The SNAP trial: a randomised placebo-controlled trial of nicotine replacement therapy in pregnancy; effectiveness and safety until 2 years after delivery, with economic evaluation

Eligibility criteria

Eligible women were aged 16–50 years, between 12 and 24 weeks’ pregnant, smoked at least 10 cigarettes per day before pregnancy and continued to smoke at least five cigarettes per day. The eligible women also had an exhaled carbon monoxide (CO) concentration of at least 8 parts per million (p.p.m.). They were excluded if they had contraindications to the use of NRT including severe cardiovascular disease, unstable angina, cardiac arrhythmias, recent cardiovascular accident or transient ischaemic attack, chronic generalised skin disorders or known sensitivity to nicotine patches, chemical or alcohol dependence, known major fetal abnormalities, or were unable to give informed consent. Women could enrol in the trial only once but could participate in other non-conflicting research projects.

Recruiting centres

Participants were recruited from seven hospital antenatal clinics in the Midlands and north-west England. Initially, these were at Nottingham University Hospitals NHS Trust City Hospital campus, Nottingham University Hospitals NHS Trust Queen’s Medical Centre (QMC) campus, Sherwood Forest Hospitals NHS Foundation Trust (King’s Mill Hospital) and University Hospital of North Staffordshire NHS Trust (City General Site). Three further sites were added later to improve recruitment rates: Mid Cheshire Hospitals NHS Foundation Trust (Leighton Hospital), East Cheshire NHS Trust (Macclesfield District General Hospital) and Derby Hospitals NHS Foundation Trust (Derby City General, later to become Royal Derby Hospital).

Research midwives

In each centre, research midwives (RMs) undertook all trial-related procedures. RMs were trained in research procedures by the trial manager with input from the chief investigator and attended monthly staff update meetings. In addition, Clare Mannion, one of the trial co-investigators and a UK expert trainer of smoking cessation professionals, provided the RMs with training, to English national standards, in delivery of behavioural support to pregnant women. 36

Recruitment and consent

We used three methods of identifying and recruiting eligible women who were interested in stopping smoking.

1. Community midwives usually ask women about smoking status at their booking appointment and then refer those who would like help to stop smoking to their local NHS SSS. In some recruiting areas, women referred to NHS SSS were asked if they would be interested in finding out about the trial and the RM contacted any who expressed interest. Those who were not interested or eligible for enrolment, but who wanted to stop smoking, were seen by the NHS SSS as per normal practice.

2. Leaflets containing brief information about the trial were sent to women before their antenatal clinic or routine ultrasonography scan appointments. Women attending these clinics were then asked to complete a screening questionnaire to identify those who were eligible and interested (see Appendix 1).

3. Some women who had seen information leaflets or posters advertising the study in hospitals contacted the RMs directly.

Potentially eligible women who expressed interest in the trial were given a participant information sheet (PIS) and, after having chance to consider this for at least 24 hours and discuss with a RM, gave their written informed consent before trial data were collected. In addition to trial participation, women were asked to give consent for researchers to have access to their and their child’s medical records, for information held by the NHS to be used to keep in touch with them and to follow their health status, and also for storage of blood samples for possible use in future research. For most trial participants, consent and baseline data collection took place in their homes.

Provision of additional behavioural support and nicotine-replacement therapy to trial participants and availability of nicotine-replacement therapy to non-participants

Prior to starting the trial, we visited local NHS SSSs in the recruiting areas and discussed their service provision, as it was intended that these would provide behavioural support to women enrolled onto the trial. Primary care trusts (PCTs) in all of the trial’s recruiting areas had NHS SSSs for pregnant women, but several PCTs had recently undergone local reorganisations and there was considerable variation in the delivery of cessation services. Some services were already issuing NRT to pregnant women, despite the absence of evidence for its effectiveness. We hoped to get agreement from PCTs that, for the duration of the trial, NRT would be issued only to pregnant women within the trial, thereby ensuring that local availability did not jeopardise recruitment and/or retention of participants or interpretation of trial findings. The outcome of these visits meant that, in most trial areas (all except Derby), women’s contact details were shared with their local NHS SSS, which agreed to contact participants and offer them additional behavioural support. They also agreed that they would not normally offer participants any non-trial NRT products. Women were encouraged to ask for further behavioural support as necessary, and RMs or NHS SSS staff, guided by the manual, delivered any additional support that participants required. In Derby, where participants’ contact details were not shared with the NHS SSS, RMs provided additional support. The provision of additional behavioural support to participants and the availability of NRT to participants and non-participants who contacted the NHS SSS are shown in Table 1, which illustrates the context in which trial recruitment occurred.

Baseline

At baseline, RMs collected women’s demographic and contact details: ‘Heaviness of Smoking Index’,39 a measure of nicotine addiction recorded on a scale of 0 (lower) to 6 (higher nicotine addiction); number of daily cigarettes smoked before pregnancy; partner’s smoking status; gestation; ethnicity; age completed full-time education; parity; use of NRT in current pregnancy; height and weight. Women were also asked to provide an exhaled CO measurement, blood and saliva samples for cotinine estimation and a blood sample for future deoxyribonucleic acid (DNA) extraction (any studies using the DNA samples would require further funding and relevant ethical approvals).

One month

One month after the quit date, RMs telephoned participants and asked for details about their smoking status. This included whether or not they had smoked since their agreed quit date and, if so, how often this had occured and whether or not they had smoked in the previous 24 hours. RMs also asked about the number of times they had received additional behavioural support and whether this had been face to face, by telephone or by mobile telephone text message; the number of trial patches they had used (i.e. adherence); and whether or not they had obtained and used any additional NRT outside of the trial. Those who reported not smoking were visited for CO validation and a saliva sample (for cotinine measurement) was obtained from any women who were still using trial patches and not smoking. Non-contactable women were sent a postal questionnaire.

