The role of healthcare professionals in smoking cessation has been the subject of considerable debate (Chapman 1993). During the late 1980s there was evidence from some randomized trials to suggest that advice from motivated physicians to their smoking patients could be effective in facilitating smoking cessation (Kottke 1988). However, concern was expressed about the low detection rate of smokers by many physicians and the small proportion of smokers who routinely receive advice from their physicians to quit (Dickinson 1989).
From a public health perspective, even if the effectiveness of facilitating smoking cessation by physicians is small, provided large numbers of physicians offer advice the net effect on reducing smoking rates could still be substantial (Chapman 1993). Since that time, there have been numerous attempts to encourage physicians to routinely identify all people who smoke and to provide smoking cessation advice (Fiore 1996; Fiore 2000; Raw 1998; Taylor 1994; West 2000).
The first systematic review on this topic was published two decades ago (Kottke 1988). Since then a number of further studies have examined the effectiveness of medical practitioners in facilitating smoking cessation. Much of this research has occurred amidst a culture in which medical practitioners are playing an increasing role in health education and health promotion, and have an increasing array of options to assist people who want to quit. Doctors now have access to pharmacotherapies that have been shown to increase the chances of success for people making quit attempts, including nicotine replacement therapy (Stead 2008), bupropion (Hughes 2007) and varenicline (Cahill 2007). In some healthcare settings they can also refer patients to more intensive behavioural counselling and support, either face-to-face (Lancaster 2005a; Stead 2005) or via telephone quitline services (Stead 2006).
The primary objective of the review was to determine the effectiveness of advice from medical practitioners in promoting smoking cessation. A secondary objective (added in 1996) was to determine the effectiveness of advice from medical practitioners on reducing smoking-related mortality and morbidity. Our a priori hypotheses were:
- advice from a medical practitioner to stop smoking is more effective than not giving advice.
- the effectiveness of advice from a medical practitioner is greater if the advice is more intensive and includes follow up.
- the supplementation of advice with aids such as self-help manuals is more effective than advice alone.
- motivational advice is more effective than simple advice (added in 2001 update).
The review does not address the incremental effects of adding nicotine replacement therapy or other pharmacotherapies to advice, as these interventions are addressed in separate Cochrane reviews (Cahill 2007; Hughes 2007; Stead 2008). From 2008 it does not address the incremental effect of demonstrating the pathophysiological effect of smoking (e.g. spirometry, expired carbon monoxide), which is covered by a separate Cochrane review (Bize 2005).
Criteria for considering studies for this review
Types of studies
Randomized controlled trials. Trials where allocation to treatment was by a quasi-randomized method were also included, but appropriate sensitivity analysis was used to determine whether their inclusion altered the results. Studies which used historical controls were excluded.
Types of participants
Participants could be smokers of either gender recruited in any setting, the only exception being trials which only recruited pregnant women. These were excluded since they are reviewed elsewhere (Lumley 2004).
Types of interventions
We included trials if they compared physician advice to stop smoking versus no advice (or usual care), or compared differing levels of physician advice to stop smoking. We defined advice as verbal instructions from the physician with a 'stop smoking' message irrespective of whether or not information was provided about the harmful effects of smoking. We excluded studies in which patients were randomized to receive advice versus advice plus some form of nicotine replacement therapy, since these were primarily comparisons of the effectiveness of NRT rather than advice. We excluded studies where advice to stop smoking was included as part of multifactorial lifestyle counselling (e.g. including dietary and exercise advice).
Therapists were physicians, or physicians supported by another healthcare worker. Trials which randomized therapists rather than smokers were included except where the therapists were randomized to receive an educational intervention in smoking cessation advice, since this is the subject of another Cochrane review (Lancaster 2000).
We defined trials where advice was provided (with or without a leaflet) during a single consultation lasting less than 20 minutes plus up to one follow-up visit as minimal intervention. We defined a trial as intensive when the intervention involved a greater time commitment at the initial consultation, the use of additional materials other than a leaflet, or more than one follow-up visit. We considered adjunctive aids to advice as additional strategies other than simple leaflets (e.g. demonstration of expired carbon monoxide or pulmonary function tests, self-help manuals).
