Back to the Scottish Parliament Health Committee Report
Archive Home

Business Bulletin 1999-2011

Minutes of Proceedings 1999-2011

Journal of Parliamentary Proceedings Sessions 1 & 2

Committees Sessions 1, 2 & 3

Annual reports

SP Paper 263

SUPPLEMENTARY SUBMISSION FROM TOBACCO MANUFACTURERS’ ASSOCIATION (Part 2)

Relative Risk

Epidemiological studies generally express their findings in terms of reported estimates of relative risk (RR). This is the ratio of the incidence of the disease being studied in the group exposed to ETS (generally non-smokers living with smoking spouses), to the incidence of disease in the group not exposed to ETS (generally non-smokers living with non-smokers).

The RR reported has no direct bearing on the probability that an individual will acquire the disease in question. RR provides only an index of the strength of any association between exposure and a disease, and is always a relative term to the incidence of disease in the non-exposed group.

In case-control studies, relative risk (RR) is most often now expressed as an Odds Ratio, as in the following example:

1.26 (95% CI 1.06-1.47)

In this example, say the RR of 1.26 is the estimated risk of the disease in non-smokers living with a smoker, relative to the risk in non-smokers living with non-smokers. Were it to be less than 1.0, it would indicate that non-smokers living with smokers were less at risk of the disease than non-smokers living with non-smokers.

Confidence Interval

CI is the ‘Confidence Interval’, which is normally stated at the level of 95%. It does not mean that there is 95% certainty that the stated RR - in the above example, 1.26 – is correct. The 95% actually refers to the frequency with which the statistical test used will generate boundaries capturing the true figure. In other words, it relates to the reliability of the test, not to the parameter.

Interpreting Relative Risk

In interpreting what a RR figure means in terms of the population, it is necessary to know what the ratio or incidence of the disease is in the population not exposed to ETS: in other words what the rate of death or disease is in non-smokers living with non-smokers.

As explained in the 1988 report of the Independent Scientific Committee on Tobacco and Health, in the case of lung cancer in the UK population, the rate of death or disease amongst non-smokers living with non-smokers is generally taken to be 10 per 100,000 person-years of the population1 .

Thus, in the above example, a RR of 1.26 would then mean that amongst non-smokers with smoking spouses, the incidence of the disease would be 12.6 persons in every 100,000 person-years of the population, as opposed to 10 per 100,000 in the case of non-smokers living with non-smokers.

RR is sometimes expressed as a percentage. Most frequently is this the case when the purpose, either of researchers, publications or reporters, is to make the risk more easily comprehended by the public. The outcome is generally the reverse.

For example, when a RR of 1.26 is expressed as an increased risk of 26%, the entirely wrong impression acquired by the ordinary person is that out of every 100 non-smokers 26 will suffer from the disease. What a relative risk stated of 26% indicates is that the incidence of the disease will be 26% greater amongst non-smokers exposed to ETS by their smoking spouses than it would be had they lived with a non-smoker. Given that the rate of death from lung cancer amongst non-smokers living with non-smokers is 10 per 100,000 person years, the percentage increase in risk is from 0.010% (amongst non-smokers living with non-smokers) to 0.0126% a year (amongst non-smokers living with smoking spouses).

However, such a very small increment in risk – 0.0026% - would not make news that demands loud, clear and unequivocal headlines and sound bites. If that kind of message is not provided by the research itself, or by the professional journals publishing their work and wanting to promote their own publications, the danger is that it can then be generated by reporting that lacks thoroughness and concern for detail and accuracy.

