Introduction

Over the last decade, increasing attention has been paid to the idea of personalised medicine. Personalised medicine implies that a specific treatment (or prevention option) is not universally applied but is directed to individuals based on their personal profile (Collins and Varmus 2015). The profile may incorporate markers of risk or predictors of treatment response. Markers of risk may be genetic or non-genetic (such as mammographic density). Treatment response may incorporate genetic variation in proteins that metabolise relevant drugs or, in the case of cancer, by presentation of an aberrant protein target caused by a specific mutation(s). These are discussed in greater detail below.

The immediate goal of personalised medicine is to optimize care to an individual based on a personal assessment. It is hoped that adopting this strategy widely will lead to more patients to be cured or to more long-term survivors or to reduced morbidity and improved quality of life. Some patients might be spared unnecessary treatment if those who do not stand to benefit can be identified. In this scenario, there may be savings both in terms of morbidity (due to side effects) and cost.

There are many reports in the literature where individual patients benefit from full sequencing of tumour DNA, but these are usually anecdotal accounts (Stockley et al. 2016; Tsimberidou et al. 2017). To date, there is little evidence that the paradigm of personalised medicine, when applied on a wide scale, will reduce morbidity and mortality in the population, compared to a conventional approach. There is no doubt that additional assessments will increase health care costs, but there is little evidence that there are savings to be had by reducing treatments or by saving lives. It is important that we consider whether or not the introduction of personalised medicine on a wide scale can expect to improve the health statistics of the population at large. The field is vast and in the following review/commentary I focus on the prevention of breast and ovarian cancer (the area that I know best) and I will emphasize genetic markers—keeping with the theme of the journal.

In general, for personalised medicine to be effective at a population level, several conditions must be met:

  1. 1.

    The assessment should place individuals into two or more categories of risk or potential benefit that are clearly delineated and the risks should be substantially different according to category.

  2. 2.

    The test should be widely available and acceptable to those who might benefit. In some cases all individuals in the population are the target of the test and if so, the benefit of the test will be proportionate to the uptake of the test in the population. In other cases, it is only patients with a pre-existing condition that are the target population.

  3. 3.

    The intervention itself should be effective in those who test positive.

  4. 4.

    The information regarding the test and the intervention should be easily communicated by the health care provider and understood by the client.

  5. 5.

    The intervention should be acceptable to a high proportion of the clientele who test positive and who might potentially benefit.

  6. 6.

    The uptake rate of the intervention should be higher in individuals who test positive than those who test negative. For continuous measurements, the uptake of the test should correlate with the level of risk.

These conditions are similar to those put forward to population-based screening and genetic testing (Dobrow et al. 2018; Burke and Zimmern 2007).

To summarise, it may be the case that a test and or an intervention has a positive impact on an individual patient, in terms of prolonging life, delaying progression or in some cases, cure. The question raised in this essay, is under what circumstances can a test/treatment combination that has been shown to be helpful in an individual, impact on cancer incidence or mortality in the population at large. In the case of breast cancer, for example, there are 26,000 new cases annually in Canada and 5000 deaths (Canadian Cancer Society). If a new test were to reduce these numbers by as little as 1% this would be valued, but that would require a reduction of 260 cases or 50 deaths in a year.

Cancer genes in personalised medicine

A susceptibility gene for cancer is one whereby a mutation increases the risk of one or a number of cancer types, beyond the risk in the general population. The relationship between genes and phenotypes can be described in two ways; first, what is the range of cancers associated with a mutation in a particular gene, such as BRCA1 or ATM? Excess cancer risk can be expressed as a hazard ratio (carrier versus non-carrier) or the lifetime risk of cancer in carriers (penetrance). Most cancer genes are associated with susceptibility to one or more cancers. This data is particularly relevant when counselling a patient once a specific mutation is found.

Another way to define the relationship is to identify the genes which are associated with a given cancer (such as breast, ovary or colon). In this scenario, the data is most useful to the counsellor or geneticist who must decide which genes to test for based on the personal and family history of the patient. Currently, the de facto operational decision is to decide which commercial sequencing panel to request—a panel may contain a few genes up to 20 or more genes (Lerner-Ellis et al. 2015; Szender et al. 2018; Daly et al. 2017).