Delivery

When participants were admitted to hospital in established labour prior to childbirth, or as soon as possible afterwards, as at the 1-month contact, RMs or delivery suite staff ascertained smoking status, use of trial and ‘non-trial’ NRT, including reported numbers of patches used and additional behavioural support received. Exhaled CO measurements and saliva cotinine samples were obtained from women who reported not smoking for at least 24 hours before delivery. RMs telephoned those missed while in hospital, and any who reported abstinence were visited at home for biochemical validation within a maximum of 4 weeks after delivery. Maternal and infant birth outcomes were obtained from medical records. Maternal outcomes included hypertension (> 140/90 mmHg) measured on two or more occasions during pregnancy, miscarriage or stillbirth, labour onset (spontaneous, induced or no labour), mode of delivery (spontaneous vaginal, assisted vaginal or caesarean section) and antenatal or postnatal hospital admissions. Infant data and outcomes collected at delivery or after discharge included date of birth, name, hospital and NHS numbers, sex, birthweight, gestational age at birth, number of births (with number and birth order if multiple birth), live birth or stillbirth, arterial cord-blood pH (either ≥ 7 or < 7), Apgar score (either ≥ 7 or < 7), intraventricular haemorrhage, neonatal convulsions, admissions to neonatal intensive care units (NICUs) and any congenital abnormalities.

Adverse event monitoring

During each contact with participants, RMs enquired about adverse events (AEs) or symptoms the participant had experienced. RMs also obtained this information from monthly examination of medical records. They then summarised the descriptions in the case report forms and on the study database. Descriptions were used to code the AEs according to standard terms in the Medical Dictionary for Regulatory Activities (MedDRA^®) version 13.1 [www.meddra.org/; MedDRA is the international medical terminology developed under the auspices of the International Conference on Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use (ICH). MedDRA trademark is owned by International Federation of Pharmaceutical Manufacturers and Associations on behalf of ICH]. The incidence of events was analysed according to the MedDRA System Organ Class categorisation and preferred terms. Information about deaths after delivery was obtained from the NHS Health and Social Care Information Centre’s Data Linkage Service (previously Office for National Statistics) with whom all trial participants and their infants had been flagged. After starting the trial, it was realised that many relatively common pregnancy-related events were being reported as serious adverse events (SAEs), including pregnancy-related hospital admissions, premature birth and LBW and, on clinical review, none of these had been considered to be related to the study drug. The Trial Steering Committee (TSC) and the sponsor, therefore, advised that it would be appropriate to amend the protocol so that only maternal and fetal/infant deaths and hospital admissions unrelated to the underlying pregnancy would be reported as SAEs, and ethical approval was received for this. We continued to collect and monitor these along with other AEs, and the Data Monitoring Committee (DMC) reviewed data showing the distribution of all AEs and SAEs within trial groups at their meetings.

Figure 1 is a flow chart showing the data collection process from recruitment until delivery.

FIGURE 1.

Trial flow from recruitment to delivery.

Six months after childbirth

The following information was collected from participants 6 months after delivery: current smoking status, smoking status since childbirth, maternal use of NRT and NHS SSSs since childbirth, length of any maternal hospital inpatient stay after delivery lasting > 24 hours, length of any infant inpatient stay in special care after birth, numbers of additional infant hospital admissions for respiratory illness or other causes, infant feeding method and a health status measure – the European Quality of Life-5 Dimensions (EQ-5D). 42

One year after childbirth

A shorter questionnaire was sent at 1 year, primarily to maintain contact with the participant, but also to collect the following information: current smoking status, smoking status since birth of infant, infant’s respiratory symptoms, infant hospital admissions for respiratory illness and other causes, and infant feeding method.

Two years after childbirth

At 2 years, a questionnaire sent to participants, the 2-year ‘participant questionnaire’ (PQ2) asked about maternal smoking behaviour and infant development. The PQ2 used items from the Ages and Stages Questionnaire^®, Third Edition (ASQ-3™),43,44 which has been developed for assessing child development at 2 years and is valid for use from 23 months until 25 months and 15 days. It was designed to be completed by parents and to distinguish between children with suspected developmental delay and those for whom development is within the normal range. In a US population, the ASQ-3 has been reported to have a sensitivity of 92.2% and a specificity of 71.9% for detecting developmental delay at 24-months. 45 With permission from the authors and publishers, the wording of some questions was slightly adapted for our UK population. Items 1–30 on the PQ2 consisted of all 30 ASQ-3 items on child development covering five domains: communication, gross motor, fine motor, problem solving and personal–social development. Seven additional PQ2 items (i.e. PQ2 items 31–36 and 44), also taken from the ASQ-3, were mixed ‘yes’ or ‘no’/free text questions investigating both general and specific parental concerns relating to infant health and development. The PQ2 also contained items that were not from the ASQ-3 asking about infant hospital admissions, parental reports of infants’ respiratory symptoms and any medication taken for these.

Health professional questionnaire

If, at 2 years, participants did not respond to the PQ2 questionnaire, the health professional questionnaire (HPQ) was posted to their GPs. This shorter questionnaire was designed to be easily completed using medical or health visitors’ records and health professionals completing HPQs required little knowledge of the infants. If GPs could not complete HPQs, they were asked to forward these to health visitors. The HPQ contained items that corresponded to those on the PQ2 and were also intended to measure children’s disability and general health in a manner consistent with the ASQ-3. 46–48 This included open-response questions that corresponded to ‘non-domain’ ASQ-3 items included on the PQ2.

For any participants for whom we received both the completed PQ2 and HPQ, only responses from PQ2 were used in analyses.

The system we used to map the question responses to outcomes is outlined in the Derivation of composite ‘impairment’ outcome for infants section and in the statistical analysis plan (SAP) (this can be accessed at http://eprints.nottingham.ac.uk/3283/).

