Smoking represents a critical international issue for public health policy makers and strategists. According to the World Health Organisation, tobacco is the second major cause of death and the fourth most common risk factor for disease, worldwide. If current trends continue it will be causing around 10 million deaths each year by 2020, with approximately 650 million fatalities overall 1. Smoking also represents a key issue in the medical profession, as physicians play a leading role in tobacco usage prevention in the community 2, and a key position in the development of overall public health policy. Medical professionals are on the frontlines of primary health care, and research has shown that medical interventions can be effective in helping patients to quit smoking 3. In this role, physicians are widely viewed as exemplars by the community, their patients and their colleagues. Indeed, the physicians' office and hospital should be a model of non-smoking behaviour 4, and, as early as 1976, it was suggested that physicians could best persuade patients to quit if they themselves did not smoke 5.

Aside from its significant impact on patients' health, tobacco usage also represents an important occupational health issue in the medical profession. According to the International Labour Office (ILO), the promotion of smoke-free environments forms a key part of any healthy and safe workplace 6. Interestingly, some of the first epidemiological research demonstrating the adverse health effects of tobacco smoking was actually conducted among a cohort of British physicians 7. So important was Doll and Hill's 1954 study of British doctors that it was republished by the British Medical Journal 50 years later 8 and remains a milestone in public health to this day 91011. Further research from the United States (US) also supported the preliminary British findings with regard to smoking hazards 121314.

Although the dangers of smoking are now well-known throughout the medical profession, physicians have not always set a good example for patients 15. In the 20^th century for example, some physicians even advertised cigarettes 1617. Smoking rates among them were also quite high. Some of the earliest large-scale epidemiological research from the United States revealed that around 40% of physicians were smokers in 1959 5, a figure which had fallen to 21% by the mid 1970s 1819. By the mid 1980s, around 17% of US physicians were still smoking cigarettes and 8% smoking pipes or cigars 20. A large prospective study undertaken by the American Cancer Society in 1982 revealed a smoking prevalence rate of around 25% among physicians 21. Subsequent National Health Interview Surveys found that the national smoking rate for physicians in the US had fallen dramatically between 1987 and 1994, and was below 10% by the mid 1990s 222324. Similar downward trends were also seen in Scandinavia 25 and the Netherlands 2627 during the latter half of last century.

Although these investigations suggest that physicians' smoking rates are probably declining in many parts of the world, international trends have not been clearly elucidated. Furthermore, few, if any, researchers have comprehensively reviewed tobacco-smoking habits in the medical profession from a global perspective. This represents an unfortunate oversight in the literature, as data on physicians' smoking habits has at least two direct benefits for health care policy makers 15. Firstly, it can help predict how effective any potential anti-smoking campaigns in the wider community might actually be. Secondly, and perhaps most importantly, contemporary tobacco usage data allows public health policy makers to determine how soon their overall community prevalence rate might decline 15. Given such clear benefits, the aim of our investigation therefore, was to undertake a comprehensive review of international tobacco smoking surveys, which have been conducted among physicians over the past 30 years. The main research question was, what proportion of physicians are smoking in what countries, and how have their habits changed over time.

This study was conducted as an extensive international review targeting all manuscripts published in peer-reviewed journals relating to the topic of tobacco smoking among physicians. No unpublished articles were included. As the nature of research changes over time and results quickly go out of date, only manuscripts which had been published in the previous 30 years were included. As there is always some delay between conducting a study and actually having it published, the most recent investigations on this topic had been conducted in 2004, and thus the search criteria were limited to articles published between 1974 and 2004. For consistency, only English-language reports were included in the review. The literature review began in March 2006 with a Medline and CINAHL (Cumulative Index to Nursing and Allied Health Literature) internet search using the most appropriate Medical Subject Headings (MeSH) 'smoking', 'tobacco' and 'physician'. After identifying some initial studies, the search was repeated using the additional keyword variations of 'smoke' and 'doctor'. Manuscripts located using these initial criteria were subsequently examined to find additional publications in their reference list. A large proportion of manuscripts were eventually located using the latter method. Manuscripts were arranged in descending order, depending on the year in which the survey was undertaken, rather than the publication year. Where such information was unavailable, the corresponding author of the manuscript was contacted for clarification. In cases where contact with the authors was not possible or repeatedly unsuccessful, manuscripts were listed by year of publication and marked with an asterisk. If a study had been undertaken over the course of more than one year, then the most recent year was listed.