Multigene panels have by and large replaced testing for one or a few genes (Kurian et al. 2018). One of the difficulties in adopting the multigene panel approach is to gauge the precision of the cancer estimates associated with a specific mutation in a particular gene. For many genes, the actual risks associated with a mutation may be poorly specified. In some cases, it is difficult to determine the level of pathogenicity of a specific variant—in particular of a missense variant. For most genes on the panels there are none or few rigorous epidemiologic studies to allow a precise risk assessment in the clinic and the risk communication is highly probabilistic.

The panels comprise high- or moderate-penetrance genes, each associated with a relative risk of 2.0 or more (there is no precise definition). A second class of variants are single nucleotide polymorphisms (SNPs). These are common variants which have been identified to be associated with a small variation in cancer risk, typically with hazard ratios of 1.1–1.4 (Pharoah et al. 2002). The individual SNPs have been identified through genome-wide association study (GWAS) and imputation. The first GWAS for breast cancer was published in 2007 and studied 227,876 SNPs on 4398 cases and 4316 controls (Easton et al. 2007). The most recent GWAS for breast cancer was published in 2018 and studied 11.8 million SNPs on 122,977 cases and 105,974 controls of European ancestry (Michailidou et al. 2017). Individually, these SNPs may contain little useful information (and no one proposes to test patients for a single SNP) but in combination, they can be used to generate a risk profile (personal risk profile). These SNP panels may identify some individuals who carry a two-fold elevated risk or more. It is proposed that the SNP-based genetic models are now sufficiently mature that they may be incorporated into clinical practice (Pharoah et al. 2002; Mavaddat et al. 2015). In 2015, Mavaddat et al. (2015) used a risk assessment model based on 77 SNPs to generate personal risk profiles and thereby to stratify women according to their lifetime risk of breast cancer. For the 77 SNPs, the frequency of the rarer of the two alleles ranged from 0.001 to 48.2% and the corresponding odds ratios range from 0.86 to 1.36. The baseline risk of breast cancer to age 74 in the UK is 8.2% (Cancer Incidence in Five Continents 2002). Based on her personal risk score, a woman in the top one percentile had a three-fold increase in risk relative to the mean (HR = 3.36, 95% CI 2.95–3.83) or roughly 25% lifetime (Mavaddat et al. 2015). Individuals in the top five percentile group had a two-fold increase in risk relative to the mean (HR 90–95% group = 1.85, 95% CI 1.72–1.99) or roughly 15% lifetime. It is important to recognize that risk can be communicated in various ways to patients and the choice of decision might be influenced by the choice of metric used to communicate risk (Kraft et al. 2009).

Prevention

The basic options for primary prevention of breast and ovarian cancer incidence and mortality are surgical prevention and chemoprevention. Screening (discussed below) is directed at secondary prevention, i.e., not to prevent cancer but to catch it at an early and treatable stage.

Preventive surgery, breast cancer

As described above, the assumption underlying personalised medicine is that we can categorize women according to a formal risk assessment and communicate the risk effectively. Further, we anticipate that the information will be used in a rational manner, namely that the likelihood of the intervention (in this case preventive surgery) increases with the likelihood of the cancer. As a corollary, we suppose that there is a baseline level of risk below which the intervention is no longer justified. Recently, Kurian et al. (2018) evaluated a multigene panel in women with breast cancer as it pertains to the risk of contralateral breast cancer and the uptake of contralateral mastectomy. It is recommended that contralateral mastectomy should only be offered to women at high risk due to genetic predisposition (Daly et al. 2017). The authors present the proportions of women who had a preventive mastectomy of the contralateral breast among breast cancer patients in the Surveillance, Epidemiology, and End Results (SEER) registry (aged 20–79 years, diagnosed with first primary stage 0 to II breast cancer between 2013 and 2015) who had undergone genetic testing, according to risk profile. Of those with a bona fide BRCA1/2 mutation, 79% had a bilateral mastectomy. Of those with a pathogenic variant in another gene, 38% had a bilateral mastectomy; of those with a variant of unknown clinical significance, 30% had a bilateral mastectomy and of those with no mutation 35% had a bilateral mastectomy. The Kurian et al. paper tests the paradigm of panel testing for breast cancer patients and clearly fails the test, at least for genes other than BRCA1 or BRCA2. Among the women who did not have a BRCA1 or BRCA2 mutation, there appeared to be no correlation between the objective inherent cancer risk and the action taken by the patient (the 10-year risk of contralateral breast cancer for a women with breast cancer with no BRCA1/2 mutation is about 5–6%; Reiner et al. 2013). Preventive contralateral mastectomy was equally common among women with normal genetic findings as it was in women with a pathogenic variant in a gene other than BRCA1 or BRCA2. The rates of bilateral mastectomy were high in all categories and there was no lower threshold. Except in rare cases, the operation was not recommended by the surgeon. The fact that decisions regarding choice of surgery were not made with respect to the result of the genetic risk assessment or as a physician recommendation, but as a consequence of the emotional state of the patient raises doubts about the validity of the assumptions of personal medicine. The paper suggests that the research community has overestimated the potential for the health care professional to influence the patient decision using risk assessment tools.