A recent example of the misuse of science was provided by an estimate2 that claimed that ETS exposure caused the death of 49 workers in UK pubs and bars each year. This figure was arrived at by using relative risks for lung cancer, heart disease and stroke for home and workplace exposure that were used in a New Zealand review paper3 ; assuming a workforce in pubs and bars of 53,500 of which half were permanent staff; assuming that all of the workforce was exposed 100% of the time over a 6-hour shift to 3 times more smoke than would a non-smoker at home living with a smoker; and assuming that all the workers in those places were non-smokers. The review paper from which the relative risks were drawn did not claim precise predictions but only a guide dependent upon many assumptions and unknowns. The researcher’s assumptions were highly speculative, but the estimate suffers from a much larger flaw - the assumption that a relative risk for a chronic disease, which is the result of prolonged exposure over forty or so years, can be applied to a population group which is much younger (as well as one which also changes jobs frequently), with a consequently much smaller duration of exposure. The incidence of lung cancer, heart disease and stroke, below the age of 40 is very low and the age distribution of workers in the hospitality trade on average is very different from those exposed to ETS at home. As if that were not sufficient, an additional, fundamental error in the data used effectively destroys all possible credibility in the claim that was made.

Even though some may regard the public as being scientifically illiterate and mathematically innumerate, that is not a reason for the public to be misled, simply because of the perceived need to achieve headlines.

How the magnitude of a relative risk should be interpreted

In statistics, the words ‘statistical significance’, or ‘statistically significant’, have nothing to do with the magnitude of a measured difference. Statistical significance does not imply real life significance. It is a probability statement of the likelihood that the results did not occur by luck or chance if the groups were really alike; about how certain it is that the results are not a fluke.

Traditionally, conventionally and historically, a RR is considered to be statistically significant – not a fluke - when at a 95% CI it does not include 1.0, albeit that the choice of the value of 95% CI is arbitrary.

A RR finding of around 3.0 is generally considered necessary in order to establish cause. For example:

“The association between cancer occurrence and exposure to either extremely low frequency (ELF) or radiofrequency (RF) fields is not strong enough to constitute proven causal relationship, largely because the relative risks in the published reports have seldom exceeded 3.0…”4

A RR of 2.0 or less is generally regarded as being weak and not indicative of a causal association.. The nearer the RR to 1.0, the more likely is this to be the case:

“…relative risks of less than 2.0 are considered small and are usually difficult to interpret … Such increases may be due to chance, statistical bias, or effects of confounding factors that are sometimes not evident.”5 .

“…when the relative risk lies between 1 and 2 . . problems of interpretation may become acute, and it may be extremely difficult to disentangle the various contributions of biased information, confounding of two or more factors, and cause and effect.”6

“Until the 1980s, epidemiologists were concerned mainly with relative risks that exceeded about 1.5 and were often much higher. Many controversies now centre on much lower risks, a notable example being the effect of ‘passive smoking’ on lung cancer risk. The pooled data show a statistically significant effect, and all studies are consistent with a relative risk of about 1.3 (US National Research Council, 1986). In view of the many difficulties discussed above, however, it can plausibly be argued that such small effects are beyond the limits of reliable epidemiological inference (particularly for lung cancer, in which the major cause produces large relative risks), as smoking habits may be inaccurately recorded and are correlated with many other social and occupational factors, including the smoking habits of spouses. A number of spurious associations with relative risks for lung cancer of this order might thus be found in a large enough sample. The observations that short-service workers in various industries suffer elevated risks for lung cancer, which seem unlikely to be caused by their recorded occupational exposure, further illustrates the problem.”7

Yet, in the case of lung cancer and ETS, a 1997 meta-analysis8 accepted by the UK authorities found a RR of 1.26 (95% CI 1.06 – 1.47), derived amongst non-smokers living and not living with smoking spouses. That has been claimed to be a "substantial" excess risk and one warranting bans on smoking in work and public places. That is simply not correct.

In 1992, the US EPA found a RR of 1.19 for lung cancer associated with ETS. However, that was only statistically significant at a 90% CI; it was not significant at 95% CI at which it included 1.0. Nonetheless, in 1998 that report was used as a basis for listing ETS as a known human carcinogen.

