Scolaris Content Display Scolaris Content Display

Poticaji za prestanak pušenja

This is not the most recent version

Collapse all Expand all

Background

Material or financial incentives are widely used in an attempt to precipitate or reinforce behaviour change, including smoking cessation. They operate in workplaces, in clinics and hospitals, and to a lesser extent within community programmes. In this third update of our review we now include trials conducted in pregnant women, to reflect the increasing activity and resources now targeting this high‐risk group of smokers.

Objectives

To determine whether incentives and contingency management programmes lead to higher long‐term quit rates.

Search methods

We searched the Cochrane Tobacco Addiction Group Specialised Register, with additional searches of MEDLINE, EMBASE, CINAHL and PsycINFO. The most recent searches were in December 2014, although we also include two trials published in 2015.

Selection criteria

We considered randomised controlled trials, allocating individuals, workplaces, groups within workplaces, or communities to experimental or control conditions. We also considered controlled studies with baseline and post‐intervention measures. We include studies in a mixed‐population setting (e.g. community‐, work‐, institution‐based), and also, for this update, trials in pregnant smokers.

Data collection and analysis

One author (KC) extracted data and a second (JH‐B) checked them. We contacted study authors for additional data where necessary. The main outcome measure in the mixed‐population studies was abstinence from smoking at longest follow‐up, and at least six months from the start of the intervention. In the trials of pregnant smokers abstinence was measured at the longest follow‐up, and at least to the end of the pregnancy.

Main results

Twenty‐one mixed‐population studies met our inclusion criteria, covering more than 8400 participants. Ten studies were set in clinics or health centres, one in Thai villages served by community health workers, two in academic institutions, and the rest in worksites. All but six of the trials were run in the USA. The incentives included lottery tickets or prize draws, cash payments, vouchers for goods and groceries, and in six trials the recovery of money deposited by those taking part. The odds ratio (OR) for quitting with incentives at longest follow‐up (six months or more) compared with controls was 1.42 (95% confidence interval (CI) 1.19 to 1.69; 17 trials, [20 comparisons], 7715 participants). Only three studies demonstrated significantly higher quit rates for the incentives group than for the control group at or beyond the six‐month assessment: One five‐arm USA trial compared rewards‐ and deposit‐based interventions at individual and group level, with incentives available up to USD 800 per quitter, and demonstrated a quit rate in the rewards groups of 8.1% at 12 months, compared with 4.7% in the deposits groups. A direct comparison between the rewards‐based and the deposit‐based groups found a benefit for the rewards arms, with an OR at 12 months of 1.76 (95% CI 1.22 to 2.53; 2070 participants). Although more people in this trial accepted the rewards programmes than the deposit programmes, the proportion of quitters in each group favoured the deposit‐refund programme. Another USA study rewarded both participation and quitting up to USD 750, and achieved sustained quit rates of 9.4% in the incentives group compared with 3.6% for the controls. A deposit‐refund trial in Thailand also achieved significantly higher quit rates in the intervention group (44.2%) compared with the control group (18.8%), but uptake was relatively low, at 10.5%. In the remaining trials, there was no clear evidence that participants who committed their own money to the programme did better than those who did not, or that contingent rewards enhanced success rates over fixed payment schedules. We rated the overall quality of the older studies as low, but with later trials (post‐2000) more likely to meet current standards of methodology and reporting.

Eight of nine trials with usable data in pregnant smokers (seven conducted in the USA and one in the UK) delivered an adjusted OR at longest follow‐up (up to 24 weeks post‐partum) of 3.60 (95% CI 2.39 to 5.43; 1295 participants, moderate‐quality studies) in favour of incentives. Three of the trials demonstrated a clear benefit for contingent rewards; one delivered monthly vouchers to confirmed quitters and to their designated 'significant other supporter', achieving a quit rate in the intervention group of 21.4% at two months post‐partum, compared with 5.9% among the controls. Another trial offered a scaled programme of rewards for the percentage of smoking reduction achieved over the course of the 12‐week intervention, and achieved an intervention quit rate of 31% at six weeks post‐partum, compared with no quitters in the control group. The largest (UK‐based) trial provided intervention quitters with up to GBP 400‐worth of vouchers, and achieved a quit rate of 15.4% at longest follow‐up, compared to the control quit rate of 4%. Four trials confirmed that payments made to reward a successful quit attempt (i.e. contingent), compared to fixed payments for attending the antenatal appointment (non‐contingent), resulted in higher quit rates. Front‐loading of rewards to counteract early withdrawal symptoms made little difference to quit rates.

Authors' conclusions

Incentives appear to boost cessation rates while they are in place. The two trials recruiting from work sites that achieved sustained success rates beyond the reward schedule concentrated their resources into substantial cash payments for abstinence. Such an approach may only be feasible where independently‐funded smoking cessation programmes are already available, and within a relatively affluent and educated population. Deposit‐refund trials can suffer from relatively low rates of uptake, but those who do sign up and contribute their own money may achieve higher quit rates than reward‐only participants. Incentive schemes conducted among pregnant smokers improved the cessation rates, both at the end‐of‐pregnancy and post‐partum assessments. Current and future research might continue to explore the scale, loading and longevity of possible cash or voucher reward schedules, within a variety of smoking populations.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

Mogu li poticaji pomoći pušačima da se zauvijek ostave pušenja?

Dosadašnje spoznaje

Poticaji, često u vidu novca ili bonova, se naveliko koriste kako bi se potaklo pušače da pokušaju prestati pušiti, te ih se nagrađuje kada u tome uspiju. Takvi projekti se mogu provoditi na radnom mjestu, u klinikama i ponekad kao dio javnih programa u zajednici. U ovom trećem ažuriranju Cochrane sustavnoga pregleda, sada su uvrštene i studije koje uključuje trudnice. Još jedna promjena u odnosu na ranije verzije ovog sustavnog pregleda je i to da je napravljena podjela ispitivanja koja koriste poticaj i onih koji koriste natjecanje; sada se u ovom sustavnom pregledu analiziraju samo poticaji, a natjecanje će biti obrađeno zasebno.

Karakteristike istraživanja

Ispitivanje na miješanoj populaciji: Pronađeno je 21 istraživanje i obuhvaćeno je preko 8.400 ljudi; ispituju se različiti programi koji bi trebale pomoći pušačima da prestanu pušiti. Zadnje pretraživanje literature je provedeno u prosincu 2014. godine, iako su mu pridružena i 2 istraživanja objavljena nakon tog datuma. Deset je istraživanja bilo provedeno u klinikama, jedno u selima na Tajlandu, gdje su radili javnozdravstveni radnici, a ostatak je proveden na radnim mjestima. Sva osim njih 6 provedena su u SAD‐u. Sva istraživanja su pratila ispitanike barem 6 mjeseci te je provjeravano jesu li prestali pušiti na način da je analiziran njihov dah ili tjelesne izlučevine. Poticaji su uključivali kartice za lutriju, nagradne pitalice, isplatu novca, bonova za predmete i namirnice, te jamstvo povrata novca nakon 6 mjeseci za one koji su sudjelovali u istraživanju.

Studije provedene na trudnicama: Pronađeno je 9 istraživanja, 8 ih je provedeno u SAD‐u i 1 u Velikoj Britaniji. Ta su istraživanja uključila 1.600 žena koje su željele prestati pušiti u trudnoći.

Ključni rezultati
Ispitivanja na miješanim populacijama:Šest ili više mjeseci nakon početka istraživanja, ispitanici koji su primili poticaje su imali veću vjerojatnost da prestanu pušiti od onih u kontrolnim skupinama. Samo 3 istraživanja zabilježila su produljen uspjeh i nakon prestanka istraživanja. Jedno američko ispitivanje, u kojemu je plaćeno 750 američkih dolara onima koji su prestali pušiti, pokazalo je da je 3 puta više ispitanika prestalo pušiti u pokusnoj skupini naspram kontrolnih. Drugim američkim istraživanjem, u kojemu je isplaćeno 800 američkih dolara po ispitaniku koji je prestao pušiti, pokazalo je da je broj ispitanika koji su prestali pušiti, a primili su poticaj, naspram kontrolne skupine, bio dvostruko veći. Tim je istraživanjem također uspoređen učinak povrata novca, koji su ispitanici sami uložili ako su prestali pušiti. Iako je veći broj ispitanika sudjelovao u programu s nagradama nego u depozitnom programu, razmjer ispitanika koji su prestali pušiti u svakoj skupini bio je veći u depozitnom programu nego u programu s nagradama. Treće istraživanje provedeno je na Tajlandu gdje je ispitanicima vraćen njihov uloženi novac, ali su im bile isplaćene i nagrade. Iako su stope prestanka pušenja bile ohrabrujuće, depozitni program pokazao je nedostatak jer nije moglo biti isplaćeno toliko nagrada zbog prevelika broja ispitanika koji su bili voljni sudjelovati.

Ispitivanje na trudnicama: Osam od devet istraživanja pokazalo je da žene u skupinama koje su primale poticaje pokazuju najveću stopu prestanka pušenja tijekom trudnoće i dugoročno (do 24 tjedna nakon poroda). Četiri ispitivanja potvrdila su da je plaćanje pretvoreno u nagradu kao poticaj za prestanak pušenja („slučajno“ plaćanje), u usporedbi s fiksnom isplatom ukoliko su pohađale prenatalne sastanke („ne‐slučajno“ plaćanje), pokazalo veću stopu prestanka pušenja. U najvećem istraživanju, provedenom u Velikoj Britaniji, korištena je nacionalna nepušačka služba kao pomoć za prestanak pušenja te je postignuta četiri puta veća stopa prestajanja pušenja nego u kontrolnoj skupini. Jedno istraživanje u kojem je korišteno plaćanje pokazalo je da su simptomi prestanka pušenja u prva dva tjedna malo utjecali na uspjeh ispitanica u prestanku pušenja.

Kvaliteta dokaza:
Cjelokupna kvaliteta za istraživanja na starijim miješanim populacijama je procijenjena niskom, ali u kasnijim istraživanjima (iza 2000. godine) veća je vjerojatnost da će istraživanja biti bolje provedena i opisana. Uklanjanje ranih istraživanja iz izračuna malo je utjecalo na rezultate uočene nakon šest mjeseci, ali je zabilježena bolji rezultat za praćenje nakon dvanaest mjeseci. Istraživanja na trudnicama objavljena su unutar zadnjih petanest godina te su bila općenito umjerene kvalitete.

Authors' conclusions

available in

Implications for practice

Mixed‐population interventions

  • Incentives appear to boost cessation rates while they are in place. Their contribution to long‐term cessation remains to be determined.

  • The two trials recruiting from worksites which achieved sustained success rates beyond the reward schedule concentrated their resources into substantial cash payments for abstinence. Such an approach may be feasible only where independently‐funded smoking cessation programmes are already established. One of the trials has subsequently failed to translate to a real‐world application because of worker resistance and costs to the host company.

  • Deposit refund trials may be prone to relatively low rates of uptake compared to reward‐based programmes; however, people who do sign up and contribute their own money achieve higher quit rates than reward‐only participants.

  • Although these interventions risk attracting smokers motivated more by the material rewards than by the desire to quit, there was little evidence that levels of deception varied between experimental and control subjects, or that rates of disconfirmation were unacceptably high.

Pregnancy interventions

  • Rewards contingent on validated cessation may ensure sustained abstinence into the post‐partum period.

  • Incentives given for abstinence to the end of the pregnancy boost cessation rates compared with routine antenatal care.

  • There is limited evidence that non‐contingent rewards, for attendance and for supplying a biological sample, do not lead to increased rates of cessation.

Implications for research

Mixed‐population interventions

  • Substantial high‐quality trials are needed to compare incentive schemes to those interventions which do not offer rewards.

  • The scale and longevity of different reward schedules for smoking cessation should be evaluated.

  • The respective merits of cash payments versus payments in kind (e.g. grocery vouchers) should be assessed and compared.

  • The stability of successful payment schedules needs to be tested in varying populations of smokers, from different socio‐economic, regional and ethnic backgrounds.

  • Head‐to‐head comparisons between reward‐based and deposit‐refund‐based interventions need further investigation.

  • The acceptability and affordability of incentive programmes should be tested in real‐world settings, as part of the evaluation process.

Pregnancy interventions

  • Trials in pregnant women should try to isolate the effect of variable versus fixed payment schedules, separately from evaluations of contingency versus guaranteed payments.

  • Trials in pregnant women should assess the feasibility of using standard in situ delivery methods (e.g. local or national stop‐smoking services, routine antenatal care) compared to ad hoc provision of support offered as part of the trial procedure.