It is not surprising that 35% of the women with no mutation had a bilateral mastectomy. Over the past decade, the incidence of women undergoing contralateral mastectomy for unilateral breast cancer has grown rapidly (Kurian et al. 2014). Most women in the USA who have the operation do not carry an identified mutation; in California, the rate of bilateral mastectomy has increased from 2.0% in 1998 to 12.3% in 2011 (an annual increase of 14%). Despite the fact that bilateral mastectomy has not been shown to reduce mortality (Kurian et al. 2014), women often state that their wish for better survival motivated their decision.

The Kurian et al. (2018) paper describes patient decisions in women with unilateral breast cancer and cannot be directly extrapolated to women with a mutation and no cancer. It is important to document patient responses to the information they receive when a new test is introduced. In our earlier work on primary preventive mastectomy, we found that the majority of preventive mastectomies were done in Ontario among women at low risk (Metcalfe and Narod 2002). It has been anticipated that the ability to define risk using molecular markers has been refined since this publication, but in the last 15 years we have seen a rise in bilateral mastectomy rates, particularly among non-carriers. We computed the average lifetime risk of breast cancer for women undergoing preventive bilateral mastectomy to be 17%, but the same women estimated their own risk to be 76%.

Preventive bilateral mastectomy is only recommended for carriers of high-risk genes, in particular BRCA1 and BRCA2 (Daly et al. 2017). Except in extreme cases, the personal risks estimated by the polygenic model will be too low to justify preventive mastectomy. Nevertheless, a positive result (a high-risk score) may cause women to worry and some women will undergo preventive surgery to alleviate their fear. It is assumed by advocates of personalised risk assessment (SNP-based personal risk profiles) that the primary impact would be through enhanced screening, but patients may choose to pursue preventive surgery instead. It is potentially problematic if a preventive surgery performed for a women at low or moderate risk were to be considered as a benefit of personalised medicine, especially if it were not recommended by the professional but rather was motivated by enhanced cancer fear engendered by the test.

Preventive surgery, ovarian cancer

Perhaps the best example of the potential of personalised medicine to reduce cancer mortality rates comes from genetic testing for ovarian cancer. Approximately 13–15% of women with ovarian cancer carry a BRCA1 or BRCA2 mutation and these cancers are potentially preventable (Zhang et al. 2011; Norquist et al. 2016). We have shown that oophorectomy, ideally done before the woman enters the period of risk, can reduce the risk of subsequent ovarian cancer by 80% or more (there is a 2% residual risk of primary peritoneal cancer) (Finch et al. 2014b). However, for a woman to know her genetic status and qualify for the surgery, it is necessary that she have the test prior to the development of ovarian cancer. The risk of ovarian cancer in the BRCA1 carrier begins around age 35 and in the BRCA2 carrier begins around age 45 (Kotsopoulos et al. 2018).

Most women do not qualify for testing through public health programs or third-party payers prior to the development of cancer (ironically they all qualify after they have ovarian cancer) (Finch et al. 2014a). We have shown that in Ontario, where genetic testing is available to all invasive ovarian cancer patients, only 3.6% of women were eligible for testing prior to developing cancer and therefore only this small fraction that might be prevented (Finch et al. 2014a). To achieve this however, it is required that the woman and caregiver are aware of the testing criteria, that a formal family history assessment is conducted and that the patient is referred for testing appropriately. Further, the average age of testing in Canada is approximately 46 years old and this is beyond the optimum age of oophorectomy (35 for BRCA1 carriers). We have recently shown that only about one in five women with ovarian cancer in Ontario undergoes genetic testing, even though all are eligible and 80% are willing to be tested (Metcalfe et al. 2009). We estimate now in Ontario that only 1% of ovarian cancers are currently prevented through the publicly funded provincial genetic testing program (Finch et al. 2014a). A possible solution is universal genetic testing (discussed below).

