Personal genomics: information can be harmful

Information is generally considered a good thing. Information may of course lead to benefit through some action based on information. But, action can also lead to harm. We know about harm not only from everyday experience and common sense but also from the field of clinical medicine. Approaches to weigh harm vs. benefit have been developed in clinical medicine to aid doctors, patients and policy-makers in deciding whether to obtain and how to use certain kinds of information. When personal genomics is used to improve health and prevent disease, the fundamental issues relating to benefit and harm of information are exactly the same as in the clinical medicine field of prevention. If information about risk or prognosis can be harmful in clinical medicine, is there a difference just because the health information is ‘genomic?’ In this issue of the Journal, Gulcher and Stefansson argue that genetic risk information obtained from ongoing discoveries of genetic factors for common complex diseases can be useful for prevention and early detection now [1]. They make the case that risk assessment tests based on these discoveries represent tools for more cost-effective prioritization of scarce resources to address common and costly and deadly common diseases. They make the case that potential benefits of early detection are intuitive, biologically plausible and generally good for healthcare utilization and individual decision making; we contend that potential harms of such tests need to be explicitly considered and evaluated in appropriately designed studies before such tests are used on a population-wide basis in otherwise healthy people. In this commentary, we suggest that personal genomic information used for health promotion or prevention can cause both benefits and harms; we describe briefly how the field of clinical medicine has developed principles and processes to assess benefits vs. harms; and we show that genomic information used in health promotion and prevention is no different from other kinds of ‘clinical’ information and should be subjected to the same kinds of evidentiary considerations. Our fundamental argument is that the presumption of health benefits from personal genomics needs to be rigorously explored using research to assess the relative benefits and harms of genetic information before such tests become widely available in the population. The idea that information can cause harm – by action based on that information – would seem obvious. Two stories illustrate, one apocryphal, the other not. The apocryphal story is about the man who, after receiving information from an astrologer that a terrible thing will happen, lives in fear, unable to take any action. He survives to an old age, frightened, paralyzed and unfulfilled. The true story is about a relative in my (DFR) family. As a young girl, she received information that she had a heart murmur. Fearing she was an invalid, she avoided sports, exercise and strenuous activity. Years later, learning it had been a benign murmur with no health implication at all, she regretted losing years of youth that she would have lived differently. In clinical medicine, harms from information are well-established. The approaches developed to understand benefit vs. harm of information in clinical medicine may have useful application in the field of personal genomics. The motivation to develop a method to weigh benefit vs. harm came from the field of prevention. It was assumed, at one time in the past, that ‘prevention’ or ‘early detection’ of disease was good. Getting information from a chest x-ray to predict risk of dying from lung cancer, for example, was considered good because action – early surgery – could prevent death from lung cancer. Similarly, measuring blood pressure, to predict risk of dying from heart disease or stroke, was considered good because it was assumed that treatment could prevent a bad outcome. In the 1960s and 1970s, however, thoughtful clinicians and clinical epidemiologists began to wonder whether in some circumstances net harm could occur [2, 3]. What if treatment did not work? What if there were side effects caused by treatment? There might be some benefit, of course, but at least the idea of harm – and measuring it and weighing it against benefit – also had to be considered. These clinicians thought also about non ‘medical’ harms that might hurt their patients. For example, could side effects be caused simply by ‘labelling’? In a landmark randomized controlled study of ‘labelling,’ it was discovered that simply labelling subjects as having elevated blood pressure led to increased lost days of work, apart from any symptoms that might be caused by mild blood pressure or medication side-effects [4]. Examples of harms in clinical medicine are extensive, as routinely considered in the analyses of the US Preventive Services Task Force (USPSTF) [5] and others and the topic is much bigger than can be discussed here. Indeed, the entire field of ‘evidence-based medicine’ has strong roots in the