Summary of findings

Open in table viewer
Summary of findings for the main comparison. Incentives for smoking cessation

Incentives for smoking cessation

Participant or population: Adult smokers
Settings: mixed
Intervention: Incentives

Outcomes

Illustrative comparative risks* (95% CI)

Relative effect
(95% CI)

No of Participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

Control

Incentives

Smoking cessation in mixed populations ‐ longest follow‐up
Follow‐up: 6 ‐ 24 months

84 per 1000

112 per 1000
(95 to 132)

OR 1.42
(1.19 to 1.69)

7715
(17 studies, 20 comparisons)

⊕⊕⊝⊝
low1,2,3

Beyond the 6‐month assessment, only 2 studies (Halpern (A) 2015; Volpp 2009) demonstrated sustained cessation.

*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; OR: Odds ratio;

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

111/20 trials failed to conduct adequate randomisation and allocation procedures, and 6 of them were rated at high risk of bias in one or both of these domains.
2Eight of the 20 studies (generally the older trials) presented results as percentages or in graphical form, which obliged us to estimate the findings without being certain of the true denominators or event counts
3Three early trials may have confounded their findings by assessing smoking status at the same time as they delivered six‐month rewards for abstinence. In one case (Rand 1989) this was also the end of the trial.

Open in table viewer
Summary of findings 2. incentives for smoking cessation in pregnant women at longest follow‐up

incentives for smoking cessation in pregnant women at longest follow‐up

Participantnt or population: Pregnant women who smoke
Settings: antenatal clinics
Intervention: Incentives

Outcomes

Illustrative comparative risks* (95% CI)

Relative effect
(95% CI)

No of Participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

Control

Incentives

Smoking cessation in pregnant women at longest follow‐up
Follow‐up: 10 ‐ 24 weeks post partum

48 per 1000

154 per 1000
(108 to 216)

OR 3.60
(2.39 to 5.43)

1295
(8 studies)

⊕⊕⊕⊝
moderate1

1 included study (Donatelle 2002) did not contribute to the analysis because of lack of usable data

*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; OR: Odds ratio;

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

1Only 2 studies (Ondersma 2012; Tappin 2015) were rated at low risk of selection bias; the rest were at unclear risk, apart from Higgins 2004 (high risk).

Background

available in

There is a growing enthusiasm for incentive‐based programmes to change unhealthy behaviours, including smoking, weight loss, alcohol consumption and levels of physical activity (NICE 2010). Many developing countries, particularly in Latin America, operate conditional cash transfer programmes of monetary rewards for behaviour change or compliance, often targeting improvements in child and maternal health (Lagarde 2009; Paes‐Sousa 2011; Powell‐Jackson 2011). In the UK incentive schemes often focus on encouraging pregnant smokers to quit, with programmes such as 'Give It Up For Baby' (Ballard 2009; Radley 2013) and 'quit4u' (Ormston 2014), conducted in Tayside (Scotland) and awarding grocery vouchers for verified abstinence. A series of studies in the USA (Donatelle 2000a; Donatelle 2000b; Donatelle 2002; Heil 2008; Higgins 2004; Higgins 2014) and a major randomised trial in the UK (Tappin 2015) also attest to the tobacco control community's interest in the feasibility of rewarding pregnant smokers for achieved abstinence.

The workplace is also a common setting for the use of incentives. This is briefly addressed in a companion review (Cahill 2014), but we explore it more fully here. Most of the relevant studies have been conducted in the USA, in part because of the structure of the healthcare system there, which obliges employers to cover health insurance costs for their workforce. In other countries, where the state or private insurance companies are the main healthcare provider, there may be less tangible incentive for employers to take direct responsibility for the health of their workers. There are a number of advantages to offering smoking cessation support in the workplace, including the accessibility of the target population, the availability of occupational health support and the potential for peer pressure and peer support. Because of the existing salary and bonus structure, it is also relatively easy to set up a rewards system to supplement the programme, if that is the chosen mechanism. Seventy‐one per cent of large US companies (employing more than 200 workers) offer smoking cessation programmes, and 36% offer financial incentives for wellness programmes, but generally for participation in and completion of a programme rather than for a verified change in health behaviour. From January 2014, section 1201 of the Affordable Care Act (2010) has increased the maximum allowable reward to as high as 50% of the total health insurance premium for smoking cessation (Cawley 2014).

A 2012 survey by the RAND Corporation of USA wellness plans found that, among those using incentives in companies with more than 50 employees, 84% frame them as rewards only, 11% frame them as both rewards and penalties, and 2% frame them as penalties only (3% unknown) (Cawley 2014; Mattke 2013). We consider the implications of this approach in the Discussion section below.

Description of the intervention

Incentives and rewards (terms used interchangeably in studies contributing to this review) routinely feature in many smoking cessation programmes, in support of the quitting process. They can be used to encourage recruitment into the programme, to reward compliance with the process, and to reward cessation achieved at predefined stages. A variety of rewards have been used for these purposes, including cash payments, vouchers exchangeable for goods (excluding alcohol and cigarettes), salary bonuses, promotional items such as T‐shirts, pens and bags, holidays, and luxury goods such as cars or boats.

Rewards can be given for attendance at the programme and at follow‐up appointments, irrespective of subsequent smoking status (i.e. guaranteed or non‐contingent), or can be paid and scaled relative to the participant's success within the programme (i.e. contingent) (Higgins 2002). Recent trials and systematic reviews have explored variations in the type, the scale, and the scheduling of rewards (Adams 2014; Crossland 2015; Giles 2014; Jochelson 2007; Leeks 2010; Sigmon 2012b), and in their acceptability as a mechanism for behaviour change (Hoddinott 2014; Thomson 2014).

Why it is important to do this review

Over the ten‐year lifetime of this review, the debate about incentive‐based smoking cessation programmes has shifted from their feasibility (i.e. can they work?) to the relative success or limitations of the mechanisms deployed (Higgins 2012; Promberger 2012), the merits of rewards ('carrots') versus penalties ('sticks') (Adams 2014; Lynagh 2013; Volpp 2014), the extent to which achieved changes can be maintained (Jochelson 2007; Strickland 2014) , and the possibilities of unintended consequences (Marteau 2009; Thomson 2014). Although many of the older included studies may not address these issues, our review contributes to a growing evidence base which defines the rationale for incentive‐based programmes and identifies areas for further investigation. We have also revised the inclusion criteria to cover interventions in pregnant smokers, since this population is increasingly the focus of incentive‐based smoking cessation programmes worldwide.

This updated review is a modified version of our previous review Competitions and incentives for smoking cessation (Cahill 2011). As the emphasis in public health has shifted in recent years away from cessation competitions, and towards contingent and non‐contingent incentive programmes, we have now split the review into two. This updated version is restricted to financial and goods‐based incentives, while the forthcoming companion update, Competitions for smoking cessation, will cover competition‐based cessation programmes, both institutional and community‐based, including 'Quit and win' contests. Although previous versions of this review excluded interventions targeting pregnant women, we now include them in this update; another Cochrane review, Psychosocial interventions for supporting women to stop smoking in pregnancy (Chamberlain 2013) covers some of the same trials, but addresses different outcomes, concentrating on antepartum abstinence and the wellbeing of the baby rather than on the prolonged smoking status of the mother.

Objectives

available in

To determine whether incentives and contingency management programmes lead to higher long‐term quit rates. We address the following questions:

Primary outcome: Cessation:

1. Do incentives reduce the prevalence of smoking?

Secondary outcomes:

2. What are the cost implications of incentives, to employers and to the community?
3. How great is the risk of disbenefits arising from the use of incentives, e.g. false claims, ineligible applicants?

Methods

available in

Criteria for considering studies for this review

Types of studies

Randomised controlled trials allocating individuals, communities, workplaces or groups within workplaces to intervention or to control conditions.
Controlled trials with baseline measures and post‐intervention outcomes.

Types of participants

Adult smokers, of either gender, in any setting. This group, which we term 'mixed populations', currently includes participants drawn from clinics or hospitals, from the community, from academic institutions, and from worksites. We have not included trials aimed exclusively at adolescent smokers, as they are covered by other Cochrane reviews.
For this 2015 update, we also include trials conducted in pregnant women who smoke.

Types of interventions

Incentive schemes, lotteries, raffles, and contingent or non‐contingent payments, to reward cessation and abstinence in smoking cessation programmes. We have not included reports of the effectiveness of incentives or rewards to healthcare workers (physicians, nurses) for the delivery of smoking cessation interventions, or of reimbursement to patients for smoking cessation treatment costs, as these are covered in another Cochrane review (Reda 2012).

Types of outcome measures

The primary outcome for this review is cessation rates, including point prevalence and sustained abstinence. For trials in the mixed populations abstinence had to be assessed at a minimum of six months from the start of the intervention, whether or not they were biochemically validated (Hughes 2003). The gold standard is biochemically‐verified sustained abstinence for at least six months. We exclude from this review trials which do not report cessation rates.
For trials in pregnant women, we extracted smoking cessation outcomes at the closest follow‐up to end of pregnancy, and also at longest follow‐up postpartum if reported. We did not require the minimum six‐month follow‐up period.

A secondary outcome, added for the 2015 update, is adverse events or unintended consequences.

Search methods for identification of studies

We searched the Cochrane Tobacco Addiction Group Specialised Register, which includes studies identified by systematic electronic searches of multiple databases, handsearching of specialist journals, and 'grey' literature (conference proceedings and unpublished reports not normally covered by most electronic indexing systems). In addition, we used specifically developed strategies to search four electronic databases: MEDLINE, EMBASE, CINAHL and PsycINFO. Search terms included incentive*, competition*, contest*, lotter*, reward*, prize*, contingent payment*, deposit contract*. The Cochrane Register of Studies search strategy is given in Appendix 1. The most recent searches were in December 2014.

We checked reference lists of published papers, and have consulted with experts in the field to identify any relevant forthcoming and/or unpublished research.
We have contacted the authors of ongoing studies where necessary, and have recorded their co‐operation in the Acknowledgements section.

We also searched trials registries (www.clinicaltrials.gov; www.controlled‐trials.com/mrct/; apps.who.int/trialsearch/; www.anzctr.org.au) for ongoing and completed trials, and report on these in the relevant reference section (Characteristics of ongoing studies).

Data collection and analysis

Selection of studies

One review author (KC) prescreened all search results (titles and abstracts), for possible inclusion or as useful background. Two review authors then independently assessed relevant studies for inclusion, resolving discrepancies by consensus. We noted reasons for the non‐inclusion of studies, and report these in the Characteristics of excluded studies tables.

Data extraction and management

One review author (KC), using a tailored data extraction form, extracted and summarised study data and a second review author (JH‐B) checked the findings. Where available, we recorded the following information in the Characteristics of included studies table:

  1. Methods: study design, study name (if applicable), study recruitment period, country, number of study centres, study setting, study recruitment procedure.

  2. Participants: N (intervention/control), definition of smoker used, specific demographic characteristics (e.g. mean age, age range, gender, ethnicity), mean cigarettes per day, mean FTND, inclusion criteria, and any relevant exclusion criteria.

  3. Interventions: Description of intervention(s) (treatment, dosage, regimen, behavioural support), description of control (treatment, dosage, regimen, behavioural support); what comparisons were constructed between which groups, concomitant medications, and excluded medications.

  4. Outcomes: primary and secondary outcomes specified and collected, time points reported, biochemical validation, definitions of abstinence.

  5. Notes: funding for trial, and notable conflicts of interest of trial authors.

Assessment of risk of bias in included studies

We evaluated each included study for risks of bias, using the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Cochrane Handbook, Higgins 2011, Chapter 8). The domains examined for this review include:

  1. Random sequence generation (selection bias)

  2. Allocation concealment (selection bias)

  3. Blinding of participants, personnel and outcome assessors (performance bias and detection bias)

  4. Attrition bias

  5. Other potential risks of bias

We rated each domain at low, unclear, or high risk of bias, with quotations from the study report and reasons to justify our judgements. We have summarised the 'Risk of bias' judgements across different studies for each of the domains listed, and display the summary results in a 'Risk of bias' figure.

Measures of treatment effect

We have combined eligible studies using a generic inverse variance model, to produce adjusted odds ratios (ORs) using a fixed‐effect model. Adjustment is for cluster randomisation where appropriate. Raw data (event counts and denominators) for the key meta‐analyses are displayed in Appendix 2. Our primary analysis is for smoking cessation at longest follow‐up (at least six months).