The most comprehensive genetic testing program is underway in the region of Szczecin, Poland (Menkiszak et al. 2017). We evaluated the impact of twelve years of a regional population-based genetic testing program for BRCA1 founder mutations and consequent prophylactic oophorectomies on the incidence of ovarian cancer in Pomerania (Menkiszak et al. 2017). From 1999 to 2010, 43,827 women aged 35 and above were tested for three BRCA1 founder mutations. This represents approximately 8% of the women in the region. 641 women were found to carry a mutation (1.5%) and of these, 220 had a prophylactic oophorectomy. Based on the age-specific rates of ovarian cancer for women with BRCA1 mutations from Poland, we estimate that of the 220 women, 63 would have developed ovarian cancer by age 70 had they left their ovaries intact. During this same period, 1611 ovarian cancers were diagnosed in the region; therefore we estimate that approximately 2.3% of ovarian cancers were prevented between 1999 and 2015. Therefore, by testing 8% of the eligible women we estimate we prevented 2.3% of the ovarian cancers. This, in itself, is a cost-efficient strategy, but one hopes that we can do better by expanding the reach of genetic testing.

Chemoprevention

Chemoprevention, breast cancer

The concept behind chemoprevention is that the incidence and/or mortality from a cancer can be reduced through administration of a drug. These may include hormonal-based prescription drugs such as tamoxifen, aromatase inhibitors and oral contraceptives or over-the-counter preparations such as aspirin and vitamin D. Tamoxifen was approved for chemoprevention for moderate- and high-risk women based on the results of the NSABP study, published in 1998 (Fisher et al. 1998). Based on the study results, tamoxifen would be indicated for all women with a 5-year breast cancer risk in excess of 1.66%, this is a large category and includes all women over age 65. In the 20 years since the paper was published, tamoxifen has been used very little for chemoprevention (Reimers et al. 2015). There are several possible reasons for this. Tamoxifen has been shown to reduce the incidence of breast cancer but not mortality (Narod 2015). For example, in the update of the IBIS study there is a significant decline in the incidence of ER-positive cancer, 16 years from randomisation, but there is no decline in breast cancer mortality (Cuzick et al. 2015a). Despite the fact that there were 99 fewer breast cancers in the tamoxifen arm of the study (251 versus 350) there were five more deaths from breast cancer (31 versus 26). Results of other chemoprevention trials are similar (Narod 2015). Based on theoretical considerations we have discussed in detail elsewhere (Narod and Sopik 2018; Sopik and Narod 2018), it should not be assumed that a decline in the incidence of breast cancer is a valid surrogate for a decline in mortality.

The group most likely to benefit from tamoxifen were those women with atypical hyperplasia (HR = 0.44, 95% CI 0.17–1.15) (Cuzick et al. 2015b). Unfortunately it is not possible to screen the population for atypical hyperplasia as this is an incidental finding among women undergoing a breast biopsy.

It is commonly stated by women that they wish to avoid the side effects associated with tamoxifen, including menopausal symptoms, vaginal discharge and a small risk of endometrial cancer (Metcalfe et al. 2005).

There are others considerations; when tamoxifen is initiated, no biomarkers (such as bone density or serum cholesterol) can be monitored to provide positive feedback to the patients that she is benefiting. Further, when women start tamoxifen there is no immediate sense of well-being such as that experienced by women who undergo preventive surgery (Metcalfe et al. 2005). It is not clear if these obstacles are specific for tamoxifen or are relevant for chemoprevention in general.