Manuscripts were assigned a reference number based on the abovementioned criteria. As the results of some investigations were published in more than one journal article, some studies have two to three corresponding references. Smoking rates were listed as the prevalence of smoking among the entire group, and also as prevalence rates for males and females. In cases where smoking prevalence rates by gender were not stated in the manuscript itself, they were manually calculated whenever possible. For consistency, all smoking prevalence rates were rounded to the nearest whole number. Response rates for each study were also rounded to the nearest whole number for standardisation purposes. As some studies investigated multiple occupational groups that included physicians, some response rates were indicative of the entire group response, rather than just the physicians. Where authors had apparently used a convenience sample with an unspecified response rate, or the response rate was not listed, this information was also indicated on the table.

A total of 81 published studies met the inclusion criteria 282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113, as shown in Table 1. Most (n = 48) had been conducted as postal surveys, 14 were hand delivered, 4 were telephone surveys, 4 had utilised census data, 3 were conducted by personal interview, and two had been conducted at conferences. A further 4 used both postal surveys and follow-up telephone interviews, while two used postal and hand delivery, mainly to increase response rates following the postal phase. The latter technique appears to have been quite successful in some cases, with one Malaysian study 76 achieving a 100% response rate in this regard. Response rates of the published studies ranged from 27% 101102 to 100% 76, with most above 60%. Only four manuscripts had response rates below 50%, and three investigations did not list their response rate. One study from Iran appeared to have 100% participation 39, although a response rate was not clearly stated and the authors were unable to be contacted to clarify their result, despite repeated attempts. A similar situation was encountered with a Greek study 70, where no response rate was listed and the authors were unable to be contacted. Among all manuscripts included in the current review, sample sizes ranged from 45 29 to 10 807 77, with an encouraging proportion having over 1000 respondents. Particularly large surveys of physicians' tobacco smoking habits were published from the Doctor's Health Study in the United Kingdom 77, the Physicians Health Study in the United States 666768, and also from New Zealand census data 54106111; one of the few countries in the world which includes tobacco smoking questions on their census form 54.

One confounding factor across many investigations however, was a lack of standardisation regarding the definition of 'current smoker'. Although most authors referred to their subjects as simply being either smokers or non-smokers, some used recall periods ranging from one week to one month in their definition of the term 'current'. Others listed no recall period. This may have arisen due to the inherent difficulties in assessing smoking habits over time, and the fact that most investigations simply described the point-prevalence of tobacco smoking among the surveyed group. Not all physicians smoked cigarettes either, with a study of Hispanic physicians in the United States 29 finding that 7% of their subjects smoked cigars, and none smoked cigarettes. In 1990 Doll et al 77 also revealed that a large proportion of their British physicians only smoked pipes or cigars, similar to Fowler et al's 89 earlier finding in the same country. Another confounding factor was that some studies appeared to use convenience samples, rather than true random sampling. Furthermore, a certain proportion of manuscripts did not clearly describe their sample group or their entire research methodology in detail. Nevertheless, such investigations were in the minority, with a large proportion of all manuscripts located during this review having reasonable sample sizes in the hundreds, and sufficiently high response rates to allow confidence in the published data.

A large proportion of all research on physicians' tobacco smoking (52 of 81 studies) appears to have been conducted since 1990 282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081. Twenty-three studies had been undertaken between 1980 and 1989 828384858687888990919293949596979899100101102103104105106107, with the remaining 6 investigations conducted prior to 1980 108109110111112113. By country, 18 manuscripts in this review originated from the United States, 7 from Japan, 6 from Italy, 5 from the United Kingdom, 5 from Australia, 3 from New Zealand, and the remainder from other areas. When investigated from an international perspective, the overall prevalence of physician's smoking appears to have followed two distinct trends during this time. First of all, most developed countries appear to have experienced a steady decline in physicians' smoking rates over the past 30 years. Since the year 2000 for example, four separate studies 29303642 have shown the prevalence of smoking among American physicians to be lower than 10%.

Three investigations of their Australian counterparts in the 1990s 565758 revealed a prevalence of around 5%, while in New Zealand, analysis of census data also suggested a similar rate during this period 54. Tobacco use among British physicians has been well-studied longitudinally 101177, although their smoking rate does not appear to be as low as the abovementioned countries, possibly due to the relatively large number who continue to smoke pipes and cigars, rather than cigarettes. Nevertheless, overall tobacco consumption has still declined markedly, with Doll et al 77 revealing that the absolute proportion of British physicians who smoked cigars, pipes or cigarettes fell from 62% to 18% between 1951 and 1990. As the British Doctor's Study follows the same cohort longitudinally, this represents one of the clearest reductions in absolute tobacco smoking rates among physicians.