Unit of analysis issues

Several mixed‐population studies were cluster‐randomised, i.e. allocated by group, community, or workplace. We have used the intraclass correlation coefficient reported by Martinson 1999 (ICC for percentage quit smoking in Worksite) to obtain an adjusted estimate of the effect size for the studies that were cluster‐randomised. Wherever possible, and whether or not the trialists themselves used this approach, we have used an intention‐to‐treat analysis, deploying as the denominator the numbers randomised in their original groups.

Dealing with missing data

We contacted investigators or study sponsors in order to verify key study characteristics and to obtain missing numerical outcome data where possible (e.g. when a study is reported as abstract only). Where this was not possible, and the missing data were thought to introduce serious bias, we have explored the impact of including such studies in the overall assessment of results by a sensitivity analysis.

We have generally treated participants who dropped out or who were lost to follow‐up after randomisation as being continuing smokers. We note the proportion of participants for whom the outcome was imputed in this way, and whether there was either high or differential loss to follow‐up between the groups. The assumption that 'missing = smoking' will give conservative absolute quit rates, and will make little difference to the risk ratio unless drop‐out rates differ substantially between groups. Where possible, we conducted an intention‐to‐treat analysis, including all smokers randomised.

In trials of pregnant smokers, we have followed the convention observed in most of the trials, and not included in the denominator women whose pregnancies were uncompleted because of termination or foetal death.

Assessment of heterogeneity

We have used the I² statistic to assess statistical heterogeneity, given by the formula [(Q ‐ df)/Q] x 100%, where Q is the Chi² statistic and df is its degrees of freedom (Higgins 2003). This describes the percentage of the variability in effect estimates that is due to heterogeneity rather than to sampling error (chance). A value greater than 50% may be considered to indicate substantial heterogeneity.

Assessment of reporting biases

As there are a sufficient number of included studies (ten or more contributing to the outcome), we have created a funnel plot to assist in identifying possible publication bias, methodological flaws, or small‐study effects. We have searched for and report on studies we know to have been completed but for which results are unavailable.

Subgroup analysis and investigation of heterogeneity

The pregnancy trials address a number of questions about the efficacy of contingent payments for smoking cessation:

  1. Does the size of the reward make a difference?

  2. Do stepped rewards (increasing with prolonged abstinence) work differently from a schedule of fixed payments?

  3. Does 'front‐loading' of the reward schedule, to counteract the stronger temptation to relapse in the early stages of quitting, work better than an incremental schedule?

  4. Does participant‐initiated verification of abstinence work better than a researcher‐controlled process?

We have explored these issues where possible through subgroup analysis of the included trials.

'Summary of findings' tables

We have created 'Summary of findings' tables using the following outcomes:
Mixed‐population studies: Smoking cessation at longest follow‐up (summary of findings Table for the main comparison).
Pregnancy trials: Smoking cessation at longest follow‐up (post‐partum where available) (summary of findings Table 2).

We have used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the quality of the body of evidence as it relates to the studies which contribute data to the prespecified outcomes. We have used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook, using GRADEpro software. We justify all decisions to down‐ or upgrade the quality of the evidence using footnotes, and have made comments to aid reader's understanding of the review where necessary.

The Tobacco Addiction Group's Glossary of tobacco‐related terms is given in Appendix 3.

Results

Description of studies

Interventions in mixed populations

We retain 14 of the 19 studies which had met our inclusion criteria and were included in the 2011 version of this review. The other five studies formerly included (Gomel 1993; Klesges 1986; Klesges 1987; Koffman 1998; Maheu 1989) will now be covered by the separate review Competitions for smoking cessation. From our latest searches, we have identified six further community‐based studies (Alessi 2014; Drummond 2014; Giné 2010; Ledgerwood 2014 (previously listed as an ongoing study); Secades‐Villa 2014; White 2013) which meet our inclusion criteria. We also include Halpern (A) 2015, which was published after we had conducted our searches, giving a total of 21 included studies (8413 participants) for this update. We have also identified 15 additional studies which we add to the excluded studies, 13 ongoing studies from trials registries, and one published protocol for a forthcoming trial (Etter 2012).

Outcomes
All the included studies rewarded smoking cessation, either alone or in combination with recruitment or participation or both (See the Characteristics of included studies table for full details).

Raw outcome data, particularly in the older studies, were often difficult to extract, with eight of the 21 studies presenting results as percentages only, in tabular or graphical form. Nine trials followed up participants for a maximum of six months (Alessi 2014; Crowley 1995; Drummond 2014; Ledgerwood 2014; Paxton 1980; Paxton 1981; Rand 1989; Secades‐Villa 2014; Tevyaw 2009), three for between six and twelve months (Gallagher 2007; Giné 2010; Volpp 2006), four for twelve months (Halpern (A) 2015; Paxton 1983; Shoptaw (A) 2002; Windsor (A) 1988), two for 13 to 18 months (Volpp 2009; White 2013), and three for 24 months (De Paul 1994; Glasgow 1993; Hennrikus 2002).

Settings
Ten of the studies were set in clinics or health centres (Alessi 2014 [residential substance use disorder clinic]; Crowley 1995 [COPD patients]; Drummond 2014 [injection drug users' clinic]; Gallagher 2007 [psychiatric patients, including people with schizophrenia]; Paxton 1980; Paxton 1981; Paxton 1983; Secades‐Villa 2014 [Addictive Behaviors Clinic]; Shoptaw (A) 2002 [narcotic abuse patients]; Volpp 2006), one in villages served by community health workers (White 2013), two in an academic institution (Ledgerwood 2014; Tevyaw 2009), one in an urban community (Giné 2010), and the rest in worksites. Most of the studies were based in the USA, apart from the three Paxton studies (UK), Secades‐Villa 2014 (Spain), Giné 2010 (the Philippines), and White 2013 (Thailand).

Incentives
Three studies used lottery or raffle tickets as the incentive (Alessi 2014; Crowley 1995; Ledgerwood 2014), with the latter comparing a traditional contingency management schedule with an enhanced version guaranteeing rewards during the first week of the cessation programme. Nine studies rewarded verified abstinence with cash payments (De Paul 1994; Drummond 2014; Gallagher 2007; Halpern (A) 2015; Rand 1989; Shoptaw (A) 2002; Volpp 2006; Volpp 2009; Windsor (A) 1988). Two studies (Glasgow 1993; Hennrikus 2002) combined cash payments to individual quitters with one or more site‐wide prize draws.

Six studies (Giné 2010; Halpern (A) 2015; Paxton 1980; Paxton 1981; Paxton 1983; White 2013) tested a system of deposits refunded for abstinence over the course of the programme. The Paxton studies compared possible reward schedules by varying the timing and the amount of deposits and repayments. Giné 2010 set up non‐interest‐bearing savings accounts for intervention participants, and topped up contributions throughout the course of the programme, with successful quitters recovering all the money at programme end. White 2013 adopted a similar model, but paired up participants in teams, and gave them additional rewards if both partners achieved abstinence, as well as returning savings to individual quitters. Halpern (A) 2015 included both individual and team‐based deposit‐refund arms in a five‐arm trial, with participants contributing a proportion of the available rewards for abstinence.

Three studies compared the effects of fixed automatic payments with payments contingent upon cessation, with the guaranteed payment regimen generally serving as the control condition (Crowley 1995; Rand 1989; Tevyaw 2009).

One study (Secades‐Villa 2014) allocated cumulative points for consecutive confirmed tests for abstinence, which translated to vouchers for local goods and services.

One study (Glasgow 1993) also included lottery tickets for a prize draw for smoking 'buddies', who supported those smokers trying to quit.

Although all the studies rewarded smoking cessation as the primary outcome, several added incentives for other performance indicators. Participation and compliance were rewarded by Alessi 2014; Crowley 1995; De Paul 1994; Gallagher 2007; Giné 2010; Glasgow 1993; Halpern (A) 2015; Hennrikus 2002; Ledgerwood 2014; Volpp 2006; Volpp 2009.

Cessation methods
Only one trial (Glasgow 1993) of the 21 did not deploy any kind of cessation support programme.

Three of the earlier studies used aversive smoking as part of a multi‐component programme (Paxton 1980; Paxton 1981; Paxton 1983). Four included nicotine replacement therapy to support their participants (Crowley 1995; Gallagher 2007; Shoptaw (A) 2002; Volpp 2006).

The 13 remaining studies all used some form of multi‐component support programme. Four studies primarily offered a self‐help programme (De Paul 1994; Ledgerwood 2014; Rand 1989; Windsor (A) 1988), five offered individual or group counselling (Alessi 2014; Hennrikus 2002; Secades‐Villa 2014; Tevyaw 2009; White 2013), one combined standard counselling with spirometry feedback and information on local smoking cessation services (Drummond 2014), while Volpp 2009 and Halpern (A) 2015 steered all participants towards locally‐available smoking cessation resources. Halpern (A) 2015 also provided self‐help materials, and encouraged those participants with company‐sponsored health benefits to access available behavioural support and NRT if they wished to. Both Giné 2010 and White 2013 added encouragement and peer pressure from deposit‐collectors to support participants during the quitting period.

Interventions in pregnancy

For this update, we also include, as a separate group, nine trials (1783 women) conducted in pregnant smokers. As part of this grouping, we also identified nine excluded studies, five ongoing studies from trial registries, and two protocols for forthcoming trials. One ongoing study (Tappin 2015; CPIT study) published its results as we were completing the update, and after the date of the last search, but we have included the findings in this review.

Outcomes
All the included studies reported abstinence at the end of pregnancy, with six of the nine tracking participants into the post‐partum stage. Two trials (Donatelle 2000b; Donatelle 2002) referred simply to 'abstinence', without further definition of the type or duration. In all cases, rewards were available only for biochemically‐verified abstinence. Two trials rewarded smoking reduction as well as complete abstinence, with Tuten 2012 setting percentage reduction targets to be met for rewards, while Higgins 2014 allocated higher‐value vouchers for breath samples below 4 ppm rather than 6 ppm in the early stages of the trial. Because our concern here is for the mothers' long‐term cessation, our primary outcome of interest for this group is abstinence at the longest available assessment point, but we also report abstinence rates at or around the end of pregnancy for all the trials. See tabulated results in Analysis 1.2.

Settings
Eight studies were conducted in the USA, mostly in public or private antenatal clinics, obstetric practices, and Women, Infants, and Children programme sites. One trial (Tuten 2012) in methadone‐maintained pregnant women was conducted in the Center for Addiction and Pregnancy in Baltimore. The largest study (Tappin 2015; the Cessation in Pregnancy Incentives Trial), with 612 participants, was UK‐based, in an NHS health board area (Greater Glasgow and Clyde) with three major maternity hospitals. This study was mediated through routine use of the UK Stop‐smoking Service.

Incentives
In all cases the rewards were vouchers for goods or services, rather than cash payments. Three trials (Donatelle 2000a; Donatelle 2000b; Donatelle 2002) delivered monthly rewards contingent upon proven abstinence. Four trials evaluated the allocation of incremental rewards, with the voucher reset to baseline value in the case of relapse or missed visits, but restored to previous levels if abstinence was re‐established (Heil 2008; Higgins 2004; Higgins 2014; Tuten 2012). Ondersma 2012, using a computer‐based intervention, shifted the onus of testing to the participants, who could present themselves as often as they wished for verification of abstinence, and could win up to five USD 50 gift cards over the course of the programme. Tappin 2015 awarded vouchers up to a value of GBP 350 for achieving staged cessation targets, and a further GBP 50 for engaging with the programme and setting a quit date. Donatelle 2000a also rewarded a social supporter, in tandem with the participant smoker. Non‐contingent rewards, roughly equivalent to the value available to the intervention group, were given to control participants in three trials (Heil 2008; Higgins 2004; Higgins 2014), while Tuten 2012 incorporated a group on a schedule of non‐contingent rewards generated from an earlier pilot study. Donatelle 2000a gave a USD 5 voucher to all participants for each of three attendances during the trial. Tappin 2015 gave all participants a GBP 25 shopping voucher for supplying primary outcome information (34 to 38 weeks gestation) and a biological sample if appropriate.

Cessation methods
All the trials offered a programme of practical cessation support, in addition to the routine care delivered by the host clinics. Three trials (Donatelle 2000b; Donatelle 2002; Ondersma 2012) used the 5As approach (Ask, Advise, Assess, Assist, Arange), while five trials offered self‐help materials. Tuten 2012 also included a brief motivational interviewing feedback session for all participants. The UK trial (Tappin 2015) referred all participants to the UK Stop‐smoking Service, which routinely conducted a one‐hour cessation session, four weekly phone calls, and free NRT if the woman requested it.