It is now accepted that, due to low uptake rates, tamoxifen chemoprevention has had little impact on reducing the risk of breast cancer in the general population. Tamoxifen has not been particularly well-received among women at moderate risk—perhaps its benefit will be more readily demonstrable among women at high risk (who derive greater potential benefit than do low-risk women, but suffer side effects in similar numbers). Can the use of tamoxifen chemoprevention have greater impact if used for women with a BRCA1 or BRCA2 mutation, as befits a personal medicine paradigm? There are no prospective studies which evaluate the effectiveness of tamoxifen for cancer prevention in these high-risk subgroups. A large study of contralateral breast cancer in BRCA1 or BRCA2 carriers demonstrated a strong benefit; the multivariate odds ratio for contralateral breast cancer associated with tamoxifen use was 0.50 for carriers of BRCA1 mutations (95% CI 0.30–0.85) and was 0.42 for carriers of BRCA2 mutations (95% CI 0.17–1.02) (Gronwald et al. 2006b). Similar results were found in a pooled analysis of data from three observational cohorts of BRCA1/2 carriers (Phillips et al. 2013); the hazard ratio for contralateral breast cancer associated with tamoxifen use for BRCA1 carriers was 0.38 (95% CI 0.27–0.55) and for BRCA2 carriers was 0.33 (95% CI 0.22–1.05).

For the sake of the argument—assuming it is effective—we can consider the case of tamoxifen chemoprevention for carriers of BRCA1 mutations. It is possible to estimate the number of breast cancers which are prevented annually by prescribing tamoxifen to BRCA1 carriers in Ontario. First, we assume that 4% of breast cancers in Ontario are attributable to BRCA1 and BRCA2 mutations, and, of these, one-half (2%) are due to BRCA1. Assume 10% of unaffected women in Ontario with a mutation are aware of their carrier status. Of women who know that they are a carrier and who are unaffected with breast cancer, we estimate that 8% take tamoxifen (Metcalfe et al. 2008). Based on studies of contralateral breast cancer we assume tamoxifen will cut their breast cancer risk in half (Gronwald et al. 2006b; Phillips et al. 2013). To sum up, we estimate that one in 12,500 cases of breast cancer (0.02 × 0.10 × 0.08 × 0.50) are prevented annually in Ontario through the administration of tamoxifen to BRCA1 carriers. There are approximately 8000 breast cancers diagnosed in Ontario every year and this deficit corresponds to fewer than one breast cancer case per year.

In this analysis, tamoxifen chemoprevention fails to meet many of the criteria for effective personalised medicine described in the Introduction. Genetic testing is not widespread, the drug effectiveness has been questioned and the intervention is not acceptable to most women in the target population. If we are to do better, we must identify a better drug, but we must also ensure that genetic testing has a greater reach. Further, we should not assume that there will be a high uptake of any chemoprevention drug without empiric evidence. In a recent survey, many more women indicated that they were interested in taking a chemoprevention drug than who actually took chemoprevention (Liede et al. 2017).

Chemoprevention, ovarian cancer

Oral contraceptives have been shown to be effective for preventing ovarian cancer in carriers and non-carriers—reported risk reductions range from 30% (for non-carriers; Collaborative Group on Epidemiological Studies of Ovarian Cancer et al. 2008) to 50% (for carriers; Kotsopoulos et al. 2015). There are several difficulties in promoting the expanded use of oral contraceptives as a population health measure against ovarian cancer. First, oral contraceptives are already in wide use—in some studies up to 80% of the women had some exposure (Collaborative Group on Epidemiological Studies of Ovarian Cancer et al. 2008). Second, the lifetime risk of ovarian cancer is only 1.4% and it is not clear that there is sufficient worry about cancer at this level of risk to consider population-based prevention options. Third, the majority of ovarian cancers occur in women in their sixties, decades after the typical ages at which oral contraceptives are consumed. It is unlikely that a woman in her twenties will consider the potential for a risk reduction of less than 1% for getting ovarian cancer far in the future when she is contemplating taking the pill. Further, there is some evidence that oral contraceptives increase the risk of breast cancer in women with a BRCA1 mutation if initiated before age 20 and if taken for more than 5 years (Kotsopoulos et al. 2014). There is also a small increase in breast cancer risk in non-carriers who take the pill (Mørch et al. 2018). However, the protection against ovarian cancer with the pill is lifelong, whereas the risk of breast cancer is modest and returns to normal upon cessation of the pill (Kotsopoulos et al. 2015, 2014; Collaborative Group on Epidemiological Studies of Ovarian Cancer et al. 2008; Mørch et al. 2018).

There has been some interest in using aspirin as a preventive agent for ovarian cancer in BRCA1 and BRCA2 carriers (Tsoref et al. 2014). A randomised trial of aspirin had no effect on the incidence of colorectal carcinoma among carriers of genes predisposing to Lynch syndrome (Burn et al. 2008).