concern about the balance of benefit and harm. Simply stated, the field of ‘evidence-based medicine’ tries to describe, for any action (e.g. getting information about risk of cancer determined by a screening test and acting on it; deciding about therapy), what benefit and harm result and how they weigh against one another. Methods to gather evidence about benefits and harms are transparent, quantitative and explicit (including different ‘levels’ of evidence) as are the methods to weigh evidence, and to reach practical recommendations [5]. The approach includes methods to avoid conflict of interest, for example, among commercial or professional groups who might be invested in certain procedures or tests. Using this kind of approach, chest x-ray information about lung cancer is considered to confer more harm than benefit (because surgery leads to harm and of morbidity without clear demonstration of benefit from reduced lung cancer mortality), while getting information about blood pressure and treating it may, in general, lead to net benefit. Beyond the many detailed analyses of specific problems, the USPSTF developed important principles for assessing tests used in screening or prevention among generally healthy populations (as would often be the case in screening, prevention, and personal genomics). One principle that became apparent to the USPSTF, as they grappled with these issues, was that the burden of proof is particularly high – to know that we are providing ‘more benefit than harm’– if one proposes to test otherwise healthy people. The rules and burden are lower in dealing with sick people, where, at an extreme like for treating advanced cancer, desperate diseases may require desperate measures. Another principle is that the main ‘outcome’ of interest is a health outcome like morbidity and mortality. The outcome of patient ‘satisfaction’ or ‘empowerment’ is not particularly important in itself, in part because it is so easy for patients to be wrong (and for doctors, for that matter). Reasoning from anecdote or personal experience, rather than evidence and formal analysis, it is easy for individuals to believe that they are making the right decision and are empowered by information, when in fact they are not. A classic example is the decision about penicillin for sore throats. Patients (and doctors) may believe, when the patient gets better following treatment, that treatment was the cause. Empirical evidence shows that a vast majority of sore throats get better by themselves and that treatment may result in net harm [6]. Similar dynamics operate in prostate cancer screening, where treatment may cause substantial harm by making men impotent and incontinent in high double-digit rates [7]. Nevertheless, men routinely say they ‘would do it again’ because they think they have ‘beaten cancer’ [8]. As the dynamics and forces that create personal satisfaction or empowerment work entirely independently of whether treatment actually works [7], we must be extraordinarily cautious before using ‘satisfaction’ or ‘empowerment’ as a component of outcome. Indeed, in the field of personal genomics, one observer has written that “society should not succumb to fantasies about ‘empowered’ individuals making free, informed choices in an unregulated genomic marketplace” [6]. It is hard to see how genomic information is different from clinical information, about risk factors for disease or markers of prognosis. Indeed, some genetic information is explicitly considered to be potentially harmful; as one observer writes, ‘For adult-onset conditions such as macular degeneration, other debilitating neurologic conditions, or common malignancies, the potential emotional turmoil and need for support and guidance before and after genetic risk testing is evident’(sic) [9]. In contrast – and for reasons not at all apparent to an observer in the field of clinical medicine –‘direct-to-consumer marketing of genomic profiles seems to have escaped the careful vetting that accompanies the introduction of new biomedical technologies’ [9]. Genetic or genomic information may be particularly worrisome – and possibly harmful – if it predicts a bad outcome with a strong degree of certainty. (Whether an action based on information leads to net benefit is a separate question; the main concern here is ‘can harm occur.’). But information that is uncertain, incorrect or simply misinterpreted can be just as harmful, as demonstrated by the two illustrations at the beginning of this essay. Harms could result from treatments, procedures or surgeries that have side effects and harms that can include psychological, social and economic harms affecting tested individuals and their relatives and unnecessary costs to the healthcare system [10]. When Gulcher and Stefansson suggest that improved accuracy or ability to predict risk will be beneficial, they do not consider the possibility of harm [1]. For example, they state that improved accuracy of SNP testing, providing information beyond family history in predicting risk of prostate cancer, may save lives and be