Summary of findings fromThorax reviews

Tables 1 and 2 summarise the findings of theThorax series. Overall there is a very consistent picture with odds ratios for respiratory illnesses and symptoms and middle ear disease of between 1.2 and 1.6 for either parent smoking, the odds usually being higher in pre-school than school aged children and higher for maternal smoking than for paternal smoking. However, for lower respiratory illness in infancy and for wheeze and cough in schoolchildren the effect of paternal smoking in households where the mother did not smoke was statistically significant. This latter observation suggests that much of the observed association with maternal smoking is probably due to postnatal rather than prenatal (intrauterine) exposure. Because smoking by the mother during pregnancy is almost invariably associated with postnatal smoking, any additional influence of prenatal maternal smoking will be difficult to resolve using epidemiological studies. Except for sudden infant death syndrome (SIDS), the risks associated with parental smoking were largely independent of measured confounding variables, which suggests that residual confounding by unmeasured factors is unlikely to be important.

In June 1998 we re-ran our original search strategy to identify publications since April 1997; this identified 29 articles containing data not included in the original reviews. These are commented on separately in the relevant sections, but the quantitative meta-analyses have not been updated.

Below we summarise the papers published since our original reviews and consider whether any changes in our conclusions are warranted.

Comparison with Californian EPA review

Table 3 contrasts the methods used in ourThorax reviews and those of the Californian Environmental Protection Agency9 ,40 and table 4summarises the conclusions of the Californian review. Despite the different approach the conclusions are qualitatively and, from a public health perspective, very similar. The main differences are a difference in interpretation of the inconsistent data on allergic sensitisation (we hold by our view that allergic sensitisation is not related to in utero or ETS exposure) and greater emphasis in the Californian review on the relationship between exposure to ETS and the incidence of asthma. This arises because the Californian review includes prevalence studies in its assessment of incidence, and also because there is no clear distinction between the incidence of lower respiratory infections and wheezing illness in infancy and the development of later onset asthma.

Value of a systematic quantitative approach

At the end of this series it is worth considering the value of the approach we took to reviewing the evidence. Meta-analyses of observational studies raise a number of difficulties compared with randomised controlled trials.45 Indeed, some have argued that “the meta-analysis of published non-experimental data should be abandoned”.46 Shapiro argues that meta-analysis is popular because it offers the Holy Grail of attaining statistically stable estimates for effects of low magnitude. This is dangerous, he argues, because, where many studies produce only modest increases in risk, those increases may be due to the same biases in all the studies. In our own case only the twofold increase in the risk of SIDS and possibly lower respiratory infection in infancy are large enough to make confounding unlikely when the relative risk is considered in isolation. However, the approach we adopted of comparing unadjusted and adjusted relative risks in each study permits a more comprehensive evaluation of confounding effects. Although this does not address the possibility of a bias common to all studies, the latter is unlikely if there is consistency of evidence from studies of different design and locations. While such dangers exist, it would seem even more dangerous to rely on a single large study or on narrative reviews. In our view the presentation of all studies on a single graph is an extremely valuable summary of the evidence, even where heterogeneity in effects is so large as to render meta-analysis irrelevant. It is a separate argument to decide whether residual confounding in all studies may explain the findings. Undoubtedly the definitive demonstration of cause and effect requires randomised trials, a point we pick up below.

In our view the major advantage of a systematic quantitative approach is that it has produced a useful corrective to a narrative approach which gives undue weight to highly valued and well published studies. This was particularly valuable in assessing the effect of maternal smoking on allergy. The major disadvantages were the large amount of work needed to extract comparable data and the need to reduce analysis to lowest common denominators. Here the main issue was the variety of ways in which exposure was assessed. Nevertheless, it was possible in most instances to compare maternal with paternal/other household smoking and to gain some insight into dose-response. The effect of adjustment for confounding variables was unimportant except for outcomes other than SIDS. Given the variety of confounders adjusted for and methods of adjustment used, this was fortunate.

Whenever systematic reviews of trials are carried out the quality of studies should be assessed and the sensitivity to inclusion of poor studies investigated. In our reviews the lack of easily agreeable criteria meant we included all studies where possible. For example, response rates were often lowest in some of the “best” studies because they were longitudinal. Equally it was often difficult to extract necessary information from some of the most influential studies, particularly those published early on. Fortunately the consistency of the evidence meant that estimates of effect were little altered by exclusion of specific studies.