Primary outcome to delivery

The primary outcome was self-reported, prolonged and total abstinence from smoking between the quit date and delivery, validated by exhaled CO and salivary cotinine (COT) at childbirth. Occasional minor lapses (no more than five cigarettes in total between the quit date and delivery) were not counted as a return to smoking; this is consistent with standard criteria for assessing outcome in cessation studies. 49 CO readings of ≤ 8 p.p.m. and COT of < 10 ng/ml indicated not smoking. 38 The method used for deriving the primary outcome from responses at 1 month and delivery is detailed in Box 1.

Secondary outcomes

1. Smoking outcomes monitored until delivery

   1. self-reported, prolonged abstinence from smoking between the quit date and 1 month

   2. self-reported, prolonged abstinence from smoking between the quit date and 1 month with biochemical validation (exhaled CO) (this outcome was not listed in the trial protocol).

   3. self-reported, prolonged abstinence from smoking between the quit date and delivery

   4. self-reported, prolonged abstinence from smoking between the quit date and delivery, with biochemical validation of this at both 1-month follow-up and delivery

   5. self-reported smoking cessation for the previous 24-hour period at delivery validated by exhaled CO and saliva cotinine estimation.

2. Smoking outcomes monitored after delivery

   1. self-reported, prolonged abstinence from smoking between the quit date and 6 months after delivery

   2. self-reported smoking cessation for the previous 7-day period at 6 months after delivery.

3. Birth and maternal outcomes

   1. miscarriage (non-live birth prior to 24 weeks’ gestation) and stillbirth (non-live birth at 24 weeks’ gestation or later)

   2. neonatal death (i.e. from live birth to 28 days)

   3. post-neonatal death (29 days to 2 years)

   4. individualised birthweight z-score (i.e. birthweight adjust for gestational age, maternal height, maternal weight at booking and ethnic group)

   5. unadjusted birthweight and birthweight as z-score

   6. Apgar score

   7. cord blood pH

   8. gestational age at birth

   9. intraventricular haemorrhage

   10. neonatal enterocolitis

   11. neonatal convulsions

   12. congenital abnormality

   13. NICU admission

   14. infant ventilated > 24 hours

   15. elective termination

   16. elective termination undertaken for fetal morbidity judged incompatible with fetal/infant survival

   17. maternal mortality

   18. mode of delivery

   19. hypertension in pregnancy (> 140/90 mmHg at least twice).

Outcomes 2 years after delivery

1. Infant impairment

   1. Defined as presence of disability and/or behaviour and development problem(s) and categorised as:

      1. survival with no impairment: two of the outcomes in the protocol at 2 years were (1) behaviour and development and (2) disability. These outcomes were combined for analysis and reporting purposes and the mapping of protocol outcomes on to those listed above is fully described in the next section (see Derivation of composite ‘impairment’ outcome for infants) and the SAP (http://eprints.nottingham.ac.uk/3283/)

      2. survival with definite impairment (this outcome was not listed in the trial protocol)

      3. survival with suspected impairment (this outcome was not listed in the trial protocol).

   2. Survival with ‘no impairment’ was the primary outcome at 2 years.

2. Infant respiratory symptoms

3. Smoking outcomes

   1. self-reported, prolonged abstinence from smoking between quit date and 2 years after delivery

   2. self-reported smoking cessation for previous 7-day period at 2 years after delivery.

Derivation of composite ‘impairment’ outcome for infants

Derivation of infant respiratory problems

Infants were judged to have a respiratory problem if, at 2 years, any of questions 38–42 of the PQ2 and/or question 7 of the HPQ indicated this. On the PQ2, these questions included hospital admissions for respiratory problems, problems with chest or breathing (yes/no, free text), wheeze or whistling in chest (yes/no, frequency), doctor diagnosed asthma (yes/no), asthma medications taken (yes/no, inhaler description free text). The HPQ asked whether or not the child has problems with their chest or breathing (yes/no, free text).

Maternal smoking outcomes after delivery

We assessed these in a manner consistent with Russell Criteria49 and with smoking outcomes reported at delivery. The derivation of smoking outcomes is described below:

1. Positive outcome for ‘Self-reported, prolonged abstinence from smoking between quit date and 2 years after delivery’: the participant must have met the criteria for prolonged abstinence at delivery (i.e. positive primary outcome), plus one of the following responses

   1. ‘smoked since 2-year-old was born’ = ‘No’

   2. or

   3. ‘how often have you smoked’ = ‘five times or less’

   4. If any participant questionnaires were completed at 6 and/or 12 months, these should all have had the same responses as above for a positive outcome.

2. Smoked in last week (self-reported smoking cessation for previous 7-day period at 2 years after delivery):

   1. ‘smoked since two year old was born’ = ‘No’

   2. or ‘smoked in last week’ = ‘No’

3. Smoked in last 2 years (self-reported prolonged abstinence from smoking between delivery and 2 years) (this outcome was not listed in the trial protocol)

   1. ‘smoked since two year old was born’ = ‘No’

   2. or

   3. ‘how often have you smoked’ = ‘5 times or less’

   4. If any participant questionnaires had been completed at 6 and/or 12 months, these should all have had the same responses as above for a positive outcome.

Outcomes collected on 6-month and 1-year questionnaires

Smoking status and respiratory outcome items were included on 6-month and 1-year questionnaires. These were not listed in the study protocol and only smoking outcomes are reported later. In addition, at 6 months, the EQ-5D questionnaire items were also used and questionnaire items asked about length of stay in hospital and/or special care unit after delivery, infant hospital admissions and medications taken; these data were intended for the Health economics analysis (see Chapter 4).

The 6-month and 1-year questionnaires included questions about breast and bottle-feeding and, again, these items were not included in the original study protocol and data are not presented in this report.