worthwhile. But the benefit of prostate cancer screening is hardly clear. The evidence cited by Gulcher and Stefansson about SEER data (which are observational) and Scandinavian data (which are not about screen-detected cancer) cannot provide a strong argument for early detection by prostate-specific antigen (PSA) screening. Furthermore, the most-important evidence about efficacy, from the two recent clinical trials or experiments [11, 12] shows that benefit of screening is low at best, if it exists at all. As discussed in an accompanying editorial [13], the challenge for patients and providers in deciding about whether to screen is to weigh the small degree of possible benefit vs. the known frequency and severity of the harms of impotence and incontinence. In fact, recent modelling analyses of the harms and benefits of prostate cancer screening in high risk persons (based on family history) ‘provide a sobering illustration of the frequency of harms men are likely to experience if they participate in PSA screening. The risk of having a false alarm rises strongly with age and with increasing familial risk’ [14]. Challenges of weighing harms vs. benefits similarly apply to the other medical screening, problems like for breast cancer, discussed by Gulcher and Stefansson [1]. Are we unfairly characterizing the results of the prostate cancer screening trials and focusing too much on harms? In his recent New England Journal editorial, Barry noted that, ‘Serial PSA screening has at best a modest effect on prostate-cancer mortality during the first decade of follow-up. This benefit comes at the cost of substantial overdiagnosis and overtreatment. It is important to remember that the key question is not whether PSA screening is effective, but whether it does more good than harm’ [13]. Similarly, the chief medical officer of the American Cancer Society said about the trial results, the ‘benefits of prostate cancer screening... are ‘modest at best and with a greater downside than any other cancer we screen for’ [15]. Benefits vs. harms is what the discussion is about, and should be about. Gulcher and Stefansson suggest that the REVEAL study results [16] indicate that genetic information does not cause harm, and that we ‘fail to cite any studies that show it is dangerous for people to receive their results of genetic risk assessment for common diseases’. The absence of evidence is not evidence of absence. Furthermore, extensive experience and research studies in the field of clinical medicine suggest that harm can occur from something as innocuous as labelling, as demonstrated by the RCT reporting increased absenteeism from receiving information about ‘high blood pressure’ [4]. For personal genomic tests, empirical studies are needed to assess both benefits and harms of providing such information to otherwise healthy people. In considering the impact of harms simply from fear or from specific action-based on information, we know that people routinely overestimate their risks, even when the degree of risk is ‘certain’ (meaning well-understood) and low. Among persons with Barrett’s oesophagus, who have roughly a 0·5% per year of developing oesophageal adenocarcinoma, patients estimate their average 1-year risk of cancer to be 13·6%– twenty times higher [17]. If the ‘disease’ of Barrett’s oesophagus is reported to insurance carriers, a practical real-world harm of increased life insurance rates may occur [18]. In the face of these ‘harms,’ it is not clear that there is any improved outcome of reduced cancer mortality from doing screening for Barrett’s oesophagus or for doing surveillance among persons with established Barrett’s oesophagus [19, 20]. To simply believe that genomic information has no harm and that information is somehow always positive or neutral would seem to ignore both common sense and extensive real-world experience. While genomic information may be uncertain, uncertainty is not the main problem in using genomic information. Although actions taken by individuals or their healthcare providers based on uncertain or faulty information can lead to more harms than benefits, reducing a test’s uncertainty or increasing its accuracy is not sufficient to establish net benefit. Even information that is ‘certain’ may cause harm. There are many sources of uncertainty. First, a personal genomic test may be inaccurate or interpreted differently in different testing platforms. A recent report described that when personal genomic results of tests conducted by two direct-to-consumer companies were compared among 5 people, there was less than 50% agreement of disease prediction for seven diseases tested [21]. Second, a test may not be highly predictive, explaining only a small portion of the heritability of common diseases [22]. Uncertainties also result from the lack of generalizable data from different populations and ethnic groups, the complexities of gene-gene and gene-environment interactions and from the lack of direct genotype risk markers [23]. Third, a prognostic assay may ‘evolve’, even if it is basically accurate