[an error occurred while processing this directive] [an error occurred while processing this directive]

How safe is safe? Communicating risk to decision-makers

by R A Kempton [adapted from an article in Science and the Scottish Parliament - Proceedings of a Symposium of the Edinburgh International Science Festival, 7 April 1998 organised by the Scottish Agricultural Science Agency - Edited by R K M Hay ]



The safety and well-being of its people is a chief responsibility of Government. Indeed, Cicero described it as the supreme law. How this responsibility should be interpreted in the complex world of the 21st century is hotly debated and will be a recurring issue for the new Scottish Parliament. Objective guidelines for dealing with real and potential risks will be of benefit to many aspects of Parliament's work, not just in areas of science and technology, such as health, agriculture and environmental protection, but also in, for example, social work and the law.

The BSE crisis has provided Government with a testing and traumatic experience in risk management and communication. Reassuring statements such as "Beef is absolutely safe to eat" (Government Chief Medical Officer, May 1990) and "I am absolutely certain that British beef is wholly safe" (Agriculture Minister Douglas Hogg, December 1995) were discredited when evidence was produced suggesting possible transmission of the BSE agent to humans. This led to an admission that "the term safe is not the same as zero risk" (Government Chief Medical Officer, March 1996). More recently, legislation to reduce the risk of transmission, by banning the sale of beef-on-the-bone to consumers, has been criticised as an over-reaction to what is now considered a miniscule additional risk of around one in a billion of contracting new-variant CJD (nvCJD).

Influenced by this and earlier food scares, Government intends to establish a Food Standards Agency with responsibility for protecting public health by promoting a safer food supply and ensuring that consumers have the required information to choose a safe and healthy diet. This two-pronged approach to risk management, which emphasises the importance of both legislation and public communication, represents a major advance in policy. The Guiding Principles of the Agency (CM3830, 1998) (Appendix 4) are to achieve a "coherent and consistent" approach to risk analysis and decision making which accepts that "complete freedom from risk is an unattainable goal, and safety (is) related to the level of risk that society regards as reasonable .... in comparison with other risks in everyday life". For this more open policy on food safety to succeed, and to be introduced into other policy areas, decision makers, be they government or members of the public, need to be better educated in interpreting estimates of risk. This paper aims to contribute towards the education process.


Understanding risk

Risk can be defined formally as the chance (probability) of the occurrence of a particular adverse event or hazard (RS, 1992). The National Lottery, introduced into the UK in 1994, has helped to increase awareness of the laws of probability among the general public, though it has also highlighted some serious misconceptions. Most risks do not, however, follow the simple rules of a lottery. In particular, the probability of an event may differ between individuals; it may not increase proportionately with exposure (the size of stake); the outcome from an event may be considerably delayed in time (as with nvCJD); and there may be a multiplicity of possible outcomes which cannot be sensibly expressed on a single (monetary) scale. We illustrate these complexities using some well studied areas of risk.

Table 2.1 compares the deaths from different forms of transport in the UK for the period 1986 - 90. Death rates are standardised for differences in exposure, which is identified with distance travelled. These figures may thus be used for comparing the risk of travelling a fixed distance by different forms of transport. Travel by coach or rail is seen to be safer than travel by car. However, the population statistics used to derive Table 2.1 produce average risks which may be quite inappropriate for a particular individual and journey. For example, the risk of deaths among car drivers will depend on age and sex (with highest risks among young adult males), on the type of road travelled (with motorways safer than A-roads and rural roads) and the level of alcohol in the blood (DETR, 1997). For low-risk drivers, the risk from travelling by car may thus be comparable to travelling by train.

Table 2.1 : Number of fatalities per billion km travelled in the UK by different means of transport 1986-1990 (RS, 1992)
Car 4.4
Rail 1.1
Coach 0.45
Car 4.4
Motorcycle 104
Bicycle 50
Pedestrian 70

The expected outcome from a particular hazard in a given time interval is determined by both risk and exposure. For travel accident statistics, we have the simple relationship:

Expected number of Deaths = Unit Risk x Exposure

For other hazards, the number of deaths may increase exponentially with exposure, or there may be a safe limit below which death rates are unaffected by exposure (Fig. 2.1). To assess the size of the impact of a hazard, we must take account of both risk and exposure. Despite its lower risk, the number of car deaths per year in UK is eight times the number of bicycle deaths, due to the population's much greater exposure to car travel.

Figure 2.1: Possible relationships between death rate and level of exposure to hazard.