Analysis to primary outcome point (delivery)

Analysis was on an intention-to-treat (ITT) basis and participants who, for any reason, had missing outcome data were assumed to be smoking. The proportion of women who reported prolonged abstinence from smoking immediately before childbirth was compared between treatment groups by logistic regression and adjusted for recruitment centre as a stratification variable. Statistical significance was assessed using the likelihood ratio test. The primary analysis adjusted for no further variables as multivariate analysis results, and therefore overall conclusions, can be sensitive to decisions concerning which variables to adjust for and how these are specified. Nevertheless, we planned a secondary analysis adjusting for baseline COT (continuous variable), maternal education (in years) and partners’ smoking status (binary variable), as adjusting for potentially important prognostic factors can improve the precision of treatment effect estimates. 51 Other smoking cessation outcomes were analysed similarly.

Fetal and maternal birth outcomes were compared on an ITT basis. For binary outcomes, ORs were obtained using logistic regression adjusted for recruitment centre and also using the likelihood ratio test (with Fisher’s exact test used and stratification by centre ignored when numbers of events were small). For continuous outcomes, we compared means between groups with adjustment for recruitment centre using multiple linear regression.

For fetal outcomes, primary analysis was of singleton births only to allow for the fact that observations will be non-independent and that non-singleton births are likely to have different birth outcomes. However, we undertook a sensitivity analysis including multiple births, with clustering of outcomes accounted for using an approach previously published. This adapts the methodology previously created for use with cluster randomised controlled trials (RCTs), assuming that each woman is regarded as the ‘cluster’ and her number of offspring the cluster size. 52

In all analyses, a p-value of < 0.05 was taken to indicate statistical significance and 95% CIs were calculated.

Analysis at the 2-year follow-up point

The ASQ-3 does not require adjustment of an infant’s age to allow for prematurity once he or she reaches 24 months of age; therefore, as the questionnaire was sent shortly before the child’s second birthday, no adjustment to infant ages was made in analyses.

Maternal characteristics at baseline and delivery and infant birth outcomes at delivery were compared between those participants and infants who did and did not have outcomes ascertained at 2 years after delivery. We also compared maternal and infant characteristics according to whether follow-up at 2 years was by PQ2, HPQ or neither.

Analysis of early childhood outcomes was on an ITT basis with participants analysed in the treatment groups to which they were randomised. Participants with no live birth (i.e. miscarriage, stillbirth or elective termination) or those where the pregnancy outcome was unknown were excluded from the ITT analysis, but postnatal infant deaths were included in the denominator for developmental outcomes. The primary analysis was restricted to singleton births to allow for the fact that observations will be non-independent and that multiple births may have very different outcomes. For the primary outcome, survival with no impairment, a complete case analysis was compared with an analysis using multiple imputation to deal with missing values. Multiple imputation was carried out using the ‘mi’ commands in Stata and in our multiple imputation we included all of the complete baseline and the treatment code, and used 20 imputations. Using this approach, multiple imputation was also used for the other developmental outcomes: suspected and definite developmental impairment. The infant impairment and respiratory outcomes at 2 years were analysed as binary indicators of presence or absence of the outcome. The ORs for the effect of treatment group were obtained by logistic regression adjusting for centre as the stratification factor. In a subsidiary analysis, multiple births were included and clustering accounted for by the same method as in analysis at delivery. We also conducted sensitivity analyses comparing the results of analyses based on parental responses only and those based on a combination of parental and health professional responses.

Smoking outcomes were also analysed on an ITT basis, with all women analysed in the treatment groups to which they were randomised and all non-respondents assumed to be smoking. ORs for the effect of treatment group on smoking cessation outcomes were also obtained by logistic regression. As at delivery, the primary analysis adjusted only for centre, but we carried out sensitivity analysis that also adjusted for baseline COT, partner’s smoking status and age at finishing education.

We tested the assumption that those missing at follow-up were smokers by exploring alternative associations for the relationship between smoking status and ‘missingness’. 53 In this analysis, we defined the OR for the association between quitting and being missing as the informatively missing odds ratio (IMOR) and we looked at the effect on the size of the treatment effect on smoking abstinence outcome by varying the size of this OR between 0 and 1. In the main analysis, the assumption that those missing at follow-up are smokers is equivalent to assuming that IMOR equals 0 (i.e. that all those who are missing are smokers). We altered this OR up to IMOR equals 1, which is equivalent to assuming that there is no association between being missing data and smoking status. We carried out this analysis using the mean score method to estimate the treatment effect under the pattern mixture model, logit[E(y|r,m_y)] = α₁ + β₁r + m_yδ, where m_y is an indicator of whether the outcome is missing or otherwise, r is the treatment effect, and exponential (δ) [the OR between outcome y and m_y (IMOR)] is varied in the range 0–1. 54 α₁ and β₁ are estimated using the subgroup with outcome data, missing values of y (the outcome) are replaced by invlogit(α₁ + β₁r + δ), the mean of this new y variable is calculated for the intervention and control arms (say a1 and a0), and ORs calculated from these mean values accordingly [a1/(1 – a1)]/[a0/1 – a0)].

Secondary analysis

A priori, we planned to investigate whether or not there was any relationship between self-reported nicotine patch use in pregnancy and the presence or absence of developmental impairment in infants at 2 years. For this, we conducted an exploratory regression analysis with absence of impairment at 2 years as the dependent variable and the number of nicotine patches women reported having used when asked at delivery as an explanatory variable. ‘Suspected’ and ‘definite’ impairment categories were combined into one group representing infants who did not have impairment-free survival at 2 years. We investigated the possibility that baseline maternal characteristics may have a confounding effect on any relationship and adjusted for confounders as appropriate. For those in the placebo group, we set adherence with nicotine patches as zero. Additionally, if data on adherence were not reported at delivery, we imputed 0 days use of nicotine patches.

