; 95% CI, 0.47–0.93; P = .019), “concerned about chances of serious problems with a smear in the future” (OR = 0.70; 95% CI, 0.51–0.95; P = .026), and “anxious about a recent smear test” (OR = 0.81; 95% CI, 0.66–0.98; P = .036). No differences seen between women at intervention vs control practices in “concern about their smear result” (OR = 0.75; 95% CI, 0.45–1.24; P = .25).
	Emmons et al30 Weinstein et al31 2004	Patient	33% of all participants in the study had less cancer worry and 17% had more cancer worry after using the Harvard CRC Risk Tool (comparative data between groups not reported).
Knowledge	Emery et al27 2007	Clinic	There was a nonsignificant increase in cancer knowledge in patients referred from intervention practices vs from control practices: BC knowledge mean difference = 0.11 (95% CI, −1.05 to 1.27) and CRC knowledge mean difference = 0.64 (95% CI, −1.01 to 2.29).
	Wilson et al29 2006	Clinic	No difference seen in patient knowledge between groups for items “Stress is a major cause of BC” (23% vs 23%, P = .98); “Having one close relative with BC always increases your risk considerably” (88% vs 91%, P = .71); and “Minor injury to the breast can cause BC” (20% vs 23%, P = .78).
	Holloway et al32 2003	Clinic	85% of women at control practices incorrectly agreed that “cervical cancer is among the top 4 female cancers in the UK” compared with 22% of women at intervention practices (OR = 0.05; 95% CI, 0.02–0.11; P <.0001).
	Schroy et al22 2011	Patient	DA groups and DA+YDR group both had increased knowledge scores vs control: intervention group 1 (DA): mean = 3.2; SD = 2.6; intervention group 2 (DA+YDR): mean = 3.0; SD = 2.5; control: mean = 0.8; SD = 2.2 (P <.001). No differences seen in knowledge scores between DA and DA+YDR groups.
Satisfaction	Schroy et al22 2011		Patient satisfaction was higher for DA or DA+YDR vs control: intervention group 1 (DA): mean = 50.7; SD = 6.2; intervention group 2 (DA+YDR): mean = 50.5; SD = 6.2; control group: mean = 46.7; SD = 7.9 (P <.001). Satisfaction did not differ between DA and DA+YDR groups.
Clinicians
Appropriate screening and/or referral	Emery et al27 2007	Clinic	Increase seen in referral rate to cancer genetics clinic in intervention practices; mean difference = 3.0 referrals per 10,000 patients per practice per year (95% CI, 1.2–4.8; P = .002). Referrals from intervention practices were more likely to be consistent with referral guidelines and therefore “appropriate” vs control practice referrals (OR = 5.2; 95% CI, 1.7–15.8; P = .006).
	Wilson et al29 2006	Clinic	No difference seen between groups in appropriateness of referrals: intervention 58%, control 48% (RR = 1.18; 95% CI, 0.88–1.37).
Clinician confidence	Emery et al27 2007	Clinic	Clinicians’ confidence in managing people with a family history of cancer increased in intervention practices vs control practices (P <.0001).
	Wilson et al29 2006	Clinic	No change seen in clinician confidence between groups for the following about BC risk: “taking appropriate family history” (60% vs 61%, P = .93); “knowing which patients need to be referred” (40% vs 33%, P = .27); “reassuring low-risk patients” (57% vs 52%, P = .46); and “being able to answer questions” (23% vs 22%, P = .77).

AM = adjusted mean; BC = breast cancer; CRC = colorectal cancer; DA = decision aid; OC = ovarian cancer; OR = odds ratio; RR = risk ratio; YDR = Your Disease Risk.

^aParticipant has the intention to schedule or order a screening test.

^bParticipant has completed a screening test.

Recruitment proportions of eligible study participants varied from very low (14% to 25%)^{24–26,30,31} to very high (93% to 95%).^28,32 Contacting eligible patients by mail and following them up with a telephone call yielded a low recruitment. ^{24–26,30,31} More successful recruiting (93% to 95% of eligible participants) was achieved with a dedicated research assistant or practice nurse recruiting eligible patients in the primary care waiting room before their appointment.^28,32 Similarly, when the intervention was delivered by a practice nurse, 75% of patients completed a risk assessment,²⁸ but when clinicians were required to complete training, engagement was low (12% of intervention general practitioners attended).²⁹

Outcomes

Study outcomes are shown in Table 3 and discussed in detail below.