IARC’s 1998 report9 was a case-control study of lung cancer and exposure to ETS in 12 centres from 7 European countries that the researchers claimed provided “the most precise available estimate of the effect of ETS on lung cancer risk in Western European populations.” It reported no overall statistically significant increase in risk of lung cancer from ETS in any of the situations where people were exposed to ETS. The conclusions of the study stated:

“Our results indicate no association between childhood exposure to ETS and lung cancer risk (0.78 (95% CI 0.64-0.96)). We did find weak evidence of a dose-response relationship between risk of lung cancer and exposure to spousal (1.16 ( 95% CI 0.93-1.44)) and workplace ETS (1.17 (95% CI 0.94-1.45)). There was no detectable risk after cessation of exposure.”

In other words, not only were relative risks found to be low, but at the 95% Confidence Interval they included 1.0, indicating that they were not statistically significant. The following observation was also made in the report:

“The available literature on ETS exposure from the spouse and lung cancer is large. However, only six studies are available from Europe; two of them, conducted in Greece, showed a twofold increase in risk for women ever married to a smoker. Of the other studies, one from Scotland provided very unstable risk estimates of the same magnitude as the Greek studies and two – one from the UK and the other from Sweden – provided little evidence of an association.”

The results were within the range at which the IARC itself concluded that unequivocal results may be forever unachievable. Yet after negative reporting of the results by the media, IARC insisted that the findings “add substantially” to previous evidence of the risk between ETS and lung cancer. A WHO press release then implied that the results proved a link between ETS and lung cancer, a highly problematic conclusion given their own guidelines of epidemiological best practice10 .

It is difficult to see how it could be claimed that the study adds substantially to the case against ETS and much less does it prove a link between ETS and lung cancer. The interpretation of such weak evidence is not in line with the official interpretation of very similar findings of other supposed health risks.

For example, a major study11 of the supposed link between electric power lines and childhood leukaemias produced a RR of 1.24, with a 95% Confidence Interval of 0.86 - 1.79. The researchers concluded that this provided “little evidence” of a link between power lines and leukaemia. The US National Cancer Institute went further, declaring that the study showed magnetic fields “do not raise children’s leukaemia risk”.

Another study12 of women with breast implants found a RR for hospitalisation for connective tissue disorders of 1.3 with a non-significant 95% CI of (0.7 – 2.2), again close to the IARC passive smoking study. But whereas the IARC findings were claimed to prove a link between ETS and lung cancer, in the breast implant study they were found not to be associated “with a meaningful excess risk of connective tissue disorder”13 .

What is absent is an explanation as to why the low RRs that have been reported in respect of lung cancer and ETS - with 95% CIs often including 1.0 and any excess risk capable of being accounted for by only modest degrees of bias and confounding, or by inadequate statistical adjustment for such factors - are regarded by some as providing incontrovertible proof of a causal link. And also why the interpretations of ETS RRs are not in line with the general guidance provided in 1998 by the Government in answer to a Parliamentary question, albeit incorporating an incorrect explanation of a CI:

“Relative risk provides a measure of the strength of association between a factor and an illness. It is an important way of measuring increases or decreases of risk over time or between different groups by comparing the incidence of an illness or hazard within a population to some baseline (for example, if drinkers are twice as likely to suffer from a particular disease as compared with the general population, a factor of 2 may be cited). A stronger association of greater than 2 is more likely to reflect causation than is a weaker association of less than 2 as this is more likely to result from methodological biases or to reflect indirect associations which are not causal. The significance of any such number does though need to be considered in context and from a number of viewpoints.

First, there is a statistical significance: in other words, what confidence is there in the number itself. This will depend on the quality and extent of the available data. Scientists usually express these by giving a confidence interval: rather than by saying that the relative risk factor is 2, they will say that (for example) one can be 95 per cent certain that it lies between 1.6 and 2.4.

Even when the strength of an association is precisely determined, it is insufficient in itself to confirm a direct causal link between possible cause and illness. The strength of an association is only one of several criteria which must be considered in the assessment of causation. Other criteria include:

the cause must precede the effect;

the biological plausibility of the association - is the association consistent with other knowledge e.g. experimental evidence?

the consistency of the finding – is the same result obtained from different studies using different methodologies elsewhere?

the presence of a “dose-response” relationship – an increased response to the possible cause being associated with an increased risk of developing the illness.