Types of outcome measures
The principal outcome used in the review was smoking cessation rather than reduction in withdrawal symptoms, or reduction in amount of cigarettes smoked. Thus we excluded trials that did not provide data on smoking cessation rates. In each study we used the strictest available criteria to define abstinence . That is, we used rates of sustained cessation rather than point prevalence abstinence where possible. Where biochemical validation was used, we classifed only those subjects meeting the biochemical criteria for cessation as abstainers; and where participants were lost to follow up, they were regarded as continuing smokers. We required a minimum follow up of at least six months for inclusion, and used the longest follow up reported. A secondary outcome was the effect of smoking advice on subsequent mortality and morbidity.
Search methods for identification of studies
We identified trials from the Tobacco Addiction Group specialised register. This has been developed from electronic searching of MEDLINE, EMBASE and PsycINFO and the Cochrane Central Register of Controlled Trials (CENTRAL) together with handsearching of specialist journals, conference proceedings and reference lists of previous trials and overviews in smoking cessation. We used the following MeSH terms to identify potentially relevant trials in the register: 'physician-patient-relations' or 'physicians' or 'family-practice' or 'physician's-role'. Trials with the words 'GP' or 'general practice' or 'physician*' in the title or abstract were also checked. For this update of the review the register was searched in September 2007.
Data collection and analysis
In all versions of this review data two people independently extracted data from the published reports. For this update, GB and LS extracted new data. Any disagreements were resolved by referral to a third author. For each trial, we documented the following aspects:
- country of origin.
- study population (including whether studies randomized only selected, motivated volunteers or all smokers, unselected by motivation to quit).
- eligibility criteria.
- nature of the intervention (including the nature, frequency and duration of advice, use of aids, and training of therapist).
- details of study design (including method of allocation, blinding, study structure).
- outcome measures.
- validation of smoking status.
In trials where details of the methodology were unclear or where results were not expressed in a form that allowed extraction of the necessary key data, we wrote to the individual investigators to provide the required information. In trials where patients were lost to follow up they were regarded as being continuing smokers. Reports that only appeared in non-English language journals were examined with the assistance of a translator.
We assessed the methodological quality of the studies included in the review using the scheme described in the Cochrane Handbook which involves assessing the quality of the random allocation (i.e. control of selection bias at entry). This is the only type of bias which has been empirically shown to result in systematic differences in assessment of the effect size (Schulz 1995). A three point rating scale was used, with a grading of: A if the effort to control selection bias had been maximal (e.g. by telephone randomization, or use of consecutively numbered, sealed envelopes); B if there was uncertainty about whether the allocation was adequately concealed (e.g. where the method of randomization was not stated), and C if the allocation was definitely not adequately concealed or was not used at all.
We expressed results as the relative risk (intervention:control) of abstinence from smoking at a given point in time, or for mortality and/or morbidity, together with the 95% confidence intervals for the estimates. This is a change from previous versions of this review, in which results were expressed as an odds ratio. This change takes in to account the fact that most clinicians find the relative risk more straightforward to interpret than the odds ratio.
We estimated pooled treatment effects using the Mantel-Haenszel fixed-effect method. We now use the I
Studies that used cluster randomization (with the physician or practice as the unit of allocation) were included in the meta-analyses using the patient level data, but we assessed the effect on the results of excluding them. Where reported, we have recorded the statistical methods used in studies to investigate or compensate for clustering.
Description of studies
We include forty-one trials, published between 1972 and 2007 and including more than 31,000 participants. Twenty-six trials with 22,000 participants contributed to the primary comparison between advice and a no-advice or usual care control.
Seventeen studies compared a minimal advice intervention with a control intervention in which advice was not routinely offered. Eleven studies compared an intervention that we classified as intensive with a control. Fourteen studies (thirteen of which did not have a non-advice control group) compared an intensive with a minimal intervention, and one study compared two intensive interventions (Gilbert 1992). One study compared an intervention based on the 4As model (Ask, Advise, Assist, Arrange follow up), delivered in two different styles (Williams 2001). Some studies tested variations in interventions and contributed to more than one comparison. These are described and the meta-analyses to which they contribute are identified in the Table 'Characteristics of included studies'.