Screening for breast cancer

The premise of breast cancer screening is that mammography can detect cancers at a stage when they are small and more likely to be node-negative than are cancers that are identified as a palpable mass. There continues to be controversy about the benefits of mammographic screening—the majority of clinicians and radiologists believe that formal screening is associated with a mortality reduction as large as 30% (Tabar 2011; Nelson et al. 2016)—however, many epidemiologists believe the benefits of screening have been over-sold (Biller-Andorno and Jüni 2014; Bleyer and Welch 2012; Baum 1997). For them, the benefit of screening has been over-estimated because of over-diagnosis. Recently, we have challenged the paradigm that the canonical size-survival relationship that is seen is cross-sectional studies can be used to infer a mortality benefit from downstaging cancers through screening efforts (Sopik and Narod 2018). This paradigm is discussed in detail elsewhere (Narod and Sopik 2018; Sopik and Narod 2018). In any case, the benefit of screening is critical to advocates of personalised medicine because, outside of those women who opt for preventive surgery, personalised prevention of breast cancer is reliant on the premise that breast screening works, and if we are convinced that screening works, then it is sufficient to show that downstaging of breast cancer is an indicator of success of a screening program (i.e., it is not necessary to demonstrate a reduction in mortality).

Two large prospective studies are now under way to evaluate this paradigm; the American WISDOM study (Shieh et al. 2017) and the Canadian perspective study (Personalized Risk Stratification for Prevention and Early Detection of Breast Cancer 2018). It is proposed in these studies that the personalised risk profile based on a SNP-based model supplemented by personal history and mammographic density can be used to tailor screening. In the event that a woman’s risk score places her in the top percentile, she is offered more frequent mammography screening, earlier screening (e.g., from age 40) or more intensive screening (e.g., with MRI). We can conduct a thought experiment to evaluate the expected benefit of offering high-risk women screening from age 40 instead of from age 50. Assuming that regular mammography screening reduces mortality by one-third, we can calculate the expected impact of personal risk profiling at a population level. Assume that 1% of these women with the SNP test are positive (three-fold risk increase) and qualify for mammography at age 40 instead of age 50. For them, the risk of cancer between age 40 and 50 is now 3% instead of 1% (the baseline risk). Assume the baseline case-fatality is 20% for breast cancer patients in this age group (Liu et al. 2015). If 100,000 women go for the test, then we expect 1000 to be positive at the one percentile cutoff. We expect that 30 of these will develop breast cancer between age 40 and 50. Assuming a mortality rate of 20% in the absence of mammography then six will die of breast cancer. Assuming a mammography benefit of 30% in reducing mortality then we expect to prevent two deaths among the 100,000 women. We expect that of 100,000 women, 2000 will die of breast cancer in the absence of the genetic test, and with the introduction of the genetic test, 1998 women will die of breast cancer. This is the best case scenario. If the test is offered by a private laboratory and if only one in ten women is willing to pay for the test, then the expected benefit is two deaths among one million women. And this assumes a benefit of mammographic screening of 30% mortality reduction. If the actual hazard ratio were less than this, then the benefit would be even less. It is a testament to the faith of the research community in the canon of personalised medicine that these studies have been funded despite their remote chance of success. Further, this funding represents a substantial opportunity cost, in that it has diverted resources from alternatives that might have a greater chance of success.

Population genetic testing

For the two genetic testing scenarios described above (high-risk genes and SNPs), the benefit is maximized as testing becomes widespread. This is true of any population-based screening program. In this sense, testing is universal but the intervention is personalised, based on the result of the test. The probability of having a mutation depends on the family history of the patient but if we rely solely on the family history of cancer to guide our testing efforts then the impact of testing will be too small to be noticeable at the population level (Finch et al. 2014a). This is because most women with a mutation will not meet the minimal testing criteria and others will find the referral process burdensome and slow.

In the case of SNP-based genetic classifiers (personal risk profiling) family history cannot be used to guide testing. This is because the risk scores in the mother and daughter will only be loosely correlated and the effects of family history and the risks score appear to be independent, so a family history of breast cancer is not going to help us select women for testing. It cannot be assumed that those with a positive family history of cancer will have a much greater risk score than those who do not. To be effective, SNP testing must be offered to the whole population. Therefore, the benefit of a program at the population level is proportional to the number of women who opt for testing.