Protocol amendments

Brief details of protocol amendments made after the start of recruitment, but prior to breaking treatment allocation codes, are listed below.

1. We originally anticipated that women would be sent the PIS before their clinic appointments so that they could then be recruited and consented when they attended for their antenatal scan appointment. However, it was soon realised that, overall, this was not a practical option and that most women were being enrolled on home visits. Therefore, rather than posting the PIS to all women, we added the option of just sending leaflets containing brief information about the trial, with the full PIS later posted to women who had been identified and contacted using the screening questionnaire.

2. Small changes were made to the protocol to clarify trial processes and allow for minor variations in practice in the different centres due to different local arrangements in, for example, prescribing and dispensing practices, local clinic arrangements, follow-up cessation support and time spent in hospital after delivery.

3. Ambiguities in the primary outcome measure were addressed and clarified in the protocol, including the time window in which data collected could be used for analysis, how self-report and biochemical validation data of smoking cessation contributed to a positive primary outcome and what constitutes ‘prolonged abstinence from smoking’. Secondary outcomes were clarified to distinguish and define fetal death at different gestations, i.e. miscarriage and stillbirth.

4. We obtained ethical approval to send a ‘congratulations on the birth of your baby’ card to women after delivery.

5. The content of the questionnaires sent at 6 months, 1 year and 2 years after delivery was finalised and approved, along with a questionnaire to be sent to participants when they could not be contacted 1 month after their quit date. We also decided to include incentives of shopping vouchers and a colouring competition to help improve response rates for the 2-year questionnaires. 41

6. Once the trial started, we realised that as many pregnancy-related conditions required hospital admissions this was resulting in a large number of SAE reports, none of which were felt to be related to the study drug. Therefore, the sponsor and TSC recommended that we should extend the list of conditions in the protocol that did not need to be reported as a SAE (any deaths of the mother or fetus/infant were still reported). All these AEs were still collected and reviewed by the DMC so that unforeseen impacts of NRT could be monitored.

7. Following further stability data from the manufacturer of the nicotine and placebo patches used in the study, the shelf life was extended from 24 to 42 months.

Trial extensions

The HTA granted two time extensions to the application, adding a total of 12 months to the original length of the trial. This was necessary as the start of recruitment was slightly delayed and the overall recruitment period took 10 months longer than the original estimate of 24 months. Careful budgeting of trial resources funded the majority of this extension, but a small addition to the budget was also awarded.

Recruitment and flow of participants through the trial

Participants were recruited between May 2007 and February 2010 (Figure 3 and Table 2). We initially estimated that recruitment would take 24 months, but after the first 6 months, target figures were revised in line with actual recruitment figures and the recruitment period extended by 10 months.

FIGURE 3.

Cumulative trial recruitment.

The Consolidated Standards of Reporting Trials (CONSORT) diagram (Figure 4) summarises the process of recruitment and flow of participants through the study to the primary follow-up point. Approximately 21,000 women were informed of the trial by questionnaires that were distributed and completed in antenatal clinics. The majority of these (around 18,590) were excluded without further contact either because the screening questionnaire showed that they were not eligible (usually because they were not smokers), or because they had no interest in joining the trial.

FIGURE 4.

The CONSORT diagram showing flow of participants to delivery. From Coleman et al. 56 Copyright © 2012 Massachusetts Medical Society. Reprinted with permission. DNA, did not attend.

Of the 2410 women who expressed interest in the trial and were assessed for eligibility, 1051 (43.6%) were randomised: 521 were assigned to receive NRT and 530 were assigned to receive placebo patches (see Figure 4). One woman was mistakenly enrolled for a second time in a subsequent pregnancy; her second enrolment in the placebo group was removed from all analyses, giving a final sample size of 1050 (529 in the placebo group).

Protocol breaches were discovered for 13 other participants, but after consideration of violation details it was decided that these were not serious and would have no significant impact on trial participants or the scientific integrity of the trial. These participants, therefore, remained in the trial and their data were used in analyses; details of these protocol breaches are given in Appendix 4.

Two additional problems affecting 27 participants occurred within one site pharmacy, but, again, these were judged to have no significant impact on participants or trial integrity, and details are given in Appendix 4.

Of 1050 pregnancies, 1038 were singleton and 12 were twin.

At 1 month after their quit date, 866 women (82.5%) provided outcome data and, of these, 573 (66.2%) responded by telephone, 19 (2.2%) responded by questionnaire and 274 (31.6%) attended face-to-face consultations with RMs.

At delivery, 981 (93.4%) participants provided smoking outcome data, but 46 (4.4%) who could not be contacted within the necessary time frame, 10 (1.0%) who withdrew consent and 13 (1.2%) who experienced fetal or infant death (including one elective termination) were not asked for their smoking status.

For most participants who reported that they were non-smokers, biochemical validation was obtained. The validation rates at 1 month were 89% (116/131) in the NRT group and 85% (63/74) in the placebo group. At delivery, validation rates were 89% (58/65 women) in the NRT group and 92% (45/49) in the placebo group.

The ascertainment rate for birth outcomes was more complete than smoking outcomes as these were obtained from participants’ hospital notes. Figure 5 summarises numbers of births within participants and completeness of birth outcome ascertainment.

FIGURE 5.

Completeness of birth outcome data. From Coleman et al. 56 Copyright © 2012 Massachusetts Medical Society. Reprinted with permission.

Further details on the numbers of participants who were followed up and for whom outcome data were obtained at 1 month and at delivery are presented by study centre in Table 3.

Smoking outcomes at delivery

For self-reported (i.e. non-validated) abstinence, there was a slightly larger but still non-significant difference in quit rates: 12.5% with NRT compared with 9.3% with placebo (OR 1.40, 95% CI 0.94 to 2.07) (see Table 5). At 1 month, the validated abstinence rate was significantly higher in the NRT group than in the placebo group (21.3% vs. 11.7%, respectively; OR 2.05, 95% CI 1.46 to 2.88). Similar findings were found for adjusted analyses with all smoking outcomes.