The definition of what constituted 'advice' varied considerably. In one study (Slama 1995) patients were asked whether they smoked, and were given a leaflet if they wanted to stop. The control group were not asked about their smoking status until follow up. In all other studies the advice included a verbal 'stop smoking' message. This verbal advice was supplemented by provision of some sort of printed 'stop smoking' material (27 studies), or additional advice from a support health worker or referral to a cessation clinic or both. Four studies described the physician intervention as behavioural counselling with a stop smoking aim. One study compared motivational consulting (based on information from theoretical models) with simple advice (Butler 1999). In two studies the smoker was encourage to make a signed contract to quit (BTS 1990A; BTS 1990B). One study provided an incentive (a telephone card) to those who successfully quit (Higashi 1995). Three studies included an intervention which involved a demonstration of the participant's pulmonary function (Li 1984; Richmond 1986; Segnan 1991), or expired air carbon monoxide (Jamrozik 1984). One study, using a cluster design, compared information and a letter alone to advice from a paediatrician to mothers of babies attending well-baby clinics with a view to reducing exposure of the children to passive smoke (Wall 1995). One study (Unrod 2007) used a computer-generated tailored report to assist with cessation, and a further recent study (Meyer 2008) compared brief advice to the use of computer-generated tailored or no intervention.
In the analysis we aggregated groups allocated to brief advice alone with those allocated to brief advice plus brief printed material. We did this with the view that advice plus provision of printed material is a practical approach in the primary care setting. In the two studies which directly compared the additional benefit of offering printed material none was observed (Jamrozik 1984; Russell 1979). Studies which provided a smoking cessation 'manual' were classified as offering an intensive intervention, but there is only weak evidence that self-help materials have a small benefit when combined with face-to-face support (Lancaster 2005b). The intensive intervention subgroup also included studies that offered additional visits.
The follow-up periods during the trials varied considerably, with a tendency towards shorter follow-up periods amongst the older studies. Definitions of abstinence were variable and frequently not stated.
Risk of bias in included studies
Randomization and Allocation
Of the 41 included trials, 10 that allocated small numbers of clusters of people to interventions and their procedures for randomization and allocation concealment are considered separately below. In the 31 other trials, there was typically little information about the way in which the randomization schedule had been generated. Only eight (25%) were rated A for having provided information about use of sealed envelopes (or some more secure method) to conceal the allocation sequence and minimize selection bias at entry, and none used an independent secure centralized randomization process. Only one of these A-rated studies contributed to the primary comparison. Ten studies (32%) used methods open to bias such as allocation by day of attendance or birthdate. In the remaining 13 (42%) individually randomized trials, insufficient information was provided on the method of randomization and allocation concealment.
Ten trials (Haug 1994; Hilberink 2005; Janz 1987; Lang 2000; Meyer 2008; Morgan 1996; Unrod 2007; Russell 1983; Wall 1995; Wilson 1990) had as the unit of allocation the physician, practice or clinic or week of attendance, rather than the individual smoker. In some of these it was unclear whether or not bias in the identification and recruitment of the individual smokers could be avoided. Some studies reported post-randomization dropouts of clinics or physicians. In one study (Meyer 2008) each practice provided each of the three treatment conditions for a week, in the same order with a gap between recruitment periods. We note in the Included Studies table where authors had allowed for or ruled out an effect of clustering, and we used sensitivity analyses to test the contribution of the cluster-randomized trials to the meta-analysis.
As required by the inclusion criteria, all trials assessed smoking status at least six months after the start of the intervention. Twenty-nine of the 41 studies (69%) had a longer follow-up period, typically one year, the longest being three years. Since the interventions generally did not require a quit date to be set, the definitions of cessation used are less strict than are typically found in trials of pharmacotherapies. About half the studies defined the cessation outcome as the point prevalence of abstinence at the longest follow up, and the other half reported sustained abstinence, which typically required abstinence at an intermediate follow-up point as well. Validation of all self-reported cessation by biochemical analysis of body fluids or measurement of expired carbon monoxide was reported in ten studies (24%) (Ardron 1988; BTS 1990A; BTS 1990B; Gilbert 1992; Li 1984; Marshall 1985; Segnan 1991; Slama 1990; Vetter 1990; Williams 2001), but only three of these contributed to the primary analysis. Validation in a sample of quitters was reported in three (Russell 1979; Russell 1983; Unrod 2007). One study used biochemical validation at 12 months but not at 18 month follow up (Haug 1994), and one study used biochemical validation or confirmation by a relative/friend (Richmond 1986). One study adjusted rates based on the deception rate found in a subsample where validation was performed (Fagerstrom 1984). No biochemical validation was used in the remaining 25 studies (61%).