In both scenarios (high-penetrance genes and SNPs) we expect that the benefits can be maximized by testing all eligible women in the population. Our early studies in Poland and in the Jewish population of Ontario have shown that his can be cost-effective and acceptable strategy (Metcalfe et al. 2010; Gronwald et al. 2006a). We have shown that it is reasonable to offer genetic testing for all women in the Ashkenazi Jewish population (Metcalfe et al. 2010) and that one can market genetic testing directly to women via a notice in a magazine without incurring harm (Gronwald et al. 2006a). However, these were not ‘real-world’ studies in that the patient did not pay for the testing. Currently the cost of genetic testing through several American laboratories is sufficiently low that, for the majority of women, money is not an obstacle. Based on the lowered cost of testing and the expected benefits, we and others have argued in favor of population-based genetic testing (Levy-Lahad et al. 2014; Akbari et al. 2017). In Canada, we have initiated TheScreenProject which provides genetic testing for BRCA1 and BRCA2 to all Canadian women at a cost of 165 USD (Akbari et al. 2017). It is offered online direct-to-consumer with a saliva kit being shipped directly to the patient. The ScreenProject offers an inexpensive accurate and universal test with minimal patient burden with the potential to decrease cancer rates in the population; however, to date, accrual has been relatively low and we cannot expect to have an impact on Canadian cancer rates unless testing becomes widespread. In the first year since initiation we have identified 16 carriers through the ScreenProject. At the same time there were 2800 deaths from breast cancer and 1800 deaths from ovarian cancer in Canada. The reasons for the low uptake are not known but it is likely that the test is not yet well known. It is possible that if an effective chemoprevention drug were available for BRCA carriers, then the incentive to be tested would be greater.

Treatment

Personalised medicine has also been adapted to cancer treatment. Genetic testing of tumour DNA or measuring protein expression within tumour cells may predict response to therapies. It has been known for decades that ER-expression predicts response to tamoxifen and HER2 expression predicts response to trastuzumab (Hoskins et al. 2009; Schmidt et al. 2016). However, given the lack of newer biologic treatments in the 20 years since trastuzumab was introduced, these advances should not be considered to be proof of principle. Several studies are now attempting to identify other targets within breast cancers so that we can extend the individualized approach (Schmidt et al. 2016; André et al. 2014; Le Tourneau et al. 2015; Tsimberidou et al. 2017). As treatment becomes more ‘personalised’ the proportion of patients who might benefit diminishes and the impact on cancer mortality becomes small. The greatest impact on cancer mortality has come from the introduction of the chemotherapies cyclophosphamide and Adriamycin which are ubiquitous in their use (Narod et al. 2015).

A non-specific goal of genomic medicine is to help decide who, among women with a low risk of metastases, might benefit from chemotherapy. Tests for patterns of gene expression based on a panel of genes (e.g., the 70-gene panel Mammaprint or the 21-gene panel Oncotype Dx) fall into this category. In general, these tests are useful to identify women who might safely avoid chemotherapy—so the benefit should be measured in terms of reduced morbidity rather than reduced mortality.

The advent of next-generation sequencing has given us a position from which to view the genomic landscape of breast cancers. Most breast cancers harbor at least one mutation, and many cancers have many more (Stephens et al. 2012). The top ten genes ranked according to prevalence of mutations are PIK3CA (27% of all breast cancers in the COSMIC database); TP53 (23%); CDH1 (11%); GATA3 (11%); MED12 (10%); KMT2C (9%); ESR1 (6%); NCOR1 (4%); PTEN (4%) and ARID1A (4%%). Of these, three (TP53, CDH1, and PTEN) are also susceptibility genes. A gene/cancer combination is said to be “actionable” if there is an investigational or approved therapy for a cancer conditional on the genotype (Wagle et al. 2012; Harismendy et al. 2013; GC001 Panel v2.3 HOTSPOT ANALYSIS 2016). This is a valid concept for research but it is not appropriate that actionable be conflated with demonstrated clinical benefit. A few studies have surveyed the genomic landscape of a number of cancers and have divided those into actionable or not (André et al. 2014; Le Tourneau et al. 2015; Tsimberidou et al. 2017). Depending on whether a drug was available and given there were some few patients for whom tumour shrinkage was noted. Historically, the highest standard for evaluating drug efficacy is the randomised trial. For personalised genomics-based treatment we are asked to accept that ‘response to treatment’ is also a measure of success. A patient might have a positive response to treatment but no increase in life expectance. As one prominent cancer researcher describes it (Stand Up to Cancer Canada 2015) “We will give our new drugs to patients for whom other treatments have not worked, and monitor them to see whether a significant percentage of these patients responds to one or more of these drugs. By respond, we mean that we hope these patients’ tumours shrink, or at least that their disease holds steady rather than getting worse”. Genomic-based trials typically target women who have disseminated breast cancer and who have progressed on standard treatment. For these women lifespan might be extended but in the absence of cures, the mortality rate from breast cancer in Canada will remain unchanged. In order for us to see a decline in the number of deaths at a national level statistically it is necessary that these drugs be tested in the adjuvant setting, where the possibility of cure is substantial.