Birth outcomes at delivery

Table 6 shows outcomes for singleton births including deaths, mean birthweight and rates of preterm birth, LBW and congenital abnormalities, and these were mainly similar in the two study groups. However, there were significantly more deliveries by caesarean section in the NRT group than in the placebo group (20.7% vs. 15.3%). Analyses that included twin births gave very similar findings.

Adverse events

Other AEs are shown in Table 7 and, apart from skin reactions at the patch site (97 participants in NRT group reported skin reactions compared with 28 in placebo group), rates were similar in the two groups. In total, 46 participants in the NRT group and 32 in the placebo group stopped using the patches permanently owing to AEs. Further information on the distribution and nature of AEs is in Appendix 5.

Adherence with patches and use of smoking cessation support

Adherence with trial patches was low in both groups and rates are shown in Table 8. Only 7.2% of women (35/485) assigned to receive NRT and 2.8% (14/496) assigned to placebo reported using trial medications for more than 28 days. Additionally, although 111 NRT group participants were abstinent at 1 month and had this validated by a RM (see Table 5), only 72 of these (65%) accepted a second month’s supply of patches. The corresponding figure for the placebo group was 47% (29/62). Only 2.5% of NRT group participants (12/485) and 2.2% (11/496) of placebo group participants reported using ‘non-study’ NRT for ≥ 20 days.

As per protocol, RMs attempted to contact participants and provide behavioural support on the quit date, 3 days after their quit date and 1 month after their quit date. In the NRT group, 368 (70.6%) were successfully contacted via a telephone call from a RM on their quit date and 386 (74.1%) 3 days after their quit date. A total of 69 (13.2%) were not successfully contacted on either day, while 302 (58.0%) had a call on both days. In the placebo group, 378 (71.5%) were successfully contacted on their quit date and 381 (72.0%) 3 days after their quit date. A total of 67 women (12.7%) were not successfully contacted on either day, while 297 (56.1%) received a call on both days. At 1 month, 428 (82.1%) participants in the NRT group received support from a RM either face to face [164 (31.5%)] or by telephone [264 (50.7%)]. In the placebo group, 419 (79.2%) participants received RM support [110 (20.8%) face to face, 309 (58.4%) telephone only]. The lower rate of face-to-face support delivered to the placebo group at 1 month reflects the fact that this was delivered at consultations arranged to validate abstinence from smoking. Support was received by participants on all three occasions (i.e. quit date, 3 days after their quit date and 1 month after their quit date) for 268 participants (51.4%) in the NRT group and 250 participants (47.3%) in the placebo group (see Table 8). Support was not received on any of these occasions for 27 participants (5.2%) in the NRT group and 28 participants (5.3%) in the placebo group.

Participants also reported little additional face-to-face or text message contact with, or support from, smoking cessation advisors who worked for local NHS SSS (see Table 8). Support by telephone was more common and both groups reported receiving a median of two telephone contacts from advisors.

Follow-up of participants and infants after delivery

Follow-up of participants and infants began after the first birth in July 2007 and continued until December 2012.

Two CONSORT diagrams are presented for the follow-up period from delivery until infants were 2 years old; participants indicated as withdrawn in either diagram were not sent further questionnaires after their withdrawal.

Participants’ follow-up

The first diagram (Figure 6) is for all participants and summarises questionnaire distribution and response rates at the three follow-up time points, plus reasons for withdrawal. This provides denominators and follow-up data that are relevant to participant outcomes (e.g. smoking behaviour). By the final follow-up point, 150 participants [14.3% of the original study population: 73 out of 521 (14.0%) NRT, 77 out of 529 (14.6%) placebo] had withdrawn from the study, or a completed 2-year questionnaires had not been received from either the participant or their GP. This includes the 122 participants whose reasons for withdrawal are shown in Figure 6, plus 28 participants with fetal deaths or for whom no birth details were obtained. These are described in Infants’ follow-up.

FIGURE 6.

The CONSORT diagram showing flow of all trial participants during the 2-year follow-up period. a, 26 HPQs were returned in participants who had already returned a PQ2 (n = 12 NRT, n = 14 placebo) and these were not included in subsequent analyses. Reproduced with permission. Cooper S, Taggar J, Lewis S, Marlow N, Dickinson A, Whitemore R, et al. Effect of nicotine patches in pregnancy on infant and maternal outcomes at 2 years: follow-up from the randomised, double-blind, placebo-controlled SNAP trial [published online ahead of print 11 August 2014]. Lancet Respir Med 2014. doi:10.1016/S2213-2600(14)70157-2.

Infants’ follow-up

The second diagram (Figure 7) provides similar details for the 1010 live infants that were known to be born from singleton pregnancies and provides information on follow-up that is relevant to the assessment of infant outcomes at 2 years. This shows the 14 fetal deaths recorded when birth outcome data were analysed (also described in Figure 5), plus 14 participants for whom no infant birth details were obtained. Of those with no birth details, 10 withdrew consent before delivery (three NRT, seven placebo) and four were lost to follow-up (one NRT, three placebo); for ITT analyses of infant outcomes, it has been assumed that 14 live singleton infants were born to these participants. Four singleton infant deaths occurred between birth and the 2-year follow-up (two NRT, two placebo), of which one was known about and recorded when outcomes at delivery were analysed (see Table 6). For simplicity, and as primary analyses for infant outcomes did not include infants carried in multiple birth pregnancies, twin infants are excluded from Figure 5. However, sensitivity analyses included twins and either PQ2 or HPQ responses for 18 twin infants (6 NRT, 12 placebo) from nine of the 12 twin pregnancies were obtained and used in these.

FIGURE 7.