Effects of interventions
Advice versus no advice
When all 17 trials of brief advice (as part of a minimal intervention) versus no advice (or usual care) were pooled (Comparison 01.01.01), the results demonstrated a statistically significant increase in quit rates; relative risk (RR) 1.66, 95% confidence interval (CI) 1.42 to 1.94). Heterogeneity was low (I
More intensive versus minimal advice
The direct comparison between intensive and minimal advice in 15 trials (Comparison 02) suggested overall that there was a small but significant advantage of more intensive advice (RR 1.37, 95% CI 1.20 to 1.56), with little evidence of heterogeneity (I²=32%). In the subgroup of 10 trials in populations of smokers not selected as having smoking-related disease, the increased effect of more intensive intervention was small and the confidence interval only narrowly excluded 1 (RR 1.20, 95% CI 1.02 to 1.43). No individual trials in this subgroup showed a significant benefit and there was no evidence of heterogeneity (I
Number of follow-up visits
The direct comparison of the addition of further follow up to a minimal intervention showed a just significant increase in the odds of quitting in the pooled analysis, although none of the five studies individually detected significant differences (RR 1.52, 95% CI 1.08 to 2.14, Comparison 03.01.01). This analysis did not include one study of the effect of follow-up visits (Gilbert 1992), because the control group received more than minimal advice, including two visits to the doctor. In this study, there was no significant difference in biochemically validated cessation rates between the two visit group and a group offered a further four follow-up visits.
Indirect comparison between subgroups of studies suggested that an intervention including follow-up visits had a slightly larger estimated effect compared to no advice than an intervention delivered at a single visit . The RR for cessation when follow up was provided was 2.22 (six studies, 95% CI 1.84 to 2.68, I
Use of additional aids
Indirect comparison between 10 studies in which the intervention incorporated additional aids such as demonstration of expired carbon monoxide levels or pulmonary function tests or provision of self-help manuals and 17 where such aids were not used did not show important differences between subgroups (Comparison 04).
Comparisons between different types of advice
In a single trial of motivational counselling (approximately 10 minutes) compared with brief advice (2 minutes) a significant benefit was not detected, but the point estimate favoured the motivational approach and confidence intervals were wide (Butler 1999, RR 1.97, 95% CI 0.6 to 6.7). Quit rates were low in both groups, but motivational advice appeared to increase the likelihood of making a quit attempt. This study also contributes to the comparison between intensive and minimal advice.
One trial comparing brief advice using an autonomy-supporting style to advice given in a controlling style did not detect a significant difference. Quit rates were high in both groups and the point estimate favoured a controlling style (Williams 2001, RR 0.51, 95% CI 0.19 to 1.32). Both interventions took about 10 minutes and this trial does not contribute to the intensive versus minimal comparison.
One study included a comparison between brief advice and personalised computer-generated tailored letters. At two-year follow up, rates of sustained six month abstinence did not differ significantly (Meyer 2008, RR 0.95, 95% CI 0.64 to 1.41).
Effect of advice on mortality
Only one study (Rose 78-92) has reported the health outcomes of anti-smoking advice as a randomized single factor intervention. At 20-year follow up, in the intervention compared to the control group, total mortality was 7% lower, fatal coronary disease was 13% lower and lung cancer (death plus registrations) was 11% lower. These differences were not statistically significant, reflecting low power and the diluting effects of incomplete compliance with the cessation advice in the intervention group, and a progressive reduction in smoking by men in the control group. After 33 years of follow up differences in rates for most causes of death were not significant but there was a significantly smaller number of deaths from respiratory conditions. The age adjusted hazard ratio was 0.72 (95% CI 0.54 to 0.96).