Accuracy of Patient Risk Perception

Overall, there was limited evidence that risk assessment tools altered patients’ risk perception, except in specific subgroups. For example, in the Family Healthware Impact trial,²⁶ there was a significant increase in accuracy of risk perception in those patients who underestimated their risk of colorectal cancer at baseline, but not in women who underestimated their risk of breast cancer. The trial of the Harvard Colorectal Cancer Risk Assessment Tool specifically tested different risk presentation formats. In people who either underestimated or overestimated their risk at baseline, accuracy of risk perception was improved by either absolute risk alone or absolute risk plus relative risk formats, compared with the control patients.^26,30,31 The Genetic Risk Assessment on the Internet with Decision Support (GRAIDS) trial and a trial assessing cervical cancer risk found no significant differences in risk accuracy for colorectal cancer,²⁷ breast cancer,²⁷ or cervical cancer.³²

Patient Behaviors

Four trials explored screening behavior outcomes, including screening intentions, patient booking/planning a screening test, and patient completing a screening test.

Schroy et al²² tested a pair of interventions. Intervention 1 was a shared decision-making tool, and intervention 2 was a combined shared decision-making tool plus a risk assessment tool (Your Disease Risk), comparing them both with usual care. Immediately postintervention, intentions to order a screening test and intentions to complete a screening test were higher in both intervention groups relative to the control group (P <.001) with no difference between the 2 interventions. During the 12-month follow-up, participants using the decision aid alone (intervention 1) were more likely at every time point to book a test than the control group, and similarly, the decision aid increased the likelihood of completing a screening test when used alone, but not when combined with risk assessment.²³

In the Family Healthware Impact trial, there was an increase in colorectal cancer and breast cancer screening in both groups, but no difference in screening rates between the intervention and control groups at 6 months. Of further note, the relatively high rates of cancer screening at baseline in both groups suggested a ceiling effect.²⁴

Holloway et al³² trialed the effect of a risk tool on reducing time intervals between cervical screening, which, at the time of the trial in the United Kingdom, was recommended every 5 years. In the short term, women in the intervention group intended to have screening less frequently, but at 5 years of follow-up, there was no significant difference.

In contrast, Campbell et al²⁸ tested a risk tool with women in primary care in Australia to identify underscreened women and encourage them to have risk-based cervical screening. At 6 months of follow-up, women in the intervention group were no more likely to have had a cervical screening test than those in the control group.

The Family Healthware Impact trial also assessed impact on lifestyle behaviors.²⁵ The risk tool provided age-specific and sex-specific health messages to participants based on their family history of heart disease, stroke, diabetes, colorectal cancer, breast cancer, and ovarian cancer. After 6 months, participants in the intervention group were significantly more likely to have increased their daily fruit and vegetable intake, and their physical activity.

Patient Cancer Worry

None of the 3 trials that measured cancer-related anxiety found any increase after risk assessment. The GRAIDS trial recruited patients who had discussed concerns about their familial cancer risk with their general practitioner.²⁷ Patients referred to cancer genetics services from practices that used the GRAIDS tool had a lower cancer worry than patients referred from the control practices. In the cervical screening trial of Holloway et al,³² women receiving the intervention were less likely to be “fearful” of cervical cancer, less “concerned about chances of serious problems with a smear in the future,” and less “anxious about a recent smear test.”

In the trial of the Harvard Cancer Risk Assessment and Communication Tool, 33% of participants reported feeling less worried about getting colorectal cancer, but 17% reported increased worry about the disease after using the tool.^30,31 These associations were seen regardless of whether the risk was presented as absolute risk, relative risk, or combined risk. There were no comparable control data in this trial for cancer worry.

Patient Knowledge

Patient knowledge was measured by understanding of population cancer risk, causes of cancer, and screening guidelines. Schroy et al²² found that both intervention groups had improvements in their knowledge of colorectal cancer screening guidelines, rationale, and goals. Women in the cervical screening trial had a greater understanding of screening guidelines and, in particular, screening intervals recommended for cervical screening as a result of the intervention.³² Wilson et al²⁹ found no differences in patient knowledge between groups despite patient and clinician education.