Risk of bias in included studies

Mixed populations studies

Randomisation
Eighteen of the included studies were described as randomised (Alessi 2014; Crowley 1995; De Paul 1994; Drummond 2014; Gallagher 2007; Giné 2010; Glasgow 1993; Halpern (A) 2015; Hennrikus 2002; Ledgerwood 2014; Rand 1989; Secades‐Villa 2014; Shoptaw (A) 2002; Tevyaw 2009; Volpp 2006; Volpp 2009; White 2013; Windsor (A) 1988), with ten of them also using stratification (Alessi 2014; Crowley 1995; De Paul 1994; Glasgow 1993; Halpern (A) 2015; Hennrikus 2002; Ledgerwood 2014; Volpp 2006; Volpp 2009; White 2013). In the remaining three studies randomisation was not used, with the Paxton trials allocating individual attenders to the next available treatment group.

We consider that seven studies (Drummond 2014; Halpern (A) 2015; Ledgerwood 2014; Volpp 2006; Volpp 2009; White 2013; Windsor (A) 1988) conducted adequate randomisation procedures (sequence generation and allocation concealment), and a further four to have followed adequate procedures for sequence generation but possibly not for allocation concealment (Alessi 2014; Gallagher 2007; Secades‐Villa 2014; Shoptaw (A) 2002). Five studies (De Paul 1994; Glasgow 1993; Hennrikus 2002; Rand 1989; Tevyaw 2009) gave insufficient detail for the integrity of the randomisation to be assessed. The remaining studies either used inadequate randomisation procedures or did not use randomisation at all. Summary assessments of the risk of bias for key items in each study are shown in Figure 1. A sensitivity analysis excluding two studies (Gallagher 2007; Paxton 1980) which did not conceal allocation did not alter the overall findings (analysis not shown).


Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Blinding
Because of the explicit mechanism of rewards, only four trials reported any attempt to blind participants, trialists or assessors (Crowley 1995; De Paul 1994; Tevyaw 2009; Volpp 2006). See the relevant 'Risk of bias' tables for details.

Drop‐outs and losses to follow up
Fifteen studies (Alessi 2014; Crowley 1995; De Paul 1994; Drummond 2014; Gallagher 2007; Giné 2010; Glasgow 1993; Halpern (A) 2015; Ledgerwood 2014; Rand 1989; Secades‐Villa 2014; Volpp 2006; Volpp 2009; White 2013; Windsor (A) 1988) treated programme drop‐outs and losses to follow‐up as continuing smokers, and conducted the analyses on an intention‐to‐treat basis, i.e. the denominator included all persons randomised at the start of the trial in their original groups. The majority of studies (11/21) were deemed to be at low risk of attrition bias or at unclear risk (9/21); only Rand 1989 was rated at high risk, with significantly more contingent participants lost than non‐contingents or controls, although reasons for drop‐out were largely unrelated to the cessation intervention, e.g. heart attack, pregnancy, relocation for work.

Outcomes

All the included studies used some form of biochemical validation procedure. Twenty tested levels of cotinine (a metabolite of nicotine) in blood, saliva or urine, either at baseline to confirm initial smoking status (Crowley 1995; Gallagher 2007; Halpern (A) 2015; Ledgerwood 2014; Secades‐Villa 2014; Shoptaw (A) 2002; Windsor (A) 1988), to validate reports of abstinence (Alessi 2014; Crowley 1995; De Paul 1994; Drummond 2014; Gallagher 2007; Glasgow 1993; Halpern (A) 2015; Ledgerwood 2014; Secades‐Villa 2014; Tevyaw 2009; Volpp 2006; Volpp 2009; White 2013; Windsor (A) 1988), among claimants of rewards (Hennrikus 2002) or among random samples of quitters (Hennrikus 2002; Paxton 1980; Paxton 1981; Paxton 1983). Giné 2010 tested all participants at six months and at one year, regardless of their claimed smoking status. Eleven trials (Alessi 2014; Crowley 1995; De Paul 1994; Drummond 2014; Gallagher 2007; Glasgow 1993; Ledgerwood 2014; Rand 1989; Secades‐Villa 2014; Shoptaw (A) 2002; Tevyaw 2009) included verification by testing breath samples for carbon monoxide (CO) levels, often to supplement the cotinine testing.

In order to test the robustness of the cessation interventions, we have included in our review only those studies which followed up participants for at least six months from the beginning of the intervention. One of the trials, however, (Rand 1989) delivered their final cessation rewards six months into the programme, which was also the end of the designated follow‐up period, thereby confounding the intervention rewards with testing at the longest follow‐up. A sensitivity analysis removing this trial made no difference to the six‐month result.

We constructed an exploratory funnel plot for the main meta‐analysis (Analysis 2.1, abstinence at longest follow‐up), but did not detect indications of publication bias, i.e. that studies with negative findings might be under‐represented.

Appropriateness of analysis
Three trials which used a cluster‐randomised design made due allowance for this in their analyses, either by testing for intraclass correlation (De Paul 1994; Glasgow 1993), or by including worksite as a random effect (Hennrikus 2002). De Paul 1994 randomised by company, but analysed by individual participant.

Crowley 1995 collapsed the three‐way groupings for the six‐month follow‐up results, since differences between groups were by then negligible, and Windsor (A) 1988 collapsed the incentive/non‐incentive groupings for analyses after the six‐week assessment, as contradictory differences had emerged between the two pairs of groups. Our own analysis of Windsor (A) 1988 and Windsor (B) 1988 compares the effect of the smoking cessation components with and without the incentives. Drummond 2014 detected no effect of spirometry on cessation, and no interaction between the lung age groups and the contingency management groups, and therefore collapsed the four‐way structure into a dichotomous CM versus no‐CM comparison. Giné 2010 recruited in three waves, across two different towns on the Philippine island of Mindanao, and revised the allocation procedure for the third wave, but analysed all participants as randomised individuals; they also collapsed the original stratification between participants who went to the bank to make their deposits and those who were visited by bank officials to make their payments.

Pregnancy studies

Randomisation

Of the nine included trials in pregnant women, six gave insufficient information to assess whether the randomisation procedure put the trial at risk of selection bias. We judged Ondersma 2012 to be at low risk of bias for sequence generation, as this was conducted by computer for the cessation intervention and by a random number generator (www.randomization.com) for the contingency management component. We also rated Tappin 2015 at low risk for selection bias; they deployed a block method (block size of four) conducted by the Glasgow Clinical Trials Unit for this purpose. We judged Higgins 2004 (the pilot for the Heil 2008 study) to be at high risk of selection bias, since the first 37 participants were allocated consecutively, while the remaining 16 were 'randomised', as part of the learning curve for study staff in trial procedures.

Blinding

Blinding of participants and personnel was generally not practicable, as recipients of rewards and those administering them would be aware that this was the intervention group. Ondersma 2012 and Tappin 2015 were the only trials that we rated at low risk of bias for blinding, since they both ensured that outcome assessment was conducted by research assistants or helpline staff, unaware of the participant's cessation intervention status.

Drop‐outs and losses to follow up

We deemed five of the pregnancy trials to be at low risk of attrition bias, and three to be at unclear risk; Donatelle 2000a was ranked at high risk of attrition bias, having lost 36% of the intervention group by two months post‐partum, and 52% of the control group, although the authors report that this level of depletion was not unusual for the antenatal clinic in question.

Outcomes

All the trials reported abstinence at or near the end of pregnancy, except for Ondersma 2012, who reported abstinence at the end of the 10‐week programme (before delivery). Seven of the trials used point prevalence abstinence as their main measure, with Donatelle 2000b and Donatelle 2002 simply citing 'abstinence', without further elaboration. Four trials (Heil 2008; Higgins 2004; Ondersma 2012; Tappin 2015) also reported measures of continuous abstinence. Five trials (Donatelle 2000a; Heil 2008; Higgins 2004; Higgins 2014; Tappin 2015) assessed smoking status post‐partum; Tuten 2012 reported the mean number of cigarettes per day at six weeks post‐partum, but not verified abstinence. We have in every case preferred the strictest definition of abstinence at the longest follow‐up. All the trials except for Donatelle 2000a validated claims of cessation using a combination of carbon monoxide monitoring and urinary or salivary cotinine. Donatelle 2000a tested salivary thiocyanate in self‐reported quitters, as well as salivary cotinine.

Appropriateness of analysis

Although we routinely prefer to conduct an intention‐to‐treat analysis (including all participants randomised), for these trials we have excluded from the denominators any predefined withdrawals due to termination or foetal demise, where these were reported. Tappin 2015 excluded three control participants from the denominator, as they had withdrawn immediately after randomisation and had withheld their data from inclusion in analyses; we have adjusted our calculations accordingly.
There were no other analysis issues in the pregnancy trials, apart from in the MISS Project (Donatelle 2002), which reported an enrolment of 298 participants, but gave results for only 293.

Effects of interventions

See: Summary of findings for the main comparison Incentives for smoking cessation; Summary of findings 2 incentives for smoking cessation in pregnant women at longest follow‐up

Mixed populations

Cessation

Details of the results for the 21 mixed‐population included studies in this review are tabulated in the Analyses section (Analysis 1.1), and are displayed graphically where data were available in Analysis 2.1.

We have conducted a meta‐analysis of 17 of the included studies (20 comparisons), grouping by longest follow‐up and by evaluation points (six to 24 months; Analysis 2.1; Figure 2). The primary result at longest follow‐up (adjusted odds ratio (OR) 1.42, 95% confidence interval (CI) 1.19 to 1.69; 7715 participants; I² = 33%) is presented in summary of findings Table for the main comparison. The six‐month result also demonstrated an unequivocal benefit for the incentives groups, with an adjusted OR of 1.72 (95% CI 1.43 to 2.08; 6945 participants; I² = 51%).The forest plots indicate that only the Halpern (A) 2015 (rewards versus usual care), Volpp 2009 and White 2013 trials demonstrated a significant advantage for incentives at all evaluation points. Although White 2013 reports a significant advantage at 14 months (longest follow‐up) for the deposit contract group over the controls (OR 2.21, 95% CI 1.16 to 4.23; 200 participants; analysis not shown), we have not included this finding in the 12‐month meta‐analysis, as abstinence at this stage was by self report only, and not biochemically or witness verified. We combined the two reward‐based arms of the Halpern (A) 2015 trial and the two deposit‐refund‐based arms, and compared each combined grouping against the usual care control condition. The rewards‐based interventions delivered significantly more instances of sustained abstinence at both the six‐ and 12‐month assessment points. For Volpp 2009, in which assessments began at three or six months after enrolment, abstinence at nine or 12 months is included in the six‐month forest plot, and abstinence at 15 or 18 months in the 12‐month forest plot. We have excluded Crowley 1995 and Hennrikus 2002 from the formal analyses, since no extractable data were available on programme participants at follow‐up; Paxton 1981 and Paxton 1983 were excluded because all the participants (experimental and control) received incentives. Shoptaw (A) 2002 and Windsor (A) 1988 are each presented with two separate comparisons, since each study evaluated incentives with two different interventions in a factorial design.

Three studies (De Paul 1994; Rand 1989; Windsor (A) 1988/Windsor (B) 1988) paid out their final reward to coincide with the six‐month follow‐up, which may have compromised that finding. A sensitivity analysis excluding these studies made little difference to the result (OR 1.73, 95% CI 1.36 to 2.19).


Forest plot of comparison: 2 Smoking cessation: mixed populations, outcome: 2.1 Smoking cessation in mixed populations.

Forest plot of comparison: 2 Smoking cessation: mixed populations, outcome: 2.1 Smoking cessation in mixed populations.

We conducted a sensitivity analysis to assess the impact of guaranteed rewards (vouchers or cash for abstinence) versus the chance of winning a reward (lotteries, prize draws), on the assumption that the level of certainty of receiving a reward might modify the success of the intervention. To this end, we identified five trials which used a lottery or a prize draw as all or part of their incentive programme (Alessi 2014; Crowley 1995; Glasgow 1993; Hennrikus 2002; Ledgerwood 2014). Two of these trials (Crowley 1995; Hennrikus 2002) provided no usable data for meta‐analysis, and could not contribute to the assessment. Removing all three trials from the main meta‐analysis (abstinence at longest follow‐up) and removing Alessi 2014 and Ledgerwood 2014 from the six‐month meta‐analysis made little difference to the effect estimates or to the CIs, and marginally increased the I² values (longest follow‐up: OR 1.43, 95% CI 1.16 to 1.76, I² = 42%; six‐month follow‐up: OR 1.63, 95% CI 1.32 to 2.02, I² = 54%). Removing Glasgow 1993 from the 12‐month meta‐analysis moved the result to borderline significance (OR 1.28, 95% CI 1.00 to 1.65, I² = 57%, P = 0.05).