Fig 2.1

Adams (1995) has argued that, by ignoring exposure, planners and legislators have misinterpreted accident statistics as direct measures of safety. He noted the surprising reduction, by more than a half, in children killed in road accidents between the 1920s and the present day, and explained the apparent paradox as being due to the reduction in exposure arising from changes in social behaviour, some of which are a direct consequence of the perceived increase in risk of children being on the streets. Hillman (1993) notes that an immediate effect of introducing legislation for compulsory wearing of bicycle helmets in Australia was to reduce cycling mileage (i.e. exposure) by around 25%, and that this largely explains the reduction in casualties. In the UK, road accident statistics show that the number of bicycle deaths for 1996, at 34% below that for the early 1980s, already meets the overall target of a one third reduction in road casualties by the year 2000 (DETR, 1997). However, this figure is almost wholly explained by a 30% reduction in bicycle mileage over the period. Indeed, when non-fatal injuries are included, the casualty rate per km cycled has risen by 24%.

When the outcome from a hazardous event is delayed in time, assessing its risk is much more problematic because of difficulties in affirming a causative link and in estimating the overall level of exposure. The risk from cigarette smoking has been demonstrated most convincingly in longitudinal studies, where a cohort is studied over a prolonged period (Fig. 2.2). This delayed mortality explains the higher smoking-related deaths currently observed in men. Although the proportion of smokers and average number of cigarettes smoked per day are now similar for both sexes in Scotland, many women smokers have adopted the habit too recently for its fatal effects to have emerged fully.

Figure 2.2: Survival of male UK doctors in a 40-year prospective study, 1951-1991, classified according to current smoking behaviour (Doll et al, 1994).

Fig 2.2

Confusion also exists in the interpretation of relative risk. In 1995, a warning by the Committee on the Safety of Medicines, that a third-generation contraceptive pill was twice as likely to cause blood clots as the second generation pill, caused panic among many women taking the new pill, and resulted in an estimated 8,000 extra abortions and an unknown number of unplanned pregnancies (Matthews, 1998). It later emerged that the absolute annual risks of death for the second and third generation pills were respectively three and six per million, and the increase in risk was well below the risks associated with pregnancy and abortion (DoH, 1996).

Another example concerns the drug Tamoxifen as a treatment for breast cancer. On the day of this symposium, the Guardian reported, as a front page headline, the preliminary results from a large US clinical trial which suggest that the drug may halve the risk of breast cancer. The drug is itself a known carcinogen and reported to cause a threefold increase in risk of cancer of the womb, as well as having other side effects. However, the drug is still expected to reduce overall mortality among women, particularly when its use is restricted to groups with high risk of breast cancer (Fig. 2.3).

Figure 2.3 Balancing cost against benefit of treatment. A hypothetical treatment reduces deaths among patients with a specified disease by 50% but has major side-efficts that kill 10% of all those treated. The treatment, then, has an overall benefit for patients with a high risk of dying from the disease, but leads to an increase in mortality if used for low-risk groups.

Fig 2.3

In the examples above, we have considered only a single adverse outcome, namely death. This is convenient for policy makers as mortality statistics are reliable and usually readily available; in contrast, injury statistics are reported inconsistently, and measuring the seriousness of an injury is necessarily subjective. However, mortality statistics may give a false comparison of the total costs of different risks. In transport accidents, for example, the frequency of injury is much higher than death, but the relative frequency of the two outcomes varies for different circumstances. Insurers deal with multiple outcomes by assigning an economic cost for each outcome and expressing the risk as the expected total cost per year for each insurance category. In medical treatments, the definition of risk may also be extended to take account of quality of life.


Tolerance and Perception of Risks

A risk-free society is unattainable; indeed, the restriction on action that such a society would impose would be unacceptable to most individuals. But what level of risk should Government consider tolerable when considering legislation or providing advice on safety? The Health and Safety Commission considered 1 death in 10 years to be an intolerable societal risk when reviewing major hazard aspects of the transport of dangerous substances (HSC, 1991), while 1 death in 10,000 years was considered a negligible risk. Considering risks from the individual's viewpoint, the first Royal Society Study Group (RS, 1983) suggested that an annual risk as high as 1 : 1,000 may be considered tolerable if there is some compensating benefit, whereas a risk below 1 in a million would be negligible. However, the Health and Safety Executive (HSE, 1988) considered an individual risk as low as 1 : 10,000 as intolerable, requiring immediate reduction, irrespective of cost. In practice, the number of individuals exposed to a risk will influence its impact on society and, consequently, its tolerance level. HSE policy for dealing with risks which are considered tolerable, but still not negligible, is to reduce them "as low as reasonably practical", the ALARP principle.