The results of the meta-analyses in Comparison 01 were not sensitive to exclusion of either trials using cluster randomization or of trials rated as C (inadequate or not used) on their quality of allocation concealment. Only one trial contributing to comparison 01 was rated A (adequate). Comparison 2 results were not sensitive to the exclusion of studies rated C, but the marginally significant effect in the unselected population subgroup was lost if inclusion was restricted to the 4 A-rated studies, or if, as already noted above, the only cluster randomized study (Lang 2000) was excluded.
The results of this review, first published in 1996 and updated in 2008, continue to confirm that brief advice from physicians is effective in promoting smoking cessation. Based on the results of a meta-analysis incorporating 28 trials and over 20,000 participants, a brief advice intervention is likely to increase the quit rate by 1to 3 percentage points. The quit rate in the control groups in the included studies was very variable, ranging from 1% to 14% across the trials in the primary comparison. However the relative effect of the intervention was much less variable, because trials with low control group quit rates generally had low rates with intervention, and vice versa. The general absence of substantial heterogeneity between trials when relative risks are compared makes for reliable estimates of relative effect. However it is more difficult to estimate the absolute effect on quitting, and the number needed to treat. Absolute quit rates will be influenced by motivation of the participants who are recruited or treated, the period of follow up, the way in which abstinence is defined, and whether biochemical confirmation of self-reported abstinence is required. Many of the trials in this review were conducted in the 1970s and 1980s, and did not use the gold standard methods for assessing smoking abstinence that would now be recommended (West 2005). Only a minority of trials used biochemical measures to confirm self reports of abstinence, and although 12 month follow up was common, many trials assessed smoking status at a single follow-up point. This will tend to lead to higher quit rates overall than in trials with biochemical validation and requiring repeated abstinence at or between multiple assessments, but there is not strong evidence that it will lead to bias in the estimates of relative effect. There were too few trials in the primary analysis to test the effect size when including only trials with complete biochemical validation. We did not find that the control group quit rates were any less variable amongst studies with a longer period of follow up and with abstinence sustained at more than one assessment.
If an unassisted quit rate of 2% at 12 months in a population of primary care attenders is assumed, we can use the confidence intervals for the minimal intervention subgroup, 1.42 to 1.94, to estimate a number needed to treat (NNT) of 50-120. If the background rate of quitting was expected to be 3%, then the same effect size estimate would translate to an NNT of 35-80. Using the pooled estimate from combining both intensity subgroups in the primary comparison would raise the lower confidence interval and reduce the upper estimate of the NNTs.
Although the methodological quality of the trials was mixed, with a number using unclear or unsatisfactory methods of treatment allocation, our sensitivity analyses did not suggest that including these trials has led to any overestimate of treatment effects. Although we noted heterogeneity in some subgroups, overall the trials showed consistent relative effects. As noted above the lack of biochemically validated cessation was the other possible methodological limitation.
Based on subgroup analyses there is little evidence about components that are important as part of an intervention, although direct comparison in a small number of trials suggest that providing a follow-up appointment may increase the effect. Indirect comparisons indicate that various aids tested do not appear to enhance the effectiveness of physician advice. However, caution is required in interpreting such indirect comparisons since they do not take account of any inherent systematic biases in the different populations from which the study samples are drawn. Direct comparison of differing intensities of physician advice suggest a probable benefit from the more intensive interventions compared to a briefer intervention, although subgroup analyses suggest that this might be small or non-existent in unselected smokers, but larger when provided to smokers in high risk groups. The effect of intensified advice in a population with established disease is however based on a small number of trials. If the marginal benefit of a more intensive advice-based intervention is based on the pooled estimate combining unselected and high risk population subgroups (RR CI 1.20 to 1.56), and assuming that the minimal intervention alone could achieve a quit rate of 3.5%, an NNT of 50-140 would be estimated for the effect of providing more support. There was insufficient evidence to draw any conclusion about the effect of motivational as opposed to simple advice (Butler 1999), or between different advice-giving styles (Williams 2001).