Three trials (Halpern (A) 2015; Volpp 2009; White 2013) detected a significant effect of rewards or incentives on smoking abstinence at their longest follow‐up, and not confounded by rewards paid out for abstinence at that time point. In Volpp 2009 at 15 or 18 months, quit rates for the incentivised and control groups were 9.4% versus 3.6% respectively (P = 0.001). A secondary endpoint in this trial was the completion of a smoking cessation programme, for which the intervention participants received a USD 100 payment. While all participants received information about local smoking cessation services, 15.4% of the intervention group enrolled in a cessation programme, compared to 5.4% of the controls (P < 0.001); 10.8% of the incentivised group completed the programme compared to 2.5% of the controls (P < 0.001). White 2013 also demonstrated a significant benefit for incentives, but in this case the endpoint was six months, and with fewer participants than the Volpp trial (201 versus 878). The six‐month quit rate for the incentivised group was 44.3%, compared with 18.8% in the controls. Participants across the trial were paired into 'teams', with each pair earning a substantial bonus of THB 1200 (≈ USD 40, about 16% of median monthly household income) at the three‐month assessment if both members of the team were quit. By this measure, the intervention teams did significantly better, with 26.2% of individuals earning the team bonus, compared with 3.6% of the controls (P < 0.001); randomly‐assigned teams performed no differently from self‐selected pairings for this outcome.

The largest trial among the included studies is Halpern (A) 2015, a five‐arm RCT covering 2538 participants drawn from employees of a national pharmacy chain and their families and friends. The interventions were predominantly internet‐based, and participants were supported by local smoking cessation resources, self‐help booklets and, for those eligible for company health benefits, free access to behavioural support and NRT if they wished. The trial tested individual and group‐based interventions, with the financial incentives either funded entirely from study resources ('rewards') or by participants paying in USD 150 of their own money and the study topping up the balance ('deposits'). All participants stood to gain around USD 800 for sustained abstinence to six months. All four incentivised arms did significantly better than the usual‐care control group at the study's six‐month primary endpoint of validated sustained abstinence, but only the reward‐based arms maintained this advantage to the 12‐month follow‐up point. As there were no significant differences between the individual‐ and group‐based arms at any time point, we have combined the four arms into two groups (rewards versus deposits) for our analyses. The rewards groups achieved a sustained quit rate of 8.1% at 12 months, compared with 4.7% in the deposits groups. This is the only included study to have conducted head‐to‐head comparisons between a rewards‐based and a deposit‐refund‐based approach, delivering an OR at 12 months in favour of the rewards arms of 1.76 (95% CI 1.22 to 2.53; analysis not shown). We further consider these findings and implications in the Discussion section below.

Costs

We consider only the three trials (Halpern (A) 2015; Volpp 2009; White 2013) that demonstrated long‐term benefits beyond the end of the programme. Among all participants in Halpern (A) 2015 achieving sustained abstinence at six months, the average cost per quitter ranged from USD 800 to USD 890. The trialists compared this outlay, even without any deposit contribution from the participant, to the estimated USD 5186 additional cost to employers of hiring a smoker rather than a non‐smoker (Berman 2014), and rated the intervention highly cost‐effective. Volpp 2009 offered no comment on potential cost benefits of incentives programmes, other than to report an estimate of savings per quitter to an employer (USD 3400 per year, MMWR 2002). Subsequent decisions, however, on the implementation of the programme in the host company are discussed below in the Summary of main results section of the Discussion. White 2013 reported that the intervention, if rolled out to the smoking population in the study area, could translate to a decrease in smoking rates of 2% to 5%, and offered an incremental cost‐effectiveness analysis. The authors estimate that the cost per quitter from the intervention was USD 281 (95% CI USD 187 to USD 562), compared with quitting with nicotine gum (USD 1780, 95% CI USD 1414 to USD 2401) or with varenicline (USD 2073, 95% CI USD 1357 to USD 4388) in Thailand. The authors note that the intervention complies with the World Health Organization's ranking of "very cost‐effective" in Thailand, i.e. less than gross domestic product (USD 8600, purchasing power parity‐adjusted in 2011; World Bank 2012).

Harms, disbenefits

Gallagher 2007, reporting an intervention in 180 people with schizophrenia or other serious mental illness, briefly considered whether smoking cessation may have worsened the participants' psychiatric symptoms, but found no evidence for this at end of intervention or at 36‐week follow‐up, using the Brief Symptom Inventory; however, the authors caution against placing too much weight on this finding, because of low power within the study to detect such differences. Alessi 2014 reported one participant being hospitalised for alcohol‐related heart, liver and lung problems, considered by the trialists not to be associated with the intervention programme. None of the other included studies reported on any harms, unintended consequences or adverse events associated with the interventions; however, we consider in the Discussion section (below) the implications of systematic deception in participants seeking to obtain unmerited rewards for abstinence.

Pregnancy

Cessation

Details of the results at longest follow‐up (up to 24 weeks post‐partum) for eight of the nine included studies in pregnant women in this review are tabulated in the Analyses section (Analysis 1.2), and are displayed graphically where data were available in Analysis 3.1 (Figure 3). The trials demonstrated a benefit for the incentives groups over the controls, with an adjusted OR of 3.60 (95% CI 2.39 to 5.43; 8 trials, 1295 women; I² = 0%). Abstinence confirmed at or near the end of pregnancy yielded an OR of 3.79 (95% CI 2.74 to 5.25; 8 trials, 1297 women; I² = 0%; Analysis 3.2), with six of the eight trials indicating a clear benefit for the incentivised arms.


Forest plot of comparison: 3 Smoking cessation: pregnancy, outcome: 3.1 Pregnancy trials at longest follow‐up (adjusted OR).

Forest plot of comparison: 3 Smoking cessation: pregnancy, outcome: 3.1 Pregnancy trials at longest follow‐up (adjusted OR).

At longest follow‐up, three trials (Donatelle 2000a; Tappin 2015; Tuten 2012) demonstrated a clear benefit for contingent rewards over usual antenatal care. Donatelle 2000a linked each participant to a 'significant other supporter', preferably a female non‐smoker designated by the participant herself, who received reward vouchers in parallel with the participant's own success rate. The incentives group achieved a quit rate of 21.4% at two months post‐partum, compared with the control quit rate of 5.9% (P = 0.0024). Tuten 2012 conducted a scaled programme of rewards for the percentage of smoking reduction achieved over the course of the 12‐week intervention. We have used abstinence rates rather than smoking reduction rates for our analyses. This trial also included a non‐contingent reward arm, wherein the woman presenting for assessment received vouchers according to a predetermined schedule based on an earlier pilot study, and not according to her smoking status at that visit. This group did no better than the women in the control arm (no quitters in either group), while the contingent rewards group achieved a quit rate of 31% (P = 0.0204). Although the remaining five trials all reported an advantage for the incentives group over the controls, the findings in each case were compatible with no overall benefit.

The largest trial (Tappin 2015; CPIT study) with 612 women, demonstrated a high success rate for the incentivised group compared to the controls. The intervention delivered Love2shop vouchers (widely exchangeable, but not for tobacco or alcohol) to successful quitters at four weeks (GBP 50), and at 12 weeks for those already quit at four (GBP 100). All participants were invited to report their smoking status at 34 to 38 weeks gestation, and those in the intervention group proving to have quit at that stage received a GBP 200 voucher. This trial also rewarded participants for engagement in the process, i.e. giving women in the intervention group a GBP 50 voucher for attending the initial smoking cessation visit and setting a quit date, and rewarding all participants who provided information and a biological sample (if appropriate) at the primary endpoint (34 to 38 weeks) with a GBP 25 shopping voucher. The authors report that the number needed to treat for an additional beneficial outcome (based on the offer of an incentive programme rather than participation in one) is 7.2.

One trial could not be included in the meta‐analysis: the MISS Project (Donatelle 2002) reported interim results only, i.e. for 298 women from a projected total of 600. We were unable to obtain further information on final numbers, or on quit rates achieved at any point.

We have not conducted a sensitivity analysis of the certainty of attainment of the rewards (see above for mixed‐population studies), as all of the included studies used guaranteed rewards for abstinence rather than lotteries or prize draws.

The pregnancy trials tested particular issues about how the incentives might work for pregnant smokers:

1. Did the size of the reward make a difference?
The MISS Project (Donatelle 2002) was designed to address this question, setting up identical reward schedules for the two experimental groups, but with one group receiving USD 25 per month for validated abstinence and the other USD 75 per month. This is an interim study report, and we have been unable so far to obtain any data on the findings of this trial.

2. Did contingent rewards (usually increasing with prolonged abstinence) work differently from a schedule of non‐contingent fixed payments?
Four trials addressed this question (Heil 2008; Higgins 2004; Higgins 2014; Tuten 2012); all four favoured conditional payments, with an OR of 6.26 (95% CI 2.35 to 16.68; Analysis 3.3; 278 participants; I² = 19%); a key condition of this finding is that the scaled payments were given only as a reward for validated abstinence, while the fixed payments were guaranteed provided that the participant attended and gave a biological sample, irrespective of her smoking status. None of the included trials compared an incremental with a fixed schedule, but with both payable only for validated abstinence, i.e. testing the role of variable rewards rather than contingency.

3. Did 'front‐loading' of the reward schedule, to counteract the greater likelihood to relapse in the early stages of quitting, work better than the standard incremental schedule?
Higgins 2014 addressed this question, by allocating to the Revised Voucher Group an additional loading of USD 296.25 early in the trial; the participant had to achieve a breath sample of carbon monoxide under 4 parts per million daily in the first week (i.e. USD 18.75 on day 1 rising to USD 33.75 by day 5), plus testing cotinine‐negative on the second Monday, and thereafter twice a week up to week six. The second test each week increased by USD 15.50 if it was negative following a negative first sample that week. The odds in favour of the revised (front‐loaded) schedule versus the standard schedule were 1.17 (95% CI 0.35 to 3.84; 79 participants; Analysis 3.4), indicating that the revised schedule made little difference to the chances of quitting.

4. Did participant‐initiated verification of abstinence work better than a researcher‐controlled process?
Ondersma 2012 differed significantly from the other included trials in pregnancy, by requiring minimal personnel or financial resources; the 5As‐based cessation intervention was computer‐delivered, and the contingency management component was initiated by the participant and moderated through a website. Participants could request a confirmatory urine test at their normal prenatal visit, rather than making multiple ad hoc visits for monitoring and testing, and could take as many computerised tests as they wished, although they were only eligible for five reward vouchers over the course of the programme. Collapsing the two CM arms and comparing them with the two cessation arms combined gave an OR of 1.70 (95% CI 0.60 to 4.82; Analysis 3.5), indicating no clear benefit to either group.

Costs

Heil 2008, while confirming that they had not conducted a cost‐benefit analysis, reported that the average cost of the incentives per participant was USD 334. The remaining included studies in pregnancy did not report on cost effectiveness. However, in their review of financial incentive programmes in pregnant women (covering five of the eight pregnancy trials included in our review), Higgins 2012 cited a report from the National Institute for Health and Clinical Excellence (NICE) which had evaluated the interventions covered by a previous version of the Chamberlain 2013 Cochrane review (Lumley 2009), and had concluded that financial incentive interventions produced the highest net cost benefit, at GBP 2261 (USD 3482) after accounting for the cost of the intervention.

Tappin 2015 reports that the short‐term incremental cost per quitter was GBP 1127, and the longer‐term cost for each quality‐adjusted life year gained was GBP 482. The latter is well below the NHS lower threshold of GBP 20,000, designated by the National Institute for Health and Care Excellence (NICE 2013) as an indicator of cost effectiveness (Tappin 2015b).

Harms, disbenefits

None of the included pregnancy trials reported on harms or unintended consequences of the interventions, although Tappin 2015 offered some evidence on the likelihood of the participants 'gaming' to receive unmerited rewards. We consider this further in the Discussion section below.

Discussion

available in

Summary of main results

Mixed‐population studies

It is notable that the two studies delivering the most robust and unequivocal evidence for the benefits of incentives for smoking cessation are the largest of the included studies, with 2538 and 878 participants respectively, delivering the most substantial rewards, and are conducted and reported to high methodological standards. However, since both trials enrolled employees of large American companies, who were predominantly white and enjoyed relatively high levels of education and income, their success may not be readily generalisable to other populations of smokers, with different regional, socio‐economic and ethnic mixes.