According to our review, such trends may not be uniform across all countries however, with physicians in some developed regions still smoking at fairly high rates. Multiple investigations from Italy 404853, Japan 41516278838485100 and France 4963 for example, have consistently documented smoking prevalence rates over 25%. A second trend is also evident in some newly developing countries, where contemporary physicians appear to be smoking at high rates in China 55, Estonia 3132, Bosnia/Herzegovina 33 and Turkey 34. In China for example, Li et al 55 reported that tobacco smoking rates among physicians have actually been increasing in recent years. A surprisingly low rate was found in Nigeria however (3%) 35, suggesting that exceptions are still possible in this latter group. The lowest overall smoking rate was documented in the United States (2%) 4280, with similar low rates also being demonstrated in Australia (3%) 56 and the United Kingdom (3%) 93. The highest smoking prevalence rate was recorded in Greece 70, where roughly half of all physicians (49%) reported themselves to be current smokers. Almost half of all Chinese (45%) 55 and Japanese physicians (43%) 100 were revealed to be current smokers in two separate studies. Similar results were also documented in Kuwait (38%) and the United Arab Emirates (36%) 73, particularly among males (among whom 45% and 44% smoked, respectively). Almost half (48%) of all male Indian physicians from one study 96 were smoking. The proportion of ex-smoking physicians is also worth considering, with prevalence rates of 8% in Australia 57, 17% in the United States 42, 23% in Wales 91 and 52% in Canada 72, being previously documented.

For current smokers by gender, the highest smoking prevalence rates were 61% among male physicians in China 55 and 34% among female physicians in Italy 53. Two investigations from France also found that one-quarter of their female physicians smoked tobacco on a regular basis 4963. Conversely, other research from China 28, Malaysia 76, Wales 91 and Hong Kong 94 revealed no female smokers at all. This may suggest a cultural reluctance for professional women to smoke in certain regions, such as Asia. An important observation during our review was the relatively large number of studies where male physicians smoked at higher rates than their female counterparts. This finding was not without exception however. In Italy for example, Zanetti et al 53 found that more women doctors smoked than men, while in Israel 52, Australia 58 and the United States 88, smoking prevalence rates were almost the same between the genders. A large proportion of manuscripts did not divide their smoking prevalence rates by sex however, making it impossible to do further gender comparisons. Many authors documented age-related differences in physicians' smoking rates, with older physicians for the most part, more likely to be current smokers. Nevertheless, in China 28, Japan 62100, Mexico 69 and India 96, tobacco usage was actually more prevalent in younger physicians. This latter result suggests that a challenge for public health policy makers in countries with high smoking rates among young physicians, still lies ahead.

Aside from overall prevalence rates, additional useful information was also obtained with regard to physician's tobacco smoking habits. Firstly, some studies simultaneously investigated the tobacco usage habits of dentists, nurses and other hospital staff while surveying doctors. Two investigations from the United States 3061 found that fewer physicians smoked when compared to dentists, while another study demonstrated very similar, albeit very low, smoking rates among the two professional groups 80. In 1979, Wyshak et al 108 found that physicians were less likely to be current smokers than lawyers. Most studies found that fewer physicians smoked when compared to nurses at the same facility, although an investigation from Finland 71 suggested the opposite situation may sometimes occur. Even so, physicians in most societies tend to give up smoking before other occupational groups and the general public for a number of reasons 15. Firstly, they may recognise the negative medical consequences more quickly than the general public. Secondly, their devotion to health naturally conflicts with unhealthy behaviours. Thirdly, tobacco smoking usually incurs a negative image in the health-care profession long before it does so in the wider community 15. In this regard, doctors are well-equipped to evaluate scientific knowledge, and can be reasonably expected to act upon new discoveries, if warranted 114. Furthermore, smoking rates in developed countries tend to decrease over time due to a generational effect, as the social climate of a country changes and more people give up smoking, doctors included 86.