It is important not to give undue emphasis to point estimates without considering the consistency and heterogeneity of results lying behind them. For this reason we believe that the figures presenting all studies as well as pooled estimates should be available. This allows distinction of lack of a consistent pattern from statistical heterogeneity where a consistent direction of effect is seen. In our reviews a consistent pattern was seen except for allergic sensitisation. In most instances fixed and random effects estimates were very similar, usually with random slightly greater, while confidence limits for random effects were wider. This reflects the greater emphasis placed on smaller studies by a random effects approach. It is therefore more susceptible to publication bias. In practice, although we were able to detect publication bias for respiratory symptoms and for effects on FEV₁ (in both instances small studies tended to show larger effects), the overall picture and estimates were little altered by excluding small studies.3 ,8

Mechanisms

Evidence relating to mechanisms could potentially assist in interpretation of the epidemiological data we have reviewed. Table 5summarises the potential mechanisms whereby maternal smoking during pregnancy or exposure to ETS postnatally might influence respiratory disease in children. However, while most of these mechanisms are plausible, remarkably little evidence exists to confirm or refute them.9

The most direct evidence on mechanisms is from acute effects on upper respiratory mucosa9 but, apart from middle ear effusion, this is least relevant to the outcomes we have considered. The most convincing epidemiological evidence relates to early lower respiratory infection in relation to postnatal exposure, yet we are lacking insights into how ETS increases the severity of these early (largely viral) infections.

An early hypothesis was that smoking parents, being more susceptible to respiratory infections themselves, might then transmit them to their children. Thus Colley in two early papers on parental smoking and respiratory symptoms looked at the effect of adjusting for parental phlegm production.47 ,48 While adjustment did not adequately explain the higher prevalence rates in children of smoking parents, this hypothesis deserves further consideration in relation to viral infections.

Studies in children which have assessed the effects of acute exposure to ETS in controlled situations are very limited, but there are weak suggestions of acute effects of ETS exposure on lung function.7 The more extensive evidence in adults has recently been reviewed.9 ,49 Coultas49reported that “most of the ETS inhalation chamber studies show slight to moderate transient effects on lung function in at least some of the study subjects. In several studies participants experienced decrements in lung function exceeding 20%.” Such acute effects might well explain the greater peak flow variability in children of smoking parents.7 Further studies to confirm these findings in children seem warranted.

The limited evidence relating ETS to bronchial inflammation and airway development is only by extrapolation from active smokers or from sidestream exposure of laboratory animals. Our review has effectively excluded allergic sensitisation as a link between ETS and asthma and casts some doubt on the BHR route. Evidence of acute effects on BHR in chamber studies in adults is limited and not consistent.9

Outstanding research issues

While the accumulated evidence for adverse effects of parental smoking on the respiratory health of children is very strong, it is based almost entirely on observational studies. There is no clear demonstration of the effect of reducing exposure. Such studies are needed, either in the form of randomised controlled trials or as observational studies focusing on parents who change their smoking habit.

While randomised controlled trials are the ideal, they would need to be large. Consider a study in which it was proposed to reduce smoking in parents of children with middle ear effusion with the outcome of interest being operative treatment. Middle ear effusion commonly resolves in about one third of cases between outpatient referral and operative treatment some 3–6 months later. We might expect perhaps 10% of parents to stop smoking with usual care and might hope to double this to 20% in the intervention group. Assuming one third of cases resolved spontaneously in children of smokers and an optimistic one half in children of those who quit smoking, we would need to randomise 33 500 children overall (16 750 to each group) to have 90% power at the 0.05 significance level. This is because the majority of parents in each group continue to smoke. The difference in outcome between the intervention and usual care groups is therefore small and the trial needs to be large to detect such a difference. Such considerations explain why there have been so few trials and those that have been carried out have been negative when analysed on an intention to treat basis.

It seems unlikely that many randomised controlled trials will take place. Nevertheless, observational studies looking at changes in health outcome in relation to changes in exposure would be valuable. For example, it would be possible to compare the outcomes in the children of the usual care group comparing the 90% whose parents continued to smoke with the children of the 10% whose parents quit. For such an analysis a sample of only 1000 would suffice.

Further cross sectional studies of lung function or symptoms are unlikely to be informative unless they can compare critical periods of exposure or look at changes in parental smoking—for example, school age versus exposure during pregnancy or early infancy for symptoms, or prenatal versus postnatal exposure for SIDS.

Future studies need to give thought to the assessment of exposure. Key issues are distinguishing between maternal and paternal smoking and looking for dose-response. Objective measures such as cotinine levels are important since actual exposure will vary between individuals and tend to decrease with age despite parental smoking habits being constant. It is also important to consider whether children from non-smoking families are a suitable group to treat as non-exposed. Any background exposure in this group which has an effect on respiratory disease will bias any comparisons between smoking and non-smoking families towards the null hypothesis of no difference. Measurement of cotinine levels will help here. The limited evidence available is not entirely consistent. Studies in British children suggest that the low levels of exposure seen in non-smoking households50 do not influence either lung function or respiratory symptoms.51 ,52 In contrast, an Italian study has reported effects on lung function in children with occasional exposure to ETS.53 However, the cotinine levels reported in this study are extremely high, even for children from non-smoking households.