at one time, but later changes as new risk markers are added. For example, the assessment of risk of type 2 diabetes in the Rotterdam longitudinal study can change depending on the numbers of markers added to TCF7L2 status. The authors showed that adding the other 17 polymorphisms caused 34% of participants to be reclassified from low risk to high risk and vice versa. In total, 39% of participants changed categories once when risk factors were updated, and 11% changed twice, i.e. back to their initial risk category [24]. While it would obviously be desirable for uncertainty problems to be resolved for any test, their solution would not assure that a test provides net benefit. Even if a test is highly accurate and highly predictive – as are some tests for autosomal dominant diseases –‘net benefit’ does not flow automatically from certainty, accuracy and ability to predict. Net benefit occurs only if information and action lead to improved outcome that is greater than harm. Outside the area of single gene disorders with high penetrance, we currently have little data on how people respond to genetic risk factor information. As suggested by several authors [21, 23, 25], behavioural research in personal genomics is necessary; one specific topic that needs to be studied is potential harm. To address the problem of weighing benefits vs. harms of genomic information, a NIH-CDC multidisciplinary panel [23] concluded that the field of personal genomics would benefit from explicit systematic assessment of genomic testing on a case-by-case basis, similar to that routinely done by the US Preventive Services Task Force, in analyses that can be understood by both consumers and providers. Similarly, the Centers for Disease Control and Prevention started an initiative, EGAPP [26], devoted to assessing genomic applications in practice and prevention. The EGAPP working group has adapted methods of the USPSTF to evaluation of genomic applications for specific intended uses, including personal genomics topics like genetic susceptibility testing for type 2 Diabetes or coronary heart disease. In contrast to these professional efforts, the effort to implement direct-to-consumer marketing, which bypasses professional organizations and considerations of benefit vs. harm, would seem to presume that harms do not exist or that they are trivial. Promotion of direct-to-consumer marketing risks appearing to do an end-run around the impartial analysis that would consider possible harms. We suggest that the public will be better-served by some kind of objective assessment – perhaps like EGAPP – that is deliberate, serious, formal and impartial. Do we need randomized clinical trials to establish the balance of benefits and harms of personal genomic information? Gulcher and Stefansson suggest that we should not wait for decade-long randomized clinical trials before using genetic risk profiles for common diseases [1]. In response, we suggest that the same ‘standards’ that apply to risk information, like a screening test in clinical medicine, should apply to genetic or genomic information. If RCT data are considered necessary to understand benefits vs. harms for a particular application, then so be it. Generally, for diagnostic tests, it can be that may not be needed if a new test is or more specific but of similar an old test if a new test is more than an old test, it can lead to the detection of of disease or from the treatment trials that only patients by the old test may not apply to these may be needed we can be that the new test the same and of disease as the old test and that is similar the of disease These principles need to be on a case-by-case basis, to personal genomic tests. The following from Gulcher and Stefansson how the argument of benefit vs. harms has been by an argument to of group of suggest that many of these markers have been as common risk factors in of of patients and of of in both and we should wait randomized clinical trials are designed and to the years or [1]. The is that, if markers have been as common risk factors in numbers of then we do not have to wait for clinical The problem is that research assessing degree of risk and accuracy of prediction a different than assessing whether action based on risk leads to more benefit than harm. A different kind of evidence – like that by a randomized controlled clinical trial – is necessary to assess benefit vs. harm. In we in an with discoveries of genomic risk the of the genetic for diseases [22]. we know even less about how consumers and providers respond to this type of information. A of the and providers are of these tests, and some providers have used results by consumers to change some of clinical In the of about and the potential ability to provide information about there seems to be an that information can be only good and cannot be In the of clinical medicine and prevention, there is evidence and experience that information and action can be In the field of personal genomics, we need to and on to best and improve the of and of at of Centers for Disease Control and