Three mixed‐population included studies in this review offer evidence that incentives can improve long‐term smoking cessation, whether conducted in the community, in healthcare settings or in the workplace. Halpern (A) 2015, the largest of the included studies, offers strong evidence that substantial financial rewards (USD 800 for sustained abstinence at six months) deliver significantly higher quit rates than usual care. This was demonstrated both for study‐funded rewards and for incentives funded partly by the participants themselves, although the deposit‐based groups' quit rate no longer reached statistical significance at 12 months (longest follow‐up).

Volpp 2009 speculates that their study population was large enough to detect an effect, and that their rewards were substantial enough (a total of USD 750 available for completion of a smoking cessation programme and sustained abstinence at 9 or 12 months) to consolidate the target behaviour change. Six months after the final payment, the incentivised group maintained a higher quit rate than the control group. While the findings of this trial and of Halpern (A) 2015 are encouraging, such a paradigm may only work in communities or situations where independent and well‐resourced smoking cessation services already operate. It is also noteworthy that, despite the results of this large and influential study, the company in which it was conducted have declined to implement the findings. Negative feedback from focus groups within the workforce, combined with a reluctance by the finance department to take on the costs of running such a programme, meant that in 2010 the USD 750 incentive model was replaced by an annual USD 620 surcharge imposed on continuing smokers. This penalty could be waived for one year, provided that the smoker completed a free‐of‐charge cessation course. Biochemical verification was also abandoned, being perceived as intrusive and too complicated to administer. The authors point out that the switch from 'carrot' to 'stick' was not an evidence‐based policy shift, and highlights the disjunction between data‐driven decisions and those based on culture and belief (Volpp 2014).

White 2013 used community‐based health workers to support smokers attempting to quit in a region of Thai villages, using a deposit‐refund intervention. The six‐month success rates were impressive, at 44.3% for the intervention group compared with 18.8% among the controls; however, the unusually high quit rate for the control group suggests that this trial‐naïve population may have represented 'low‐hanging fruit' (easy quitters), and that these findings are not readily generalisable to areas with longstanding and established tobacco control programmes.

The deposit‐refund trials merit particular attention, in the light of the real‐world constraints that have emerged from the Volpp 2009 study. If implementation of an incentives programme is compromised by the costs incurred, then the model of participants depositing and forfeiting their own money is likely to be more attractive to employers and institutions seeking affordable behaviour change interventions. However, low uptake rates compared with reward‐based interventions may limit the appeal and efficacy of such programmes (Volpp 2014); White 2013 reports a participation rate of 10.5% among eligible smokers, while Giné 2010 reports a rate of 10.6%. The three Paxton trials are of limited relevance here, as they were conducted under the auspices of the UK National Health Service, on condition that participants motivated to quit but unable or unwilling to pay the required deposit must still be allowed to participate in the cessation programme. Uptake in this case was therefore not a marker of acceptability.

Halpern (A) 2015 is the only trial in this review to have directly compared interventions funded entirely from trial resources to those funded partly by the participants themselves (USD 150). As discussed above, uptake rates proved to be a barrier, with 90% of those offered rewards accepting the intervention compared to only 13.7% of those required to put up a deposit, obliging the trialists to develop an adaptive model of randomisation in order to populate the deposit‐based arms. On an intention‐to‐treat basis, the rewards arms consistently delivered significantly more quitters than the deposit arms at all time points; however, in instrumental variable analysis which accounts for different rates of uptake (equivalent to a per protocol analysis), among participants prepared to accept either intervention the deposit arms outperformed the rewards arms, with six‐month quit rates of 53.4% and 17.1% respectively, and 12‐month quit rates of 18.5% and 8.8% respectively. This would suggest that although it may be more difficult to recruit smokers into deposit‐based programmes, once they are in they are strongly committed to the process and can achieve high quit rates.

Many of the early studies were underpowered and of variable quality. A sensitivity analysis removing those studies conducted before 2000 made little difference to the findings at longest follow‐up (OR 1.65, 95% CI 1.32 to 2.07), but shifted the 12‐month result to one of marginal statistical significance (OR 1.47, 95% CI 1.09 to 1.99). One study (Rand 1989) confounded the final delivery of cessation rewards with the final follow‐up assessment. Similarly, although De Paul 1994 reported significantly higher cessation rates for the incentives group at the six‐month assessment, this evaluation coincided with the final phase of the rewards programmes. At later follow‐up points in this study all such differences had disappeared. Encouraging early quit rates at 30 days (Volpp 2006) and at 21 days (Tevyaw 2009) dwindled to non‐significant differences by the six‐month follow‐ups. Although Gallagher 2007 achieved significantly different CO‐validated quit rates, the cotinine‐validated quit rates did not achieve clinically significant differences, suggesting that while some participants could achieve temporary abstinence for their clinic visit, the more rigorous urinary cotinine test did not indicate abstinence sustained beyond a few hours.

Glasgow 1993 reported one‐year cessation rates for HIP participants more than double those of non‐registrants (22.1% versus 9.4%, P < 0.005), but this difference had become non‐significant at the two‐year follow up; and again, the one‐year evaluation was very close to the final lottery draw for HIP participants and may well have been influenced by that proximity. Shoptaw (A) 2002 reported the same long‐term pattern of relapse, with significant benefit to the contingency management groups in the first three months rapidly vanishing over the nine‐month post‐programme follow‐up period. One recent study (White 2013) reported impressive cessation rates for their deposit‐refund programme at six months, which appeared to be sustained to a 14‐month follow‐up, but the latter finding must be considered as unreliable, since it was based on self report, without biochemical or witness validation. Halpern (A) 2015 noted a halving of the validated quit rates between six and 12 months, although self‐reported 12‐month rates remained comparable with the six‐month levels. The overall picture that emerges from these trials is that incentives work while they are in place, but that the typical relapse pattern is likely to re‐establish itself once they are removed.

Pregnancy studies

The eight trials in 1295 women contributing to the meta‐analyses confirmed the efficacy of incentives at longest follow‐up, and at or around the end of pregnancy.

Our results are slightly at variance with those reported in the Pregnancy and Childbirth review (Chamberlain 2013), as their primary outcome was abstinence around the end of pregnancy, while ours is at longest follow‐up, i.e. post‐partum where available. We also dropped prespecified withdrawals (termination, foetal demise) from the denominator, and we include four trials which do not appear in the Chamberlain 2013 meta‐analysis. Their main result (abstinence around the end of pregnancy) covers 426 women, giving a risk ratio of 3.09 (95% CI 1.34 to 7.15); this converts (for direct comparison) to an odds ratio (OR) of 4.13 (95% CI 1.50 to 11.40), using a random‐effects model (Chamberlain 2013 comparison 11, Analysis 1.4). Our own analysis of the same (secondary) outcome delivers an OR (fixed‐effect) of 3.79 (95% CI 2.74 to 5.25) in 1295 women (Analysis 3.2). Conducted as a random‐effects model, this OR becomes 3.61 (95% CI 2.60 to 5.02), i.e. still a clear finding in favour of incentives, but with three times the number of participants and a narrower confidence interval.

Incentives
The use of tangible rewards will always be a trade‐off between maximising participation and attracting smokers who are motivated more by the rewards than by the wish to stop smoking. The type and scale of the incentive has therefore been considered a critical element in the design of a cessation programme, although from the perspective of this review, which is primarily concerned with sustained or permanent cessation, the type and scale of the incentives may be less significant than the negative effects of removing them altogether. The incentives varied considerably across the studies, including cash prizes, vouchers for goods and services, prize draws, and combinations of cash and lottery tickets. Both Volpp 2009 and Halpern (A) 2015, which between them represent more than 40% of the study population for this review, provided substantial cash payments both for compliance and for prolonged abstinence, and demonstrated a sustained beneficial effect beyond the expiry of the payment schedule. White 2013, which returned the participants' own money to them provided that they were quit within three months, also demonstrated lasting benefits, but suffered from relatively low levels of uptake. All participants in this trial stood to benefit from substantial bonus payments if they could collaborate with their team partner to achieve joint cessation, but the control teams failed in this compared to the intervention teams, suggesting that social support was not of itself sufficient to boost cessation rates.

Deception
Seven studies tested for smoking status at baseline, with cotinine or exhaled carbon monoxide (CO). Five were clinic‐based, with three (Crowley 1995; Gallagher 2007; Shoptaw (A) 2002) treating disease‐specific populations of smokers; in all five studies, biochemically‐verified proof of smoking was a requirement for enrolment into the trial. One worksite‐based study (Windsor (A) 1988) also screened all participants for salivary thiocyanate as a condition of entry. Halpern (A) 2015 was the only study to monitor eligibility by testing after entry, in a 5% sample of enrolled participants who were paid USD 100 for supplying a cotinine assay. Of those asked to submit a baseline sample, nine (6%) returned a negative assay and 21 (14%) did not return a sample, suggesting that up to 20% of participants could have been non‐smokers. The rates were comparable across all arms of the trial, and sensitivity analyses adjusting for this possible level of deception made no difference to effect estimates.

All the included studies in this review used some form of biochemical verification to confirm the smoking status of those claiming abstinence. While this procedure is now the recommended gold standard for good trial design (Benowitz 2002), it is particularly important that quitters in an incentives‐based trial are shown to be truly abstinent at the evaluation points. Eligibility for cessation rewards depended in all the included studies upon biochemical confirmation of the claim of abstinence. Volpp 2006 addressed the likelihood of participants modifying their smoking behaviour in anticipation of being contacted for follow‐up assessment and cotinine testing, and concluded that this was unlikely, since although participants knew that they would be biologically tested, they were unaware of how long nicotine metabolites would be detectable or the exact date on which they would be checked.

Two of the studies in this review reported a good correspondence between claims of abstinence and their biochemical verification, with Windsor (A) 1988 noting 100% agreement for more than 600 saliva thiocyanate samples, and De Paul 1994 a 95% agreement. The Paxton studies took random urine samples to deter false reporting, and also occasionally cross‐checked smoking status with family members or friends. They reported that levels of deception were 'very low', and attributed this to having warned participants in advance about the random biochemical checks.

Three studies which targeted high‐risk smoking groups went to some trouble to control for possible levels of deception. Crowley 1995, dealing with moderately‐ill COPD patients who had been co‐opted into the programme, anticipated a measure of deception, and calculated an expected ratio of CO divided by cigarettes smoked. This confirmed a greater disparity between cigarettes smoked and numbers reported among the non‐verified self‐report group and the control group than among the intervention group, who were rewarded only for verified abstinence. Shoptaw (A) 2002, dealing with methadone‐maintained drug abusers with high levels of smoking, reported similar findings between self‐reported abstinence and its biochemical verification, but cautioned that it was possible that participants had found ways of subverting the breath‐testing schedule. Against this possibility, however, was the fact that subjects had averaged only 44% of the available prize vouchers for abstinence, suggesting that any subversion had not been particularly successful. Gallagher 2007, dealing with smokers with schizophrenia or other serious mental disorders, found considerable disparities between quit claims validated by CO breath samples (confirming abstinence for a few hours) and those validated by urinary cotinine (a few days of abstinence). This need not indicate deception, but suggests that the achieved abstinence for which the rewards were being claimed was not robust or sustained (SRNT 2002).

The two studies which found a striking disparity between self‐reported abstinence and biochemical verification of the claims were both large, worksite‐based, cluster‐randomised trials, which followed their participants for 24 months. Hennrikus 2002 reported a 33.6% mismatch between self report and confirmation at 24 months, and Glasgow 1993 a 27% mismatch at the12‐month evaluation. Both trials had 'sprung' the biochemical validation requirement on the quitters at follow‐up. The clear discrepancies suggest that people responding indirectly (not face‐to‐face) to a question about their smoking, and not expecting to have their answer checked, may be significantly more likely to say what they think the questioner wants to hear.

Recent work on incentives for pregnant smokers trying to quit has directed attention to the risks of deception or 'gaming', particularly the likelihood of delaying a quit attempt to coincide with a rewards programme, and the likelihood of misrepresenting smoking status, either to gain admission to an incentive programme or to receive unmerited rewards for abstinence (Marteau 2013). Ierfino 2015, in a cohort of 239 pregnant smokers, found no evidence of gaming to enter the programme, but detected a 4% level of deception to win vouchers for abstinence.