The CONSORT diagram showing flow of participants with live singleton births. These data were used for birth outcome analyses during the 2-year follow-up period. a, 26 HPQs were returned in participants who had already returned a PQ2 (n = 12 NRT, n = 14 placebo) and these were not included in subsequent analyses. Reproduced with permission. Cooper S, Taggar J, Lewis S, Marlow N, Dickinson A, Whitemore R, et al. Effect of nicotine patches in pregnancy on infant and maternal outcomes at 2 years: follow-up from the randomised, double-blind, placebo-controlled SNAP trial [published online ahead of print 11 August 2014]. Lancet Respir Med 2014. doi:10.1016/S2213-2600(14)70157-2.

Completeness of follow-up for participants’ smoking outcomes

From the full trial cohort of 1050 participants, 606 (302 NRT, 304 placebo) (57.7%) responded to the PQ2 questionnaire providing data that could be used in the analysis of participants’ smoking outcomes at the 2-year follow-up (see Figure 6).

Overall follow-up rates were similar in both groups; the response rates for all follow-up time points can be seen in Figure 6 and Table 9, and in chart form in Figure 8. Most participants required a reminder questionnaire and approximately half of those who provided data did so by completing the questionnaire over the telephone. Around three-quarters of health professionals returned the questionnaire by post, although most needed a postal reminder. The total combined response for participant or HPQs at 2 years was 900 (89.8% of those sent questionnaires at 2 years; 85.7% of all randomised participants). This figure excludes the 26 HPQs that were received for participants who subsequently returned their PQ2.

TABLE 9 - Response rates to questionnaires sent during the follow-up period

NA, not applicable.

This includes 26 HPQs that were returned for participants who had already returned a PQ2.

Questionnaire	Questionnaire sent (N)	Reminder sent n (% of questionnaires sent)	Returned by post n (% of questionnaires sent)	Completed by telephone n (% of questionnaires sent)	Total response n (% of questionnaire sent)
Participant 6-month	1015	886 (87.3)	246 (24.2)	427 (42.1)	673 (66.3)
Participant 1-year	1006	875 (87.0)	272 (27.0)	316 (31.4)	588 (58.4)
PQ2	1002	864 (86.2)	304 (30.3)	302 (30.1)	606 (60.5)
HPQ	430	306 (71.2)	320 (74.4)	NA	320a (74.4)
Combined 2-year response (i.e. PQ2 or HPQ returned)	1002	NA	NA	NA	900 (89.8)

FIGURE 8.

Numbers of questionnaires sent and returned during the follow-up period.

Characteristics of participants and infants who were followed up after delivery

Table 10 shows the maternal baseline characteristics and singleton birth outcomes for the 900 participants who had 2-year outcome data in the NRT and placebo groups. As with the whole trial cohort, these two groups had similar demographic characteristics at enrolment. Singleton infant birth outcomes were also similar in the two groups, apart from delivery by caesarean section, which, as found in the full cohort at primary outcome, was similarly slightly higher in the NRT group than in the placebo group (20.2% compared with 15.5%). The proportion of the participants providing data at 2 years who had prolonged, validated abstinence from smoking at delivery was 10.3% in the NRT group and 8.2% in the placebo group (see Table 10), which compares with 9.4% and 7.6% for NRT and placebo groups, respectively, for the full cohort (see Table 5).

Table 11 shows the same characteristics within those participants who provided 2-year follow-up data on the PQ2, those for whom a HPQ was completed instead and those for whom no data was obtained. Most of the demographic data are similar in these three groups; however, those with no follow-up data at 2 years had a slightly higher mean index of multiple deprivation (IMD) score (36.8, compared with 32.3 in those with PQ2 data and 32.2 in those with HPQ data). Their first behavioural support session may also have been slightly shorter (21.3% of this group had a behavioural support session of < 30 minutes, compared with 14.4% and 15.7% of those with PQ2 and HPQ data, respectively). Median COT levels at enrolment appear to be slightly higher in those who were followed up by HPQ (131.6 ng/ml, compared with 119.1 ng/ml in the other two groups).

Participants who completed a PQ2 were more likely to have validated abstinence between quit date and delivery (10.9%, compared with 5.8% of those with HPQ data and 4.0% of those with no 2-year data) (see Table 11).

Infant outcomes at 2 years

Infants’ developmental outcomes and any reports of respiratory problems at 2 years after delivery are shown in Table 12, separated by treatment group for all those with known outcomes, including four postnatal deaths. In Table 13, these results are broken down further to show outcomes by questionnaire and, for comparison purposes, postnatal deaths are not included in the denominators in this table.

Of the 1010 singleton infants, 891 (445 NRT, 446 placebo) had information about them returned on either the PQ2 or the HPQ questionnaire; however, owing to missing data, it was not possible to allocate developmental outcomes on one PQ2 (from placebo group) and six returned HPQs (two NRT, four placebo). Within these 891, infants born to women who had been allocated to the NRT group in pregnancy were significantly more likely to have survived with no impairment than those born to women allocated to placebo [323/445 (72.6%) of NRT group infants and 290/443 (65.5%) in the placebo group]. The OR for ‘survival with no impairment’ in the NRT compared with the placebo group obtained in the primary analysis (ITT with multiple imputation analysis) was 1.40 (95% CI 1.05 to 1.86, p = 0.023). Similar statistically significant differences in survival without impairment were also found for other analyses including the clustered analysis with twin births (OR 1.43, 95% CI 1.08 to 1.91, p = 0.013) (see Table 12) and the analysis of singleton infants using just the PQ2 responses (OR 1.52, 95% CI 1.09 to 2.11, p = 0.012) (see Table 13).