Conclusions

In this review I have covered selected topics to help us better understand the claim that the translational activities emanating from genomic discoveries will impact on cancer incidence and mortality. To date, there is little evidence to expect that personalised medicine will have a noticeable impact on these indices. There are several reasons for this, to summarise:

  1. 1.

    The proportions of cancers that occur in those women identified to be at high risk tend to be small. It is not the case that a small proportion of women can be identified that will develop the majority of the cancers.

  2. 2.

    For population-based programs, the proportion of cancers prevented correlates with the proportion of women tested and this is often a very small number.

  3. 3.

    There is little empiric evidence to date that patient decisions regarding prevention and screening are predictable and are grounded in risk.

  4. 4.

    Novel treatments for breast and ovarian cancer are almost always introduced after recurrence when the chance of cure is very small.

  5. 5.

    The primary endpoint of novel treatments for cancer is usually progression-free survival which correlates poorly with survival.

  6. 6.

    There is no empiric evidence that screening tumours for mutations has an impact on cure or survival. ‘Actionable’ is a process misrepresented as an outcome.

When considering if a personalised medicine intervention is likely to have any impact, ask:

What proportion of the diseases burden falls into the fraction who have the risk marker? (e.g., BRCA1 mutation, high personal risk score).

How likely is a woman going to have the test?

How effective is the intervention?

How likely is the woman in the high-risk category going to have the intervention?

In the case of cancer treatments, the most important question is if there is a treatment which increases the cure rate or extends life expectancy in the identified target group

There are several limitations to this report. The report focuses on breast and ovarian cancer and there is similar research underway for prostate, colon, and pancreatic cancer. The outcomes of cancer incidence and mortality are chosen as the canonical measures of success but others will argue that patient satisfaction and quality of life are as important as cure. There are several successful cancer gene projects in place which have no doubt made an impact, these include for medullary thyroid cancer, retinoblastoma and familial polyposis, but these are classical genetic syndromes and the effective prevention programs were developed prior to the era of precision medicine. There are many topic areas which have not been covered here, such as pharmacogenetics and non-malignant conditions, such as rheumatic disease, systemic lupus erythematosus etc. Personalised medicine is an emerging field and perhaps it is too early to undertake a critical review.

Nevertheless based on these data, I believe that the optimism of personal medicine is unfounded. The example which we believe most closely meets the criteria for population-based program is universal testing for BRCA1 and BRCA2 mutations. We have an accurate and inexpensive test which can be accessed online and the testing process is simple. The risks associated with the mutations are known with accuracy and are substantial, it is relatively simple to interpret a genetic test result if restricted to known pathogenic mutations in BRCA1 and BRCA2, we have effective means of prevention (preventive surgery) and the intervention is acceptable to high proportion of the women. It is important that surgery be offered judiciously as a means of reducing risk not just to mitigate anxiety. It is also desirable that the clinical attention be focused on those with a clearly positive test and we should avoid communication of uncertain results and we need to be assured that those with a negative test do not incur additional costs to the health care system. We must avoid adding tests of dubious value to the panel, such as ATM, which will no doubt increase health care costs without generating any benefit. The limits of the paradigm are that only 4% of breast cancer and 13% of ovarian cancers are due to BRCA1 and BRCA2 mutations and to date, there is a very low uptake of testing to date when offered as a direct-to-consumer product.