Tappin 2015 used residual routine blood samples (i.e. taken for non‐study‐related purposes) collected from the final 200 women enrolling in their study, to cross‐check the smoking status of self‐reporting quitters and of women with whom contact had been lost. Of 46 women in the incentive group who could not be reached for the 34/38‐week assessment, residual blood tests for 10 of them confirmed in all cases that the women were currently smoking; similarly, blood tests for three 'lost' control participants confirmed current smoking. Although 30 women in the incentive group who could not be contacted at the primary endpoint reported that they were abstinent, residual samples for two of them indicated that both were smoking. Residual bloods were also available for 18 of the 69 intervention women who self‐reported abstinence at 34/38 weeks, and had this confirmed by saliva or urinary cotinine; 78% of these samples (14/18) confirmed their non‐smoking status. Similarly, five residual samples for the 26 control participants with confirmed abstinence at 34/38 weeks corroborated 80% (4/5) of the results. While this suggests some overestimation of the true quit rates, the level of deception appeared to be similar across both groups, and confirmed the veracity of 80% of the self‐reported quitters.

Overall completeness and applicability of evidence

We have followed standard Cochrane methods to identify and evaluate the studies contributing to this review, and are confident that we have not missed any significant trials. We have sought missing or incomplete data, and have contacted authors where possible to clarify our interpretation of their work. Our inclusion of trials of interventions in pregnant smokers has extended the applicability of this review, in line with recent trends in public health approaches to incentivising behaviour change.

Quality of the evidence

Mixed‐population studies: We rated the overall quality of the evidence in this group of trials as low (see summary of findings Table for the main comparison), suggesting that further research is very likely to have an important impact on our confidence in the estimate of effect, and may change the estimate. This is largely attributable to the methods and assumptions of the older studies, and their data presentation, compared with prevailing standards in trial methodology. Eight of the trials delivered imprecise results, as percentages or graphs without a clear statement of denominators or event counts. Eleven of the 21 studies were deemed to have conducted adequate random sequence generation, with seven of them also applying adequate concealment of allocation; all but one of these trials (Windsor (A) 1988) were conducted in 2002 or later.

Pregnancy studies: The included trials covering pregnant smokers are rated as being of moderate quality (summary of findings Table 2), suggesting that further research is likely to have an important impact on our confidence in the estimate of effect, and may change the estimate. All the trials were conducted within the last 15 years, and have benefited from advances in trial methodology and reporting expectations. However, only two trials (Ondersma 2012; Tappin 2015) were deemed to have reported adequate randomisation procedures, while another (Higgins 2004) was rated at high risk of bias for failing to implement a consistent method of sequence generation. Ondersma 2012 is a notable inclusion within this group, as the trialists sought to implement a pragmatic and highly automated approach to delivering an incentive‐based intervention, not requiring multiple ad hoc visits or tests, and placing the onus upon the participant to verify and report abstinence and to initiate claims for rewards. The Tappin 2015 trial may also be viewed as a benchmark study, for the scale of recruitment (612 women) and the robustness of its findings. Furthermore, the trialists' use of residual blood samples to complement the ad hoc cotinine samples lends some support to the Russell Standard recommendation (West 2005) to assume that participants lost to follow‐up should be counted as continuing smokers.

Agreements and disagreements with other studies or reviews

A number of systematic reviews addressing incentives for smoking cessation have been published in recent years. Two reviews address smoking among other public health interventions based on incentives: Diepeveen 2013 considered the public acceptability of smoking cessation interventions (in 110 studies out of 200 covering various interventions), and reported strong support for bans and restrictions on workplace smoking, in indoor public places, restaurants and shopping centres, and for focusing on children and young people. Giles 2014 evaluated 16 studies of incentivised health behaviour change, 10 of which focused on smoking cessation, five on vaccination programmes and one on physical activity. Using most of the same studies as in our review, they demonstrated a clear benefit for smoking cessation up to six months follow‐up (RR 2.48, 95% CI 1.77 to 3.46), but a less clearcut advantage at later follow‐up (more than six months) (RR 1.50, 95% CI 1.05 to 2.14) with high heterogeneity (I² = 76%). Two further reviews were confined to smoking cessation only: Leeks 2010 demonstrated a 4.4% benefit at a median of 12 months follow‐up for worksite‐based cessation programmes which included incentives or competitions, but could not isolate the relative efficacy of the incentive component from other aspects of the programmes. Sigmon 2012b explored incentive programmes within particular high‐risk population subgroups, including substance abusers, adolescents and young adults, and people diagnosed with pulmonary disease, and also flagged up the use of shaping procedures for hard‐to‐treat smokers, the promise of developing technologies for delivery of the intervention, and varying the scale of the incentive.
Kane 2004 may now be of limited interest, as it includes only trials conducted up to 2002 and has largely been superseded. It covered smoking (11 out of 47 included studies) as well as screening, immunisation, and STD/HIV prevention programmes, and interventions for exercise, weight loss, cardiovascular health and breastfeeding.
Higgins 2012 summarised a series of six trials of incentives in pregnant smokers, conducted by two US‐based research groups and particularly addressing birth outcomes. Their review is compatible with our own findings, and states that such interventions "hold promise" as a mechanism for increasing cessation rates in this population of smokers.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Figures and Tables -
Figure 1

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Forest plot of comparison: 2 Smoking cessation: mixed populations, outcome: 2.1 Smoking cessation in mixed populations.
Figures and Tables -
Figure 2

Forest plot of comparison: 2 Smoking cessation: mixed populations, outcome: 2.1 Smoking cessation in mixed populations.

Forest plot of comparison: 3 Smoking cessation: pregnancy, outcome: 3.1 Pregnancy trials at longest follow‐up (adjusted OR).
Figures and Tables -
Figure 3

Forest plot of comparison: 3 Smoking cessation: pregnancy, outcome: 3.1 Pregnancy trials at longest follow‐up (adjusted OR).

Study

Denominator

Abstinence

Time point

Biological criterion

Quit rate

Stat sig?

Other outcomes

Comment

Alessi 2014

24 (CM)

21 (control)

7‐day PPA

24 weeks

CO < 6 ppm

cotinine < 30 ng/ml

12.5% (I)

23.8% (C)

No

% reduction in cpd; self efficacy

Raw data supplied by the author

Crowley 1995

36/49 randomised

24 hr PPA

6 months

CCO < 10ppm

N = 5/36

Chi² N.S.

Mean CO values (reduction)

Groupings were collapsed at follow‐up; of 49 at baseline, 9 dropped out, 4 had moved or died.

De Paul 1994

281 (I)
280 (SH)

PPA

24 months

CO < 9ppm

13.2% (I)
10.3 %(SH)

N.S.

PP, ITT and continuous quit rates reported at all time points

Comparison confined to I and SH groups in this review

Drummond 2014

50 (UC/LA)

50 (CM x 2)

7 days PPA

6 months

cotinine, eCO

UC/LA 1/50

CM 3/50

N.S.

CO values, Fagerstrom score, N of visits wanting to quit, trying to quit, reporting cessation, eCO‐confirmed quitting

Groupings collapsed, as lung age alone or conbined with CM produced no quitters

Gallagher 2007

60 (CR)
60 (CR+NRT)
60 (Cont)

PPA

36 wks

CO ≤ 10 ppm
SCN < 15ng/mL

7% (CR)
5% (Cont)
(based on SCN)

N.S.

CO‐validated rates higher, i.e. 37% (CR), 8% (Cont).
Reduction, psychiatric symptoms.

CR+NRT group not used in our comparison.

Giné 2010

781 (CARES)

603 (Cards)

616 (Control)

PPA

12 months

NicCheck strip (urinary cotinine) = 0

11% (CARES)

9.3% (Cards)

8.9% (Cont)

@ 0.05

6m PPA: CARES 9.7%, Cards 10%, Control 8.3%.

Cost effectiveness: USD 700 per quitter

12m assessment was 'sprung' on pts

Glasgow 1993

344 (I)
426 (C)

7 days
abstinence

2 years

CO ≤ 9ppm
Cotinine ≤ 25ng/mL

14.2% (I)
11.5% (C)

N.S.

Incentives had a sig. effect (P < 0.03) on less educated subjects (18.6% vs 8.8% @ 2yr 'probably chance').
Compared participants with non‐participants (22.1% vs 9.4$ @ 1yr, p<0.005; 21.3% vs 16.8% @ 2yr, N.S.)

27% of all abstinent claims could not be biochemically verified

Halpern (A) 2015

498 (Ind R)

519 (Coll R)

582 (Ind D)

471 (Com D)

468 (UC)

sustained

12 months

Cotinine < 10 ng/ml

anabasine/anabitine < 3 ng/ml

7.4% (Ind R)

8.7% (Coll R)

3.6% (Ind D)

6.2% (Com D)

3.4% (UC)

vs UC:

0.007

0.001

0.94

0.052

Sustained verified abstinence @ 14 days, 30 days, 6m;

Self‐reported abstinence at 12m;

per protocol analyses;

Uptake rates of assigned intervention

No differences between individual and group interventions, so both reward arms versus both deposit arms combined for analysis

Hennrikus 2002

407

7 days PP

24 months

Saliva from 149 random sample of quitters @ 24m.

19.4%
(cohort survey)

Not stated

Cohort prevalence and cessation rates (PP and continuous)
Recruitment rate
Programme format

Programme registrants' outcomes not available

Ledgerwood 2014

ECM: 36

TCM: 28

SC (Control): 17

PPA

6 months

Urinary cotinine ≤ 100 ng/mL

CO ≤ 6 ppm

4/64 (TCM+ECM)

1/17

N.S.

Prize money won; 81% CM pts earned prizes (median USD 120.56); Differences between TCM and ECM in wk 1 non‐significant

Both CM arms combined for analysis

Paxton 1980

33 (I)
27 (C)

At least 7 days

6 months

Random urine samples, but
none after 4m

?43% (I)
?45% (C)

N.S.

Weight and lung function monitored for 4 months

Quit rates presented as graphic % only

Paxton 1981

33 (D1)
27 (D2)
23 (D3)

At least 7 days

6 months

Random urine samples, but
none after 4m

?35% (D1)
?38% (D2)
?42% (D3)

N.S.

Weight and lung function monitored for 4 months

Quit rates presented as graphic % only.
Group D1 is the same data as Paxton 1980 Group I

Paxton 1983

60 (2LA)
49 (4LA)
31 (4LT)
19 (4CT)

No smoking since last measured

12 months

Random urine samples, but
none after 4m

26.7% (2LA)
26.5% (4LA)
38.7% (4LT)
36.8% (4CT)

N.S.

'thinning' deposit repayments
Participation rate
Effects of cumulative vs lump‐sum deposits

Rand 1989

17 contingent
16 non‐cont
14 control

Continuous

6 months

CO<=11ppm

1/17 contingent
1/16 non‐cont
0/14 control

N.S.

Numbers of abstinent CO samples and missed samples

Pairwise comparisons gave sig diffs @ 11ppm, but not @ 8ppm

Secades‐Villa 2014

43 CBT + CM

49 CBT

continuous

6 months

CO < 4 ppm;
Cotinine < 80 ng/ml

17/43 CM

13/49 CBT

N.S.

Treatment retention; % attending all sessions for 6m

Shoptaw (A) 2002

42 (P)
42 (RP)
43 (P+CM)
47 (P+RP+CM)

PPA

12 months

CO<=8ppm
Cotinine
<30ng/mL

4/36 (P)
2/33 (P+RP)
2/35 (P+CM)
1/38 (P+RP+CM)

N.S.

Treatment group and cocaine and opiate abuse

Quit rates supplied by authors.
P group relapsed more slowly than other groups (P = 0.0017)

Tevyaw 2009

28 (CM+MET)
27 (CM+REL)
27 (NR+MET)
28 (NR+REL)

7‐day PPA

6 months

CO<5ppm
Cotinine <15ng/mL

1/55 (CM)
3/55 (NR)

N.S.

Attendance, sample returns.

Volpp 2006

92 (I)
87 (C)

7‐day PP

6 months post‐completion (˜7.5m post‐quit date

Urinary cotinine <500 ng/mL

6/92 (I)
4/87 (C)

N.S.

Enrolment, attendance, programme completion

Denominators could be Ns enrolled (I:38, C:17). No quitters outside the enrollers.

Volpp 2009

436 (I)
442 (C)

Prolonged

15 or 18 months

Salivary cotinine < 15 ng/ml
or urinary cotinine < 2 ng/ml

41/436 (I)

16/442 (C)

< 0.001

Enrolment in SC course, completion of SC course

15 ‐ 18 months results shown in 12‐month forest plot

White 2013

131 (I)

69 (C)

7‐day PPA

6 months

Urinary cotinine

58/131 (I)

13/69 (C)

< 0.0005

PPA @ 3m (verified), 14m (self‐report).

Relative success of teams vs individuals? Yes

Choosing team partner vs random assignment? No

Did text messages help? No

Cost effectiveness; No figures given.