All these factors would be taken into account in trying to pinpoint cause.

The practical significance of risk factors, also needs to be considered and depends on how great is the underlying risk. Doubling a very small probability (risk), say 1 in 10,000,000, still results in only a very small risk of illness. Doubling a risk of, say, 1 in 100 could, depending on its nature, be more serious.

In practice, scientific judgments will be made and debated on a case-by-case basis. The Government can draw on the expertise of independent scientific advisory committees which are constituted to provide balanced judgment on the questions covered above”14 .

The factors mentioned in that important Parliamentary answer are included in the criteria that were proposed by Bradford Hill15 to guide the evaluation of a body of evidence as to whether or not an association between an outcome and a putative risk factor is causal. In the case of ETS, the study findings do not come close to meeting the Bradford Hill criteria for causality. In particular, they are not consistent, generally produce very weak or no excess risks, and rarely show dose-responses.

The nature of ETS

ETS is a mixture of the smoke released from the burning end of a cigarette (termed “sidestream” smoke) and the smoke exhaled by the smoker between puffs16 . This smoke quickly mixes with the ambient air and becomes highly diluted and, as a result, there are important differences between the level and the chemical and physical composition of the “mainstream” smoke inhaled by the smoker and ETS.

In all normal circumstances, ambient air contains a large number of substances, whether or not smoking has taken place17 . Such substances can include dust, pollen, bacteria, fungi, trace chemicals from vehicle emissions and other sources of pollutants, as well as, in certain circumstances, emissions from cooking and heating appliances. Research suggests that the types of substances found in indoor air are generally similar, with or without the presence of ETS18 .

It is extremely difficult to measure real-life ETS. The concentrations of the various substances that make up ETS are generally extremely low and many of the chemicals present in ETS are, irrespective of ETS, likely to be present in the air anyway, emanating from other sources. Moreover, ETS is a complex and constantly changing mixture, making it difficult to extrapolate total ETS exposure from the measurement of an individual chemical marker.

Nonetheless, the results of studies seeking to quantify exposure suggest that concentrations of chemicals in ETS are typically much lower than permissible exposure limits to these chemicals approved by regulators19 . Studies have, not surprisingly, also reported that non-smoker exposure to ETS is a great deal lower than the smoker’s exposure to mainstream smoke. Generally such studies have looked at exposure to nicotine, not because airborne nicotine is widely thought to cause lung cancer, heart disease or respiratory disease, but because it is almost unique to tobacco smoke and can be measured even at low concentrations.

For example, one study20 reported that, on average, in the course of a year, non-smokers had an exposure to airborne nicotine which was less than the amount delivered to a smoker by just five cigarettes with a yield of 1mg per cigarette. Another study21 of British women exposed to ETS in various settings reported that on average a non-smoker would only be exposed to the equivalent nicotine of smoking a single cigarette over a period in excess of two years.

A variety of studies which have measured the biological metabolites of nicotine have suggested ETS exposures of an average of 0.2% to 0.4% of active smoking, while estimates of particulate exposure suggest a factor of around 0.05% to 0.1%.

Measuring uptake, as compared with exposure, of ETS by non-smokers presents its own problems. The most commonly used markers are nicotine and its metabolite cotinine, which can be analysed in body fluids. Subjects do vary, however, in the rate at which they metabolise nicotine. Nicotine and cotinine are also not quantitative markers for other components of ETS. Most scientists also accept that there is a threshold for carcinogenesis and other disease processes22 .

The findings on the nature of ETS suggest that no firm conclusions can be drawn on the possible health effects of ETS without adequate supporting evidence from clinical, experimental and epidemiological studies.

A listing of ETS epidemiological studies

In the tables that follow, there are listings of ETS epidemiological studies concerning lung cancer and ETS, prepared for the TMA by the epidemiologist, P N Lee. With regard to heart disease, studies relating to the work place are listed. Further details relating to the composition of these lists, and also further detailed listings regarding heart disease, are available on the website, www.pnlee.co.uk. The overviews of the findings of those studies given below have been prepared by the TMA.