Tobacco smoking by medical speciality also revealed some interesting, though inconsistent, results during this review. One study for example, found that family physicians smoked less than physicians generally 67, while two others suggested that general practitioners smoked more often than specialists 7986. In the Netherlands, more consultants smoked than house officers 82. Trainee psychiatrists 58 and psychiatry residents 92 were the most likely to smoke in some investigations, while in others it was surgeons 70, obstetricians 111 or surgeons and obstetricians 109. Encouragingly, Kawane et al 838485 demonstrated that Japanese chest physicians had a lower smoking rate than Japanese physicians, generally. Exactly how much a physician's medical speciality influences their smoking habits is uncertain. A previous study of Malaysian doctors for example 76 found that around half were already smoking before they entered medical school. Based on the findings of multiple investigations therefore, it is very difficult to ascertain which medical speciality actually has the highest or lowest smoking rate.

Regarding antismoking practices, most physicians in a British study 47 felt they should advise patients to quit, and in France 74 over half the tobacco using physicians had made at least one serious attempt to quit smoking themselves. In Italy however 97, more than half the physicians who currently smoked had made no attempt to quit smoking, and in Japan 64 only 60% of smokers stated any intention of reducing their tobacco consumption or quitting altogether. Other authors have already suggested that Japanese physicians may not be setting a good example in this regard 115. The institutions where doctors work may also play an important role in tobacco control, with an American study 95 demonstrating that a hospital no-smoking policy was useful in helping to reduce the overall smoking rate among staff. Hospitals in the United States were the first industry to declare a national smoking ban in the early 1990s, and ones which later influenced social norms and probably reduced overall smoking rates 116. Even so, the actual hospital in which physicians work, as well as the geographical location where they live, may not always affect the smoking rates of physicians in the same country. One Italian study for example, found different smoking rates by region 40, although in Nigeria 35 smoking rates of physicians in two different hospitals were exactly same, with both being encouragingly low (3%). A doctor's smoking habit may also be associated with the smoking habits of their spouse. Previous studies from Scotland 103 and New Zealand 117 for example, revealed that around half of all male physicians who smoked, also had smoking wives. This situation may reflect assortative mating and its subsequent influence on partner choice, at least with regard to smoking.

How much a physician's personal smoking behaviour affects their professional attitude and clinical behaviour represents a critical issue in public health policy, as physicians are on the frontlines of primary health care. Although medical professionals have many opportunities to reduce the prevalence of smoking among their patients, physicians may have not yet maximized their efforts in meeting the tobacco epidemic. Doctors incur a certain responsibility as exemplars for patients with regard to healthy behaviour 118, as well as the public image they inadvertently portray outside of the work environment 119. Having any physicians who smoke may increase public scepticism, with people inclined to ask why should they stop smoking when their doctor continues to do so? 120. Continued tobacco usage by health care workers undermines the message to smokers that quitting is important 121, and as early as 1976 it was suggested that physicians could best persuade patients to quit if they themselves did not smoke 5. In 1983, Sachs 101 stated that 80% of US citizens expected their physicians to be non-smokers, and in 1984, Wells et al 109 suggested that physicians with good personal health habits counselled their patients significantly more about all health habits. As physicians gain more insight into their own health and health habits, their advice to patients becomes increasingly relevant and effective 122. Although methods for treating tobacco dependence in clinical practice have been described elsewhere 123124125, the entire process need not be overly taxing for physicians. At the most basic level, such interventions may require them to ask only two questions: 'do you smoke?' and 'do you want to quit?' 123. Nevertheless, such guidelines are not always followed for a variety of reasons.

The extent to which the professional practice of physicians is affected by their own smoking habits is very important for policymakers, and has been examined in certain investigations. One of the most marked differences in this regard was found in Greece 70, where only half the smoking physicians were involved in smoking cessation counselling, compared to 100% of their non-smoking colleagues. Several Japanese studies revealed differences in smoking-cessation advice 64 and taking a patients' smoking history 51, with both being significantly more commonplace among non-smoking physicians. Similar findings were also seen in Finland 37, while Pärna et al 31 revealed that Estonian physicians who smoked were reluctant to disturb a patient's privacy by asking about their tobacco usage. Knowledge of smoking-related damage also showed correlations with smoking behaviour in an Italian study 48, although their analysis included other health professionals as well as physicians.

Not all studies revealed differences however. In Israel for example, Samuels 52 asked physicians whether or not they advised patients to stop smoking during consultations, and found no difference between smokers and non-smokers. A longitudinal study of Chinese physicians also revealed that the effects of smoking on counselling behaviour varied between 1987 and 1996 55. In 1987 for example, smoking behaviour was an influential factor, whereas by 1996 it had ceased to be so. Other confounding issues were raised by the Chinese study. Firstly, only one-third of physicians believed that they were the most influential person who could help patients quit. On the other hand, over three-quarters of them believed that physicians can set a good example for patients by not smoking. Most disturbingly, anti-smoking counselling practices appear to be diminishing among Chinese physicians in recent years, while their overall prevalence of smoking is probably increasing during the same time 55. These discrepancies between various countries suggest that not only are physician-targeted smoking interventions urgently needed in public health, but that they should also be culturally specific.