Patient Satisfaction

Only 1 study measured patient satisfaction. In this study, the use of a decision aid with or without a risk tool improved patient satisfaction with making screening decisions compared with the control condition.²²

Appropriate Clinician Referrals, Screening, or Both

Two trials from the United Kingdom looked at the effect of risk tools on “appropriateness of referrals” to cancer genetics services by comparing them against local referral guidelines.^27,29 In the GRAIDS trial, risk assessment increased the proportion of appropriate referrals when compared with local guidelines that were implemented with the GRAIDS tool.²⁹ Although there was an increase in appropriate referrals based on the local guidelines, the actual proportion of patients found to be at high risk was no different after more detailed assessment at the genetics clinic. This finding suggests a lack of specificity of the referral guideline that is likely to be implemented more systematically using a risk tool.

Clinician Confidence

In the GRAIDS trial, clinicians’ confidence in assessing patients’ family history of cancer was increased.²⁷ In contrast, in the Scottish trial, no differences were observed in clinician confidence about family history risk assessment and referral.²⁹

DISCUSSION

This systematic review identified only 11 articles reporting trials of 7 cancer risk assessment tools in primary care. Overall, this sample represents a relatively small evidence base, especially in the context of the growing number of cancer risk tools available online. The findings suggest potentially beneficial effects of cancer risk assessment tools in terms of improving accuracy of patient risk perception and knowledge, intentions to have cancer screening, and changes in diet and physical activity, without causing an increase in cancer-specific anxiety. Effects on actual cancer-screening behaviors are less clear. Cancer risk assessment tools may also improve clinician confidence and appropriateness of referrals to cancer genetics services, although the evidence for this benefit is somewhat contradictory from only 2 trials. Risk tools were more successful when they were initiated by patient who were concerned about their family history (of cancer),²⁷ were used by a dedicated clinician,^27,32 included health promotion messages,²⁵ and included decision support within the tool.²³ Interventions were less successful when tested in trials that involved a passive system for using the risk assessment tool.²⁹

There are some important caveats. The trials included in this review were heterogeneous in terms of the precise nature of the intervention, the unit of randomization, how they were implemented, and the health care setting in which they were studied. Furthermore, some of the populations in which the tools were used were selected toward a group who had existing concerns about their risk, especially about their family history. For example, the relatively low recruitment rates in the US Family Healthware Impact trial probably were associated with response bias toward a well-educated sample with relatively high baseline rates of cancer screening. Additional methodologic weaknesses in some studies included small sample sizes and therefore potentially underpowered trials,^28,30,31 poor recruitment rates^{22–24,26,29–31} lack of clinician engagement in the intervention,²⁹ and patient-reported outcomes that may be influenced by social desirability bias.²⁸ The unit of randomization was not a clear source of heterogeneity despite greater risk of contamination in the patient-randomized trials.

Previous systematic reviews have examined the effect of patient-oriented decision aids in screening³³ and also communication of risks in screening programs.³⁴ Our review differs in terms of the nature of the interventions and the populations studied, although the findings are consistent: communication of risks is associated with increased intention to screen, and patient-oriented decision aids can increase knowledge. Two of the included studies examined different methods of communicating risk. The way risks were presented across all trials varied, and none complied with current perceived best practice in presenting risk information as recommended by the International Patient Decisions Aid Standards.^35,36

If we are to move toward risk-stratified cancer screening, primary care clinicians will require simple tools to implement validated risk models, which are likely to incorporate genomic as well as lifestyle factors. As the GRAIDS trial demonstrated, risk tools are only as effective as the underlying risk model. Ideally, tools will be able to present absolute risks and the predicted effects of behavior change or chemoprevention on an individual’s risk of cancer. Importantly, they need to be designed to present evidence in ways that highlight the risks of overscreening people at average or low risk as well as the benefit of screening in populations who are most likely to benefit.³⁷ Most of the trials to date have focused on a single cancer or those for which predictive genetic testing was relevant. Validated risk prediction models, however, exist for many common cancers that could ideally be incorporated into a single tool.

In conclusion, despite the existence of many cancer risk assessment tools, there is relatively limited evidence from RCTs of their effectiveness, especially in terms of their impact on risk-appropriate cancer screening behaviors. Risk tools may increase actual intentions to have cancer screening, but additional interventions at the clinician or system level may be needed to increase screening behavior. The results support the use of dedicated staff to maximize implementation of the intervention. The incorporation of health economic evaluation to determine the most cost-effective approaches to delivering risk-stratified cancer screening in primary care, and the potential added cost-benefit of genomic profiling within these trials, will be important outcomes to measure in future trials.