Identifying a single tolerable risk level for decision-making is difficult, as tolerance will vary considerably with individuals and context. Some individuals enthusiastically embrace very high risk sports such as hang gliding and rock climbing, while others dread flying in a commercial airliner, one of the safest forms of travel. The same individual will accept the personal risk from smoking, which is far higher than the tolerable limit of 1 : 1,000 proposed above, but be concerned about the much smaller risks associated with the carcinogenic properties of food additives or chemical residues in vegetables.

Such inconsistencies may result from differences in perceived dangers and rewards. An individual's perception and tolerance of risks with the same long-term cost may also be influenced by:

A cognitive map of different risks can be constructed by plotting the level of dread against the level of knowledge (Fig. 2.4). Dreaded risks seem to be those with potentially-catastrophic consequences that are involuntary and not easily avoided; unknown risks are often new risks, and other risks whose impact is not easily assessed. Public tolerance tends to be lowest for risks which are both unknown and dreaded, such as nuclear reactor accidents and DNA technology.

Figure 2.4: Cognitive map showing psychrometric characteristics of some new and familiar risks (redrawn from Cutter, 1993).

Fig 2.4


Communicating risk to the public

The Government is pledged to more open decision-making and the creation of a more informed society. To advance this cause, the Food Standards Agency will include a Risk Communication Unit to provide the public with clearly-presented information on food safety so that we can each make individual choices about what we eat. The desire for more information was expressed by Simon Jenkin at the height of the BSE crisis:

"I want to know, from those more knowledgeable than I, where a steak stands alongside an oyster, a North Sea mackerel, a boiled egg and running to catch a bus. Is it a chance in a million of catching CJD or a chance in ten million? I am grown up. I can take it on the chin. " (quoted in Times Higher Education Supplement, April 1996).

Can scientists hope to ever satisfy such requests? One of the difficulties the Agency will face is the communication of newly-identified, theoretical risks, where scientific information is too poor to give a confident estimate of risk level. Now that the BSE epidemic in cattle is almost over, any risk of contracting nvCJD from eating beef should be trivially small, but, because of difficulties over diagnosis and uncertainty over the average incubation period for nvCJD, we still do not know what the risks were at the height of the epidemic. This is reflected in the wide range given for the latest estimate of 100-35,000 eventual deaths from the disease (CJD Surveilance Unit, March 1998).

Knowledge of well-understood risks would, however, allow the public to assess the seriousness of new risks more effectively. The Royal Statistical Society (Smith, 1996) has promoted the construction of a "Riskometer", a simple sliding scale of risk, as an aid for comparing different risks. An example of such a riskometer appeared in The Annual Report of Chief Medical Officer of the Department of Health for 1995 (DoH, 1996) and an amended version appears in Table 2.2. The riskometer uses a Richter scale of annual probabilities of death ranging in multiples of 10 from greater than 1 in a hundred (high risk) to less than 1 in a million (negligible risk). As an anchor point, I have added the chance of winning the lottery (from 100 tickets purchased in a year): this chance is 1 in 140,000, slightly smaller than death from homicide, and classified as minimal. As noted previously, the risks of an event for a particular individual may differ substantially from these average population risks, according to age, sex and other circumstances. Nevertheless, the riskometer should prove useful for placing estimates of newly-identified risks, such as passive smoking and eating beef-on-the-bone, in the context of well-known risks.

Table 2.2 : Risk of death in any year from various causes (DoH, 1996; with additional data added in italics) .
Term used Risk estimate Example
High > 1:100 First year of life, Glasgow 1855 1:8
Moderate 1:100 - 1:1,000 First year of life, Scotland 1990 1:130
Smoking 10 cigarettes/day 1:200
All natural causes for 40 year old 1:850
Low 1:1000 - 1:10,000 All kinds of violence and poisoning 1:3,300
Influenza 1:5,000
Road accidents 1:8,000*
Very low 1:10,000 - 1:100,000 Leukaemia 1: 12 000
Accidents at home 1:26,000
Accidents at work 1:43,000
Lung cancer from passive smoking 1:50,000
Homicide 1: 100,000
Minimal 1:100,000 - 1:1,000,000 Winning Lottery (100 stakes) 1:140,000
Railway accident 1: 1,000,000
Negligible 1:1,000,000 - 1:10,000,000 Hit by lightning 1: 10,000,000
Radiation leak from nuclear plant 1: 10,000,000
Beef on the bone 1:1,200,000,000
*more recent estimate is 1:16,000

A riskometer of annual risks is not very informative for assessing risks with long incubation period. For example, while the riskometer gives a moderate risk for a smoker who smokes 10 cigarettes a day, the risk for any individual in a particular year will depend on his smoking history, possibly going back more than 40 years. Indeed, for an individual under 35, the risk of smoking-related death is negligible (Peto et al., 1994). Several alternative methods have, therefore, been proposed for expressing the risks from such long-term hazards, including the eventual likelihood of death (Table 2.3) and the average loss of years of life (Table 2.4) from the hazard.