Although our current review sourced a wide variety of manuscripts from many countries, there are a few limitations worth considering. Firstly, for practical reasons, only English language papers were included. With such a methodology, it is possible that we may have missed some manuscripts published in domestic journals and local languages. The decision to restrict our study to English language papers was a purely pragmatic one, however. We felt that formulating inclusion criteria for other languages would have been too difficult to practically achieve, given the wide variety of dialects in which contemporary research is now being published. Furthermore, including some languages but not others would also have led to a bias against manuscripts written in languages that the authors could not understand. Secondly, we assumed that, knowing the topic's importance, significant domestic findings would most likely have been published in an international, English language journal.

Nevertheless, the nature of biomedical publication itself is known to incur some inherent bias against developing countries, and this will probably be reflected in the overall number of publications available from those regions. As previously noted by Rahman and Fukui 126, the imbalance between developing and developed countries in terms of biomedical publication is not only significant, but the total share of publications originating from low-income countries may also be declining in recent years. This limitation in data equity makes it difficult to unequivocally decide why smoking rates differ between physicians in developed and developing countries. Given that such limitations occur, it is important that continued measures be taken to encourage local researchers, irrespective of nationality, to publish their findings in a globally accessible format. Furthermore, it is imperative that the knowledge gained from international smoking research be equally and consistently disseminated to physicians and other public health policy makers in all countries.

A further difficulty encountered during this review was ascertaining exactly how comparable the smoking prevalence data is from country to country, and also from study to study in the same country. As previously mentioned, one issue we encountered early on was the definition of the term 'smoker' that researchers had actually used. Another issue is reflected in the changing nature of survey methodologies, even in the same country, combined with the fact that many surveys are not originally designed for a comparative analysis of smoking rates at a later date; a point acknowledged by certain authors of longitudinal data 43. Probably the best study with longitudinal smoking data is still the British Doctor's Study 77, which has followed the same cohort of physicians for over 50 years. Future researchers investigating the smoking habits of health care workers would do well to follow this pioneering model.

Regardless of what country the data originates from, tobacco smoking by physicians remains a contentious issue in public health. Global policy making demands accurate data on this topic, from which at least two distinct benefits can be derived. Firstly, this kind of data can help predict how effective any potential anti-smoking campaigns may actually be in a particular country 15. That is, it would be difficult to convince the general public not to smoke if their physician role-models continue to do so. Secondly, and perhaps most importantly, it allows public health policy makers to determine how 'mature' a country's smoking epidemic currently is, and thus, how soon the overall community prevalence rate could be expected to decline. The current paper presented an international review of all modern literature concerning tobacco usage patterns within the medical profession. Overall, our review suggests that smoking rates in this professional group vary widely from region to region. International policy makers who are attempting to tackle the tobacco problem on a global scale, such as the WHO, will need to carefully consider these regional differences when devising potential control strategies. In this regard, high smoking rates among doctors in some countries for example, suggests that physicians may not always be the best role models from where sound policy can actually originate.

On the other hand, comparison of physician's smoking rates with other health professionals suggests that fewer doctors seem to be smoking when compared to nurses in the same countries. In addition to physicians, it has also been shown that smoking is now becoming quite rare among dentists in many areas 127. Dentists may therefore be ideally placed to work with physicians in helping to reduce the overall burden of tobacco use among patients. A multidisciplinary effort from all health care workers would seem to be an ideal goal, from a global perspective. Nevertheless, the fact that any health care workers smoke at all is unfortunate, given their undoubted status as public health exemplars. As such, further preventive efforts will need to be focussed on the personal health behaviours of physicians, particularly those in developing countries. To assist in the promotion of sound public health policy, it is important that physician's tobacco usage continues its decline in future years, so that the medical profession can remain at the forefront of anti-smoking programs and lead the way as public health exemplars in the 21^st century.

The author(s) declare that they have no competing interests.

DRS conceived the idea for the paper. DRS and PAL contributed equally to the literature search. Both DRS and PAL contributed to the evaluation of the literature. DRS drafted the manuscript. Both authors reviewed and approved the final manuscript.