Smoking cessation interventions on health-care workers: a systematic review and meta-analysis

Search strategy

This systematic review was performed based on PRISMA statement (Liberati et al., 2009) and of Cochrane Handbook for Systematic Reviews of Intervention (Higgins & Thomas, 2019). The following databases were searched until December 2018: PubMed, Scopus, Web of Science and CINAHL. Articles were retrieved using the string “(Health care personnel OR hospital staff OR medical staff OR paramedical staff OR health care worker*) AND (smoking cessation OR smoking reduction OR stop smoking)”, which was adapted according to the research criteria of each database (we used the character “*” for PubMed and WOS database). We also reviewed the reference lists of the identified articles to avoid missing relevant studies (Wells et al., 2019).

Inclusion and exclusion criteria

The following inclusion criteria were used: (1) randomized clinical trials (RCT); (2) cohort studies; (3) cross-sectional studies; (4) studies regarding health-care workers; (5) studies published in Italian, English, German or Spanish. Among included articles, there were quasi-experimental studies, i.e., research studies that resembles experimental research but is not true experimental research. Although the independent variable is manipulated, participants are not randomly assigned to conditions or orders of conditions (Cook & Campbell, 1979). The following exclusion criteria were used: (1) case-reports; (2) reviews; (3) case-control studies; (4) abstracts and author debates or editorials; (5) lack of effective statistical analysis; (6) animal studies; (7) in vitro studies; (8) studies dealing smoking cessation/reduction that did not include health-care workers as group of intervention.

No filter item about articles period of publication was applied.

Data extraction

Following the inclusion criteria, four authors (VC, BD, AS and GT) independently selected the articles on the basis of relevant titles and abstracts. The full text of each selected paper was then evaluated to determine if it was suitable for inclusion. Disagreements were solved through consensus and by discussion with a reference author (GLT). For each eligible study, the following information was independently extracted by four authors (VC, BD, AS and GT) and examined by reference author (GLT): name of the first author, publication year, title, study design, country, type of intervention, number of health-care workers involved and results.

Quality assessment

Quality of all studies included in the systematic review was assessed according to the Newcastle-Ottawa Scale (NOS) on cohort or cross-sectional studies (Wells et al., 2019) and to the Cochrane Risk of Bias Tool for Randomized Controlled Trials (Higgins & Thomas, 2019), when appropriate. Risk of bias for cohort or cross-sectional studies was performed by the Newcastle-Ottawa Scale (NOS) risk of bias assessment tool for observational studies that is recommended by the Cochrane Collaboration (Lo, Mertz & Loeb, 2014).

Statistical analysis

Risk ratios were calculated for dichotomous outcomes. For meta-analysis, cohort studies and clinical trials were considered; in particular, we have considered cohort studies those with a quasi-experimental or a before-after design. The study from Sarna et al. (2009) was excluded from meta-analysis because it was not suitable for this procedure. Due to the heterogeneous nature of the papers included, we employed random effects models in all meta-analyses. Heterogeneity was documented by the P for homogeneity values reported in Results section. Subgroup analysis and meta-regression were based on fixed effects models. EpiSheet package was used for statistical analyses.

Search results and characteristics of included studies for systematic review

The search strategy gave a total of 3,944 references (3,187 in PubMed, 13 in Scopus, 586 in Web of Science and 158 in CINAHL): of those, 369 were duplicate references. The remaining 3,575 references underwent a title and abstract analysis, after which 3,526 were excluded. As a result, 49 articles were recovered for full-text reading: among them, 26 were excluded (8 did not comply with the language criteria of inclusion and their abstract reported no useful data; 18 did not meet inclusion criteria by full-text). One article was retrieved from snowball procedure and, finally, 24 articles have been included in the analysis (Fig. 1): there were 4 clinical trials (1 from UK, 1 from Croatia, 1 from Egypt and 1 from Switzerland), 3 quasi-experimental studies (2 from USA and 1 from UK), 4 before-after studies (2 from Spain, 1 from Turkey and the remaining 1 from Vietnam) and 13 surveys (2 from UK, 1 from Switzerland, 1 from Denmark, 1 from China, 2 from Australia, 2 from USA and 4 from Spain) (Table 1).