Table 2.3 : Presentation of risk from smoking in terms of likely cause of death (Peto et al., 1994)
Among 1000 20-year-olds in UK who smoke cigarettes regularly -,
1 will die from homicide
6 will die from car accidents
250 will be killed by smoking in MIDDLE age (ie before 70)
250 will be killed by smoking in OLD age

Table 2.4: Expression of risk in terms of average loss of life from various hazards, calculated from years of life lost by those dying from the disease (columns 1 - 2) multiplied by the probability of death among all individuals at risk. (For more details see RS, 1992; p 76).
Average age at death (yrs) Average loss of life for those exposed to risk
Population Death from unexposed individuals those dying from hazard
All women Complications of
77.5 30.0 0.01
All men Motor traffic
72.0 44.9 0.30
Blue asbestos
Lung cancer
78.9 57.9 1.50
Mining accidents
68.9 50.7 2.40
Cancer of lung /
nasal sinuses
69.3 55.1 3.97
smoking 15 to
24 cigarettes a
day (male)
All conditions
associated with
76.3 61.6 5.45



Scientific and technological advances during this century have led to substantial improvements in the length and quality of human life. They have also brought with them new risks which have needed to be carefully evaluated and managed (BMA, 1987). However, commercial and public pressures result in an ever-shortening interval between the identification of a new risks will be introduced before they can be properly assessed. Furthermore, the apparent ability of scientists to measure ever smaller levels of risk, and the often sensationalised reporting of these risks, adds to public confusion and anxiety.

In recent years there has been an increase in public mistrust of the way governments deal with public health risks. A major shift in policy is required to rebuild this trust. To be successful, new policies must involve more open participation in the process of risk analysis and communication, including:

The Internet is providing government with an exciting new medium in which to develop its policy of openness. However, equally, it provides every pressure group and activist with a global platform to tell the public "what the government doesn't want us to know". This is currently seen, for example, in the large number of web sites devoted to "mad cow disease".

As I have emphasised in this paper, the communication of risks also needs to be tackled with care at the technical, statistical, level. In particular, there is a need to :

A new policy towards more open discussion of risk will bring with it many difficulties but, in this new age of communication, it is a challenge that neither scientists nor policy-makers can shirk. In Popper's (1966) words "we must plan for freedom, and not only for security, if for no other reason that only freedom can make security secure".



Adams J. 1995. Risk. London: UCL Press.

BMA. 1987. Living with risk. London: John Wiley.

CM3830. 1998. The Food Standards Agency: a force for change. Government White Paper. London: The Stationery Office.

Cutter SL. 1993. Living with risk - the geography of technological hazards. London: Edward Arnold.

DETR. 1997. Road accidents Great Britain: 1996. London: The Stationery Office.

DoH. 1996. On the State of the Public Health: Annual Report of the Chief Medical Officer of the Department of Health for the Year 1995. London: HMSO.

Doll R, Peto R, Wheatley K, Gray R. 1994. Mortality in relation to smoking: 40 years' observations on male British doctors. British Medical Journal 309: 901-911.

HSC. 1991. Major hazard aspects of the transport of dangerous substances. Health and Safety Commission Report. London: HMSO.

HSE. 1988. The tolerability of risks from nuclear power stations. Health and Safety Executive Report. London: HMSO.

Hillman M. 1993. Cycle helmets: the case for and against. London: Policy Studies Institute.

Matthews R. 1998. Hidden perils. New Scientist 21 Feb. 1998, 16 - 17.

Peto R, Lopez AD, Boreham J, Thun M, Heath C. 1994. Mortality from smoking in developed countries, 1950 - 2000. Oxford: Oxford University Press.

Popper KR. 1966. The Open Society and its Enemies, 5th edition. London: Routledge & Paul.

RS. 1983. Risk assessment. London: The Royal Society.

RS. 1992. Risk: analysis, perception and management. London: The Royal Society.

Smith AFM. 1996. Mad cows and ecstacy: chance and choice in an evidence-based society. Journal of the Royal Statistical Society A159: 367 - 384.

[an error occurred while processing this directive]