If these results are to translate into a public health benefit, the important issue will be the proportion of physicians who actually offer advice. Although 80% of the general population visit a physician annually, reports of the proportion who receive any form of smoking cessation advice vary considerably. While many of those who are not offered smoking cessation advice will quit unaided, every smoker who does not receive advice represents a 'missed opportunity'. Provision of lifestyle advice within the medical consultation is now promoted as a matter of routine, but advice on smoking may still not be offered systematically (Denny 2003; McLeod 2000). Not all primary care physicians agree that advice should be given at every consultation (McEwen 2001), and some practitioners still consciously choose not to raise smoking cessation as an issue in order to preserve a positive doctor-patient relationship (Coleman 2000), although some research indicates that satisfaction may be increased by provision of advice (Solberg 2001).
Several strategies have been shown convincingly to enhance the effectiveness of advice from a medical practitioner, including provision of nicotine replacement therapy and/or bupropion (Hughes 2007; Stead 2008). Addition of either of these forms of therapy increases quit rates 1.5 to 2-fold, and is a potentially valuable adjunct to any advice provided. Both individual and group-based counselling are also effective at increasing cessation rates amongst patients prepared to accept more intensive intervention (Lancaster 2005a; Stead 2005). Telephone counselling can also be effective (Stead 2006). National clinical practice guidelines generally advise the use of a brief intervention in which asking about tobacco use is followed by advice to quit, and an assessment of the smoker's willingness to make a quit attempt. Patients willing to make a quit attempt can then be offered specific assistance and follow up (Fiore 2000; Miller 2001; NHC 2002; West 2000).
Implications for practice
The results of this review indicate the potential benefit from brief simple advice given by physicians to their smoking patients. The challenge as to whether or not this benefit will be realized depends on the extent to which physicians are prepared to systematically identify their smoking patients and offer them advice as a matter of routine.
Implications for research
Further studies of interventions offered by physicians during routine clinical care are unlikely to yield new information about the role of advice. Work is now required to develop strategies to increase the frequency with which smokers are identified and offered advice and support.
Chris Silagy initiated the review and was first author until his death in 2001. Sarah Ketteridge developed the initial version of this review and was a co-author of the review until 1999. Ruth Ashenden provided technical support in the preparation of the initial version of this review.
Dr Chris Hyde, Iain Chalmers and Helen Handoll have all provided constructive comments on previous versions of the review which have been taken account of in the updates.
Martin Shipley provided 33 year follow-up data on the Whitehall study (Rose 78-92) via Iain Chalmers.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Index terms
Intention to treat analyses
I wonder whether the studies included were based on intention to treat analysis? If they were not, I believe that a selection of more motivated subjects has taken place even in studies where unselected populations were invited. Smokers who are not motivated to quit, do not take the same interest in such on offer. Intention to treat principles should be applied if the size of the effect should apply to whole practice populations.
In extracting data from the studies, the denominators were derived from the number of participants stated to be randomised to each condition, and participants lost to follow-up were assumed to be continuing smokers, an intention to treat analysis. Where unselected participants were recruited, the results should therefore reflect whole practice smoker populations. However, the exact way in which participants were recruited differed between trials. In some studies where the intention was to recruit unselected participants, it may be that those recruited were not typical of the practice populations.
Ann Dorrit Guassora (commenter); Lindsay Stead (author)
Last assessed as up-to-date: 13 February 2008.
Protocol first published: Issue 2, 1996
Review first published: Issue 2, 1996
Contributions of authors
Chris Silagy initiated the review and was contact author until the review was updated in 2004 following his death in 2001. Sarah Ketteridge developed the first version of the review including data extraction and drafting. Ruth Ashenden provided technical support in the preparation of the initial version of the review.
Lindsay Stead has identified trials, extracted data and drafted updates since 1998. Tim Lancaster has checked data extraction and finalised updates since 2002.
Rafael Perera gave statistical and translation support for the 2008 update. Gillian Bergson extracted data from trials for the 2008 update, and contributed to revisions of the text.
Declarations of interest
Sources of support
- University of Oxford, Department of Primary Health Care, UK.
- National School for Health Research School for Primary Care Research, UK.
- NHS Research and Development Programme, UK.
Chris Silagy was first author at the time of his death in 2001. The authors for citation were changed when the review was updated in 2004.
Medical Subject Headings (MeSH)
MeSH check words
* Indicates the major publication for the study