Quality assessment

Quality assessment of before-after studies resulted in poor, fair and good quality in 1, 1 and 1 cases, respectively. All cross-sectional studies were of fair quality (ranging from 4 to 7 points), with the exception of two (Bloor, Meeson & Crome, 2006; Martínez et al., 2018), which had a poor quality: 3 points, both. Two quasi-experimental studies (Longo et al., 1996; Rowe & Clark, 1999) were of fair quality (ranging from 6 to 7 points) and the remaining one (Sarna et al., 2009) was of poor quality. Finally, all clinical trials were of fair quality, excluding one (Mohamed, Mustafa & Del Ghoneim, 2016), which was of poor quality (Table 1).

Results for before-after studies

Results for before-after studies were varying (P for homogeneity <0.01) and depended on the type of intervention: they were positive except for one of them, which showed tiny post intervention differences (Table 1). At risk of bias analysis, three studies reached a good score: Battle et al. (1991) (6), Dao Thi Minh et al. (8) and Santinà et al. (6), whereas Bakan et al. obtained a poor score (4).

Dao Thi Minh et al. (2015) performed a pre- and post-intervention cross-sectional study in nine selected hospitals after a “smoke-free hospital model” implementation: smoking prevalence significantly decreased post-intervention, but the number of daily cigarettes smoked at workplaces among male health workers remained unchanged. Moreover, air nicotine levels in the doctors’ lounges and in emergency departments did not change post-intervention, while at other sites the levels decreased minimally. Finally, Santinà et al. (2011) assessed a plan entitled “Smoke-free hospitals” by conducting two surveys (in 2004 and in 2007): the number of smoking workers decreased significantly in all sectors, but was less marked in nursing staff. In 2007 people only smoked in smoking areas (p < 0.0001). The plan was supported by smokers and non-smokers.

Results for cross-sectional studies

Cross-sectional studies were quite homogenous about the no smoking bans and policies implementation in hospitals; the smoking bans and policies introduction led to a smoking cessation or reduction among health-care workers (Table 1). At the risk of bias evaluation, two papers were scored 3 (Bloor, Meeson & Crome, 2006; Martínez et al., 2018), three were scored 4 (Jones, Crocker & Ruffin, 1998; Poder et al., 2012; Strobl & Latter, 1998), two scored 5 points (Agusti et al., 1991; Etter, Khan & Etter, 2008); among positive evaluations, four papers scored 6 (Martínez et al., 2012; Offord et al., 1992; Stillman, Hantula & Swank, 1994; Xiao et al., 2013) and two scored 7 points, because of the lowest risk of bias (ReyesUrueña et al., 2013; Kannegaard et al., 2005 ).

Martínez et al. (2012) found that significant predictors of abstinence were smoking 10 to 19 cigarettes/day, having present low or medium Fagerström Test for Nicotine Dependence score, and using combined treatment (nicotine replacement therapy and bupropion). The study of Offord et al. (1992) obtained a significant contribution toward providing a healthful work environment and toward encouraging non-smoking behavior in staff and patients by implementing a smoke-free policy in their medical center: decrease in prevalence was the result of both smoking cessation among existing employees and less frequent regular smoking among new employees. ReyesUrueña et al. (2013) in their study demonstrated that tobacco consumption reduction coincided with measures introduced by legislative changes, even if in different manner among health-care workers. Also Stillman, Hantula & Swank (1994) found that physicians and nurses differed significantly on attitudes related to implementation and enforcement of a smoking ban (nurses were more accommodating toward smoking and less to enforce a ban). Strobl & Latter (1998) suggested that smoking policies should aim at strengthening nurses’ determination to give up as well as secure their support for the restrictions in order to assist them in changing their smoking behaviour. Xiao and his coworkers (Xiao et al., 2013) realized that health-care workers’ education was the key priority to stop smoking.

Results for quasi-experimental studies

Three studies (Longo et al., 1996; Rowe & Clark, 1999; Sarna et al., 2009) were considered quasi-experimental studies because they provided no randomization of participants. Quasi-experimental studies analyzed the effects of stop smoking interventions on health-care workers, with one exception (the authors analyzed the effect of a smoking bans). The total results, reported below, are favorable and satisfying (Table 1). The risk of bias analysis showed that two studies (Longo et al., 1996; Rowe & Clark, 1999) reached a good score (6), whereas the study from Sarna et al. obtained a poor score (3).