Lung cancer

There have been over 60 epidemiological studies of lung cancer among life-long non-smokers. The overall evidence shows no statistically significant increased risk of lung cancer in relation to ETS exposure from parents in childhood, or in social situations or to non-spousal ETS exposure at home. The overall evidence shows that lung cancer risk among non-smoking women is associated with having a husband who smokes (and vice versa but an even weaker association). However, this excess risk of well below 2.0 may be accounted for by bias and failure to take account of confounding factors and misclassification. Those studies that reported stronger associations did not adjust for age, a standard procedure to avoid bias. 80% of the studies showed no statistically significant association with smoking by the spouse and lung cancer. The largest five studies (with over 400 lung cancer cases) produced inconsistent results; one reporting a small increase in risk, three no statistically significant increase and one a statistically significant decrease in risk.

Of those studies, around 50 have examined the incidence of lung cancer in women who claim never to have smoked, but who are married to smokers (“spousal” studies), or the nearest equivalent index, such as living with a smoker. Many have reported a small increase in risk, but a significant majority have not reported overall statistically significant increases. Where a statistically significant association was reported, the magnitude of relative risk reported was so small (below 2.0) that it would generally be regarded as being too weak by normally accepted epidemiological standards to form a basis for public health policy23 .

The small increase in risk reported by various studies could be accounted for by a number of factors. For example, non-smokers living with smokers tend to have different lifestyles and diets from those living in non-smoking households. It is also not possible to be certain that all studies made appropriate adjustments for misclassification – such as when self-reporting non-smokers are in fact former or current smokers. This is especially problematic because former and current smokers not only have an increased risk of lung cancer, they are also more likely to have married smokers and thus be included among those exposed to ETS in these studies.

The data on ETS exposure at work is even less conclusive than the spousal data. Only a very small minority of the studies on ETS and lung cancer have reported an overall statistically significant increase in risk. Similarly, most studies which have looked at ETS exposure in social settings and during childhood do not report an overall statistically significant increase in risk of lung cancer.

Coronary heart disease

There have been around 30 studies of heart disease and ETS among life-long non-smokers. The overall evidence does not indicate an increased risk of heart disease due to ETS exposure in the work place. Only one study out of 18 reported a statistically significant association. Again the weak associations found between spousal smoking are generally not statistically significant and could be accounted for by lifestyle confounding factors – of which over three hundred have been reported – study design, absence of confirmation of diagnosis, and misclassification. Two of the most substantial pools of data on this subject are the databases of the American Cancer Society’s Cancer Prevention Study and the database of the US National Mortality Followback Survey. Analyses of these data sets have reported no overall association between ETS and heart disease24 .

A further large study of ETS and heart disease was published in 200325 and also showed no increase in risk.

A report of the US Surgeon General26 noted “because smoking is but one of the many risk factors in the aetiology of heart disease, quantifying the precise relationship between ETS and this disease is difficult”.

Children

There is a large body of research on ETS exposure and respiratory disorders in children. These are difficult to analyse overall as there is great disparity in study design, age ranges and subjects, the symptoms measured and methods of diagnosis. There are quite a number of reports of statistically significantly increased risk of respiratory disorders in pre-school age children exposed to ETS. It is unclear to what extent this increase is influenced by other factors more statistically common in smoking households, such as poor diet, housing conditions and quality of pre-natal care. The pattern of increased risk is not consistently replicated for children of school age, suggesting that a real effect, if one exists, is short term and is age-related.

Although smoking by parents has been associated in some studies with an increased risk of “cot death” (sudden infant death syndrome), a long list of other factors has also been reported27 . Some recent studies have reported that incidence of ‘cot death’ has been reduced by up to 50% where parents have followed government advice not to put their children to sleep in a prone position. However, no one yet fully understands the reasons or mechanisms behind this syndrome. Some have suggested that there may be some residual effects of a mother’s smoking during pregnancy, in respect of which there is strong public health advice to women not to smoke during pregnancy.