Further studies distinguishing current versus cumulative versus early (particularly in utero) exposure in relation to symptoms and lung function would help to elucidate the mechanism and inform preventive measures. While maternal smoking is the most important source of exposure in many countries, it would be valuable to see further large studies measuring dose where mothers are confirmed non-smokers. The studies from China54-57 have been particularly useful in this context. In the case-control studies of SIDS it would be important to confirm non-smoking status of mothers at interview by measurement of salivary cotinine levels. Such an approach would remove worries about reporting bias.

There is undoubtedly a need to clarify the association between exposure to ETS and BHR—here the solution would be to pool data from all studies published and unpublished. Equally, with only four published studies of peak flow variability there is undoubtedly room for publication bias. However, if there is an acute effect on lung function from ETS exposure in at least a significant minority of subjects this will be better shown by laboratory studies of acute exposure.

Finally, it would be useful to have larger and more comprehensive studies of children with cystic fibrosis. In particular, there is a need for studies of prognosis and severity.

The magnitude of the problem: attributable risks

Throughout our review series we focused on odds ratios as measures of effect since these are what studies provide and they are portable in that studies from different countries produce similar estimates. However, it is important to consider the potential size of the public health problem in any given country and the difference in absolute terms of the different health effects should be recognised. Previous reports have done this for the USA58 and for California.9 Such analyses involve a number of assumptions which may be split into two stages. At the first stage the percentage of cases in a population attributable to an exposure (PAR%) can be derived from knowledge of the relative risk (RR) and the proportion of subjects exposed in a population (p):

The PAR% for a range of relative risks and exposure prevalence rates is given in table 6. For our purposes we assume that the odds ratios in table 1 are equivalent to relative risks. Where the number of events is known it is then straightforward to move to the second stage to work out the number of cases attributable to the exposure.

Thus, for SIDS, assuming that 25% of women smoke during pregnancy and that this raises the risk of sudden infant death by 2, 20% of deaths from SIDS may be attributed to exposure to ETS (table 6). In England and Wales there are approximately 400 SIDS deaths per year59 and thus 80 deaths are attributable to maternal smoking. This may be an underestimate as rather more women smoke postnatally and it takes no account of smoking by other household members. One recent study estimated that 63% of sudden infant deaths were attributable to parental smoking,60 the high percentage arising from a combination of a high relative risk estimate in that study along with an assumption that paternal smoking had a marked effect, thereby increasing the exposure prevalence.

In contrast to sudden infant death which is a fatal but rare condition, lower respiratory infection in infancy, respiratory symptoms in older children, and middle ear disease are much more common but it is difficult to estimate the prevalence of them with any precision. While the number of attributable cases is not easily quantified, even a small relative excess implies many thousands of extra children affected by each of these conditions. The effects range from the very minor to major, and from acute to chronic. Thus, assuming a relative risk of 1.3 and a prevalence exposure of 35%, about 10% of surgical operations for glue ear are attributable to the effects of parental smoking. Given the reported 60 000 operations per year in England,61this amounts to an extra 6000 ear operations per year. The number of attributable episodes of glue ear will be far greater.

Public health issues

That exposure to cigarette smoke after childbirth, rather than solely during pregnancy, increases the risk of a range of respiratory problems in infancy as well as later in childhood, appears to alter the agenda. It broadens the problem from maternal smoking to that of family and friends, and hence policy about smoking on public transport, in restaurants, and other public places becomes an important issue. On the other hand, postnatal exposure should be easier to modify as it is theoretically feasible to keep the infant physically apart from the smoker. In practice this is difficult in pre-school children where the mother is a smoker. In particular, it would be wrong to lose sight of the fact that the major part of ETS exposure occurs within the home and that maternal smoking remains the major source in many countries. It seems likely that prevention will remain focused on reducing the percentage of parents who smoke rather than on isolating smokers or increasing ventilation.

ETS pollution is increasingly being tackled in western countries by health promotion campaigns and restrictive interventions—for example, in the workplace. However, few campaigns outside the USA have highlighted the susceptibility of children to ETS exposure. The challenge is to get the message about smoking and health risks in infancy across without making the first six postpartum months even more difficult. While in developed countries “back to sleep” campaigns have successfully altered the sleeping position of babies, smoking rates have been left virtually unchanged. In many undeveloped countries few women smoke, while male smoking rates are very high. It is likely to prove difficult to promote household changes in these groups if education is channelled through mothers because of marked imbalances of power within the family.

It is also important to view the adverse health effects from ETS exposure in context. The effects are small relative to effects of active smoking, but for children ETS exposure is not voluntary. Potentially, the link between parental smoking and uptake of active smoking by their children is of greater long term importance to a child’s health. In England parental smoking doubles the risk of smoking uptake by children62 which, since 50% of children come from smoking households, allows us to estimate that up to one third of children who smoke can be attributed to parental example. For this reason, reduction of parental smoking throughout a child’s upbringing, rather than just in the perinatal period, may pay substantial future dividends in the prevention of respiratory diseases.