Longo et al. (1996) found that workplace smoking bans could be effective in saving lives, reducing health care costs, addressing safety concerns, and decreasing operating and maintenance expenses of employers. Rowe & Clark (1999) supported the effectiveness of offering an individualized approach utilizing also salivary cotinine measurements of continuous abstinence at 6 months and 1 year and demonstrated that 24% of student and qualified nurses in the intervention groups stopped smoking compared with 7% of those in the comparison groups, reaching statistical significance. Sarna et al. (2009) demonstrated that use of Nurses QuitNet and workplace factors influenced the nurses to stop smoking.

Results for clinical trials

Since the trial studies designs were very similar, the authors wanted to search the most effective method to help health care professionals to quit smoking (P for homogeneity: 0.061). We could find two different groups of subjects included for each study: one undergoing pharmacotherapy (bupropion or nicotine) and the other one undergoing behavioral therapy or placebo. The results were positive and substantial for all studies, with the experimental group (the one traded with pharmacotherapy) having grater results in quitting smoking than control group (Table 1).

Dalsgareth et al. (2004) demonstrated that bupropion was effective as an aid to smoking cessation in a broad group of hospital employees in Denmark. Glavas, Rumboldt & Rumboldt (2003) proved that short-term transdermal nicotine system was effective in smoking cessation. Mohamed, Mustafa & Del Ghoneim (2016) found that programs promoting smoking cessation including behavioral therapy in addition to the complementary role of pharmacotherapy with bupropion enhanced the chance of success in smoking cessation. Finally, Zellweger et al. (2005) found that bupropion was effective and well tolerated in health care professionals, but prevention measures for relapse were needed to maintain long-term tobacco abstinence among healthcare workers.

Meta-analysis results

For meta-analysis procedure, we have considered cohort studies and clinical trials. We have excluded cross-sectional studies since their study-design can cause biases in the result of our study. We have considered as cohort studies quasi-experimental and before-after studies design.

Overall, we have performed meta-analysis on 10 studies (6 cohort studies and 4 clinical trials; P for homogeneity <0.01). All the considered studies just reached the limit of statistical significance. A 21% of success rate resulted from the application of smoking cessation interventions, pharmacological or behavioral. The resulted pooled RR (Risk ratio) was 1.21 (95% CI [1.06–1.38]) (Fig. 2). Then, we have considered and analyzed the studies on the basis of their design: a 24% of success rates resulted from clinical trials (pooled RR was 1.244; 95% CI [1.099–1.407]) and 19% from cohort studies (pooled RR was 1.192; 95% CI [0.996–1.426]). Considering studies that reached almost a fair quality score (n = 8) to reduce variability and to ensure the good quality of our study (i.e., fair quality by Cochrane scores, ≥6/9 for cohort studies by Newcastle-Ottawa scale), the rate of success reached 18%, being 10% for 3 clinical trials and 21% for 5 cohort studies. The pooled RR for all studies was 1.179 (95% CI [1.030–1.350]), the one for clinical trial was 1.099 (95% CI [1.011–1.194]) and for cohort studies it was 1.206 (95% CI [0.993–1.465]). As a result, after this additional analysis, success rate decreased by three points for all studies (from 21% to 18%) and fourteen points for clinical trials (from 24% to 10%), but a little increase was registered for cohort studies (from 19% to 21%). Considering Clinical Trials, two studies (Glavas, Rumboldt & Rumboldt, 2003; Zellweger et al., 2005) have confidence intervals which include unity and one study (Mohamed, Mustafa & Del Ghoneim, 2016) has a wide confidence interval; as a consequence, the meta-analysis for its results depends heavily on one single study (Dalsgareth et al., 2004).

Meta-regression analysis was completed in order to analyze the influences of the quality, design and size of the studies on our study: we have found that if we weighted studies for the number of participants, the higher the quality of the study, the lower the relative risk was. As a result, the quality of the study may interfere with its result (R² =  − 0.365).