1 Fourth Report of the Independent Scientific Committee on Smoking and Health, 1988, para 69, citing IARC Monograph on the evaluation of the carcinogenic risk of chemicals to humans, Tobacco Smoking, 1985; 38: 214, 230-232

2 Presented at the Royal College of Physicians’ conference on 17 th May 2004.

3 Woodward A & Laugesen M, How many deaths are caused by second hand cigarette smoke, Tobacco Control, 2001;10: 383-388

4 Evaluation of the potential carcinogenicity of electromagnetic fields, US EPA Review Draft, p 6-2, October 1990

5 National Cancer Institute, USA, Press Release, 26 October 1994

6 Doll, R and Peto, R, The causes of cancer, p 1219, OUP 1981 ,

7 Peto, J, Meta-analysis of epidemiological studies of carcinogenesis, Mechanisms of Carcinogenesis in Risk Identification, ed Vainio H, p573, IARC 1992

8 Hackshaw A K, Law M R, Wald N J. The accumulated evidence on lung cancer and environmental tobacco smoke. BMJ 1997; 315:980-988

9 Boffetta P et al. Multi-centre case-control study of exposure to environmental tobacco smoke and lung cancer in Europe, Journal of the National Cancer Institute 1998;90: 1440-1450

10 Isabel dos Santos Silva. Cancer epidemiology: Principles and methods, IARC 1999

11 Linet M, et al. Residential exposure to magnetic fields and acute lymphoblast leukaemia in children. New England Journal of Medicine;1997:337:1

12 Nyren O et al. Risk of connective tissue disease and related disorders among women with breast implants: a nationwide retrospective cohort study in Sweden. BMJ 1998;316:417

13 Cooper C, Dennison E. Do silicone breast implants cause connective tissue disorder? BMJ;1998;316:403

14 Baroness Jay of Paddington, Minister for Public Health, House of Lords, Written Answer, Official Report, 31 March 1998, Cols. 31-32 .

15 Hill A B. The environment and disease: Association or causation. Proc.R S Doc. Med. 1965;58: 295-300

16 Baker R and Proctor C ,The origins and properties of environmental tobacco smoke, Environmental International 1990;16: 231,245

17 Proctor C. The analysis of the contribution of ETS to indoor air. Environmental Technology Letters, 1998; 9: 553-562

18 Guerin M et al, The chemistry of environmental tobacco smoke: composition and monitoring. Chelsea, Michigan, Lewis Publishers, 1992

19 Gori G B and Mantel N, Mainstream and environmental tobacco smoke. Regulatory Pharmacology and Toxicology 1991;14: 88-105

20 Jenkins R A et al. Determination of personal exposure of non-smokers to environmental tobacco smoke in the United States. Lung Cancer 1996;14;1: Supplement p195

21 Proctor C et al. A comparison of methods of assessing exposure to environmental tobacco smoke in non-smoking British women. Environmental International 1991;17.4: 287-297

22 Kraus N, Malmfors T and Slovic P. Intuitive toxicology: expert and lay judgements of chemical risks. Risk Analysis 1992;12: 215-232

23 Dirty Water. US National Cancer Institute, Reason 1996;28:1.52

24 LeVois M and Layard M. Publication bias in the environmental tobacco smoke/coronary heart disease epidemiologic literature, Regulatory Toxicology and Pharmacology 1995;21:184-191

25 Enstrom J & Kabat G. Environmental Tobacco Smoke and tobacco related mortality in a prospective study in California, 1960-1998, BMJ 2003;326: 1057-1061

26 Reducing tobacco use, a report of the US Surgeon General, US Dept. of Health and Human Services, Public Health Service, Office of Smoking and Health

27 Thornton A J & Lee P N . Parental smoking and sudden infant death syndrome: A review of the evidence. Indoor Built Environment 1998;7:87-97