Windsor (A) 1988

95 (A)
94 (B)
95 (C)
94 (D)

Continuous

12 months

SCN<=
100 ng/mL

≃6% (A)
≃18% (B)
≃5% (C)
≃10% (D)

Not reported

Social enhancement vs self‐help manual (+/‐ incentives) gave a continuous quit rate of 14.4% at 12m, vs 5.8%.

Incentives comparison was abandoned @ 6w

Figures and Tables -
Analysis 1.1

Comparison 1 RESULTS OF INCLUDED STUDIES, Outcome 1 RESULTS TABLE: Mixed populations.

Study

Denominator

Abstinence

Time point

Biological criteria

Quit rate

Stat sig?

Other outcomes

Comment

Donatelle 2000a

112 (I)

108 (C)

7‐day PPA

8m gestation

Salivary cotinine < 30 ng/ml
Thiocyanate < 100 ug/ml

34/105; 32% (I)

9/102; 9% (C)

Chi² = 18.4; P < 0.0001

None stated

Differential losses to follow‐up; (I) 32% at 8m, vs (C) 51.5%.

Donatelle 2000a

112 (I)

108 (C)

7‐day PPA

2m post‐partum

Salivary cotinine < 30 ng/ml
Thiocyanate < 100 ug/ml

22/103; 21% (I)

6/102; 5.9% (C)

Chi² = 11;
P < 0.0009

None stated

Differential losses to follow‐up; (I) 36% at 2m pp, vs (C) 52%.

Donatelle 2000b

67 (E1)
59 (E2)
60 (C)

"biochemically confirmed abstinence"

end of pregnancy

Salivary cotinine < 30 ng/ml
Monthly CO < 5 ppm.

19% (E1)

22% (E2)

12% (C)

Not stated

None stated

Very little information available.

Donatelle 2000b

N/A

N/A

N/A

N/A

N/A

N/A

N/A

Donatelle 2002

102 (E1)

96 (E2)

95 (C)

Self report (telephone call)

8m gestation

Salivary cotinine < 30 ng/ml
Monthly CO < 5 ppm.

N.S.

Not stated.

High vs low incentives;

cost per quitter

Results are interim analysis only, based on 298 enrolled; target was 600.
No further information available

Donatelle 2002

N/A

N/A

N/A

N/A

N/A

N/A

N/A

Heil 2008

37 (I)

40 (C)

PPA

End of pregnancy;

Urine cotinine < 80 ng/ml

CO ≤ 6 ppm

15/37; 41% (I)

4/40; 10% (C)

P = 0.003

Foetal growth

Heil 2008

antepartum CA;

24 wks post‐partum

Urine cotinine < 80 ng/ml

CO ≤ 6 ppm

3/37; 8% (I)

1/40; 3% (C)

N.S.

Baby health

Total voucher earnings

Higgins 2004

30 (I)

23 (C)

7‐day PPA

End of pregnancy

Urine cotinine < 80 ng/ml

CO ≤ 6 ppm

11/30 (I)
2/23 (C)

P < 0.05

Mean voucher earnings

This was the pilot study for Heil 2008

Higgins 2004

30 (I)
23 (C)

7‐day PPA

24 wks post‐partum

Urine cotinine < 80 ng/ml

CO ≤ 6 ppm

8/30 (I)

0/23 (C)

P < 0.05

Higgins 2014

44 (RCV; E1)

44 (CV; E2

42 (NCV; C)

7‐day PPA

28 wks gestation

Urinary cotinine ≤ 80 ng/ml

CO < 4ppm or 6 ppm

18/40; 45% (E1)

14/39; 36% (E2)

7/39; 18% (C)

N.S.

Foetal growth
Birth outcomes

Higgins 2014

44 (RCV; E1)

44 (CV; E2

42 (NCV; C)

7‐day PPA

24 wks post‐partum

Urinary cotinine ≤ 80 ng/ml

CO < 4ppm or 6 ppm

7/40; 18% (E1)

6/39; 15% (E2)

3/39; 8% (C)

Foetal growth

Birth outcomes

Ondersma 2012

26 (E1)

28 (E2)

30 (E3)

26 (C)

7‐day PPA

30‐day CA

7‐day PPA

Urinary cotinine ≤ 100 ng/ml

CO < 4 ppm

6/23: 26% (E1)

2/22: 10% (E2)

5/26: 19% (E3)

1/23: 4% (C)

E1 P < 0.05

Ondersma 2012

42 (E1)

28 (E2)

32 (C)

PPA

12 wks

CO < 4 ppm

Urine sample (for cocaine)

13/42; 31% (E1)

0/28; 0% (E2)

0/32; 0% (C)

Tappin 2015

306 (I)

306 (C)

"even a puff" in past 2 wks

"even a puff" in past 4 wks

< 5 cigs in past 8 wks

4 wks

12 wks (if quit at 4)

34‐38 wks gest (all pts)

CO < 10 ppm

Cotinine: Urine 44.7 ng/ml; saliva 14.2 ng/ml

69/306 (I)

26/303 (C)

P = 0.0000

Adverse events

engagement

birth weight

cost effectiveness

3 controls dropped out after randomisation ‐ not included in denominators

Tappin 2015

306 (I)

306 (C)

still quit or < 5 cigs for since TQD

6m post‐natal (for 34/38‐wk quitters)

Cotinine: Urine 44.7 ng/ml; saliva 14.2 ng/ml

47/306 (I)

12/303 (C)

P = 0.0000

3 controls dropped out after randomisation ‐ not included in denominators

Tuten 2012

42 (E1)

28 (E2)

32 (C)

Self‐reported 24‐hr PPA

6 wks pp

none

13/42; 31% (E1)

0/28; 0% (E2)

0/32; 0% (C)

N.S.

Mean cpd

Tuten 2012

42 (E1)

28 (E2)

32 (C)

Self‐reported 24‐hr PPA

6 wks pp

none

13/42; 31% (E1)

0/28; 0% (E2)

0/32; 0% (C)

N.S.

Mean cpd

Abstinence not reported for this time point

Figures and Tables -
Analysis 1.2

Comparison 1 RESULTS OF INCLUDED STUDIES, Outcome 2 RESULTS TABLE: Pregnant women.

Comparison 2 Smoking cessation: mixed populations, Outcome 1 Smoking cessation in mixed populations.
Figures and Tables -
Analysis 2.1

Comparison 2 Smoking cessation: mixed populations, Outcome 1 Smoking cessation in mixed populations.

Comparison 3 Smoking cessation: pregnancy, Outcome 1 Pregnancy trials at longest follow‐up (adjusted OR).
Figures and Tables -
Analysis 3.1

Comparison 3 Smoking cessation: pregnancy, Outcome 1 Pregnancy trials at longest follow‐up (adjusted OR).

Comparison 3 Smoking cessation: pregnancy, Outcome 2 Pregnancy trials: abstinence at end of pregnancy.
Figures and Tables -
Analysis 3.2

Comparison 3 Smoking cessation: pregnancy, Outcome 2 Pregnancy trials: abstinence at end of pregnancy.

Comparison 3 Smoking cessation: pregnancy, Outcome 3 Pregnancy: contingent rewards vs guaranteed payments.
Figures and Tables -
Analysis 3.3

Comparison 3 Smoking cessation: pregnancy, Outcome 3 Pregnancy: contingent rewards vs guaranteed payments.

Comparison 3 Smoking cessation: pregnancy, Outcome 4 Pregnancy: front‐loading vs incremental payments.
Figures and Tables -
Analysis 3.4

Comparison 3 Smoking cessation: pregnancy, Outcome 4 Pregnancy: front‐loading vs incremental payments.

Comparison 3 Smoking cessation: pregnancy, Outcome 5 Pregnancy: Participant‐initiated CM.
Figures and Tables -
Analysis 3.5

Comparison 3 Smoking cessation: pregnancy, Outcome 5 Pregnancy: Participant‐initiated CM.

Summary of findings for the main comparison. Incentives for smoking cessation

Incentives for smoking cessation

Participant or population: Adult smokers
Settings: mixed
Intervention: Incentives

Outcomes

Illustrative comparative risks* (95% CI)

Relative effect
(95% CI)

No of Participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

Control

Incentives

Smoking cessation in mixed populations ‐ longest follow‐up
Follow‐up: 6 ‐ 24 months

84 per 1000

112 per 1000
(95 to 132)

OR 1.42
(1.19 to 1.69)

7715
(17 studies, 20 comparisons)

⊕⊕⊝⊝
low1,2,3

Beyond the 6‐month assessment, only 2 studies (Halpern (A) 2015; Volpp 2009) demonstrated sustained cessation.

*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; OR: Odds ratio;

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

111/20 trials failed to conduct adequate randomisation and allocation procedures, and 6 of them were rated at high risk of bias in one or both of these domains.
2Eight of the 20 studies (generally the older trials) presented results as percentages or in graphical form, which obliged us to estimate the findings without being certain of the true denominators or event counts
3Three early trials may have confounded their findings by assessing smoking status at the same time as they delivered six‐month rewards for abstinence. In one case (Rand 1989) this was also the end of the trial.

Figures and Tables -
Summary of findings for the main comparison. Incentives for smoking cessation
Summary of findings 2. incentives for smoking cessation in pregnant women at longest follow‐up

incentives for smoking cessation in pregnant women at longest follow‐up

Participantnt or population: Pregnant women who smoke
Settings: antenatal clinics
Intervention: Incentives

Outcomes

Illustrative comparative risks* (95% CI)

Relative effect
(95% CI)

No of Participants
(studies)

Quality of the evidence
(GRADE)

Comments

Assumed risk

Corresponding risk

Control

Incentives

Smoking cessation in pregnant women at longest follow‐up
Follow‐up: 10 ‐ 24 weeks post partum

48 per 1000

154 per 1000
(108 to 216)

OR 3.60
(2.39 to 5.43)

1295
(8 studies)

⊕⊕⊕⊝
moderate1

1 included study (Donatelle 2002) did not contribute to the analysis because of lack of usable data

*The basis for the assumed risk (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).
CI: Confidence interval; OR: Odds ratio;

GRADE Working Group grades of evidence
High quality: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low quality: We are very uncertain about the estimate.

1Only 2 studies (Ondersma 2012; Tappin 2015) were rated at low risk of selection bias; the rest were at unclear risk, apart from Higgins 2004 (high risk).

Figures and Tables -
Summary of findings 2. incentives for smoking cessation in pregnant women at longest follow‐up
Comparison 1. RESULTS OF INCLUDED STUDIES

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 RESULTS TABLE: Mixed populations Show forest plot

Other data

No numeric data

2 RESULTS TABLE: Pregnant women Show forest plot

Other data

No numeric data

Figures and Tables -
Comparison 1. RESULTS OF INCLUDED STUDIES
Comparison 2. Smoking cessation: mixed populations

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Smoking cessation in mixed populations Show forest plot

20

Adjusted Odds Ratio (Fixed, 95% CI)

Subtotals only

1.1 Longest follow‐up

20

7715

Adjusted Odds Ratio (Fixed, 95% CI)

1.42 [1.19, 1.69]

1.2 Six months

19

6945

Adjusted Odds Ratio (Fixed, 95% CI)

1.72 [1.43, 2.08]

1.3 12 months

8

4245

Adjusted Odds Ratio (Fixed, 95% CI)

1.17 [0.94, 1.46]

1.4 18 months

1

561

Adjusted Odds Ratio (Fixed, 95% CI)

1.59 [0.89, 2.83]

1.5 24 months

2

1331

Adjusted Odds Ratio (Fixed, 95% CI)

1.29 [0.93, 1.79]

Figures and Tables -
Comparison 2. Smoking cessation: mixed populations
Comparison 3. Smoking cessation: pregnancy

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 Pregnancy trials at longest follow‐up (adjusted OR) Show forest plot

8

1295

Odds Ratio (Fixed, 95% CI)

3.60 [2.39, 5.43]

2 Pregnancy trials: abstinence at end of pregnancy Show forest plot

8

1297

Odds Ratio (M‐H, Fixed, 95% CI)

3.79 [2.74, 5.25]

3 Pregnancy: contingent rewards vs guaranteed payments Show forest plot

4

278

Odds Ratio (M‐H, Fixed, 95% CI)

6.26 [2.35, 16.68]

4 Pregnancy: front‐loading vs incremental payments Show forest plot

1

Odds Ratio (M‐H, Fixed, 95% CI)

Totals not selected

5 Pregnancy: Participant‐initiated CM Show forest plot

1

Odds Ratio (M‐H, Fixed, 95% CI)

Totals not selected

Figures and Tables -
Comparison 3. Smoking cessation: pregnancy