There were no significant differences between groups for any of the other infant outcomes (see Tables 12 and 13). Using definitions of developmental impairment described in the methods, ‘definite’ impairment was identified in 48 (10.8%) of the NRT and 64 (14.5%) of the placebo groups, and ‘suspected’ impairment in 72 (16.2%) of NRT and 87 (19.6%) of placebo groups. A greater number of questionnaires that were returned by participants reported potential child development problems than the health professional ones; 253 (42.0%) of the 602 returned PQ2s were categorised with definite or suspected impairment, compared with only 18 (6.2%) of the 289 HPQ responses.

A detailed breakdown of infant developmental outcomes from the PQ2 and HPQs, including ASQ-3 domain scores and problems reported in the supplementary questions, is shown in Table 14. Overall, participants were most likely to report problems with their infant’s talking (23.1%), followed by behaviour (10.3%), but these showed no differences between groups. However, there were significant differences in the number of infants for whom all five ASQ-3 domain scores were normal: 190 (63.1%) in the NRT group compared with 163 (54.2%) in placebo (OR 1.47, 95% CI 1.06 to 2.05, p = 0.02) (see Table 14). For specific items that contributed to ‘survival with no impairment’, the only significant difference between groups was in the ASQ-3 ‘personal-social’ domain; 254 (84.4%) had normal scores in the NRT group compared with 231 (76.7%) in the placebo group. However, although not significantly different, for infants in the NRT group, both the mean scores and the number of infants with normal scores were consistently higher in every ASQ-3 domain. The only other difference was with feeding problems (see Table 14), for which fewer participants allocated to NRT reported problems with their infants’ feeding [18 (6.0%) NRT, 36 (12%) placebo (OR 0.47, 95% CI 0.26 to 0.85, p = 0.011)]. This outcome is not one used by the ASQ-3 questionnaire and did not contribute to ‘survival with no impairment’.

Respiratory problems at 2 years were reported in 132 (29.7%) and 111 (25.0%) of infants born in the NRT and placebo groups, respectively (OR 1.30, 95% CI 0.97 to 1.74, p = 0.083) (see Table 12). As with the development questions, more respiratory problems were reported on participant-completed questionnaires [194 (32.2%) of returned PQ2s compared with 49 (17.0%) of HPQs included reports of respiratory symptoms]. A more detailed breakdown of singleton infants’ respiratory problems is given in Table 15, including outcomes that were collected 1 year after delivery. The only significant difference was in the number of participants reporting at 2 years that their child had ever experienced any wheeze or whistling in their chest [74 (24.6%) NRT, 49 (16.3%) placebo (OR 1.72, 95% CI 1.14 to 2.59, p = 0.0099)].

Maternal smoking outcomes after delivery

After delivery, both point prevalence and prolonged abstinence from smoking was low and relapse to smoking gradually increased (Table 16). During the 2-year follow-up, the numbers of participants in the NRT and placebo groups reporting abstinence from smoking since their quit dates were 28 (5.4%) and 17 (3.2%), respectively, at 6 months, and 19 (3.7%) and 11 (2.1%), respectively, at 1 year. By 2 years after delivery, 15 (2.9%) allocated to NRT and nine (1.7%) allocated to placebo remained abstinent (OR 1.71, 95% CI 0.74 to 3.94, p = 0.20). Table 17 shows that participants who did not provide any smoking data at the three postnatal follow-up time points were more likely to have been smokers at delivery, which is consistent with the assumption used in analyses of smoking behaviour, i.e. that participants who were lost to follow-up should be counted as smokers.

Additionally, the sensitivity analysis that further investigated this assumption found that varying the relationship between missingness and smoking status had almost no impact on ORs that compared smoking cessation rates between trial groups at 2 years. Varying the IMOR between 0 and 1 (in an unadjusted analysis) altered the OR for the effect of treatment group on self-reported prolonged abstinence since delivery from 1.117 (IMOR = 0) to 1.107 (IMOR = 1), the OR for self-reported 7-day cessation from 1.068 to 1.059, and the OR for prolonged abstinence from smoking between quit date and 2 years after delivery from 1.713 to 1.707. This provides added reassurance that treating those with missing outcome data as smokers had no substantial impact on the study findings for smoking outcomes.

Secondary analysis: adherence with nicotine patches and impairment

Data on the presence or absence of impairment at 2 years of age were available for 884 singleton infants and this was the sample size for analysis. Reported adherence with nicotine patches was heavily skewed and skewness was increased by coding placebo group participants’ adherence with nicotine patches as zero. We analysed adherence with nicotine-containing patches in three categories: 0 days’ use and, for participants reporting at least use of one patch (1 day’s use), we used a median split within these participants (median = 10 days’ use) to create two categories of ‘up to median use’ (1–10 days) and ‘above median use’ (11–56 days). Table 18 shows the numbers in each adherence category, the distribution of survival with no impairment within these categories and crude and adjusted OR for the association between adherence categories and survival with no impairment, using 0 days’ adherence as a reference category. Adjustment was made for partner’s smoking status, which was found to be associated with outcome in the multivariable model. Results in Table 18 suggest that participants in the highest category of adherence with nicotine-containing patches were more likely to have infants who had survived without impairments (adjusted OR 1.72, 95% CI 1.22 to 2.57).

Overview

A cost-effectiveness analysis was undertaken to compare NRT patches and behavioural support to behavioural support only, for women who were smoking during pregnancy. The main outcome for the economic evaluation was biochemically validated abstinence from smoking between a quit date and delivery. As recommended by the National Institute for Health and Care Excellence (NICE),61 analyses were conducted from a NHS and personal-social services viewpoint, including direct health effects (maternal smoking cessation) and costs (or cost savings) to the NHS. Mothers were eligible for inclusion in the SNAP trial if they were between 12 and 24 weeks’ gestation and outcomes were collected at delivery (up to 42 weeks’ gestation); therefore, the time horizon of the trial was up to 7 months.

Cost estimation

There were two main components to the costing of the control and intervention strategies: first, the costs of the inputs required for the interventions and second, the resources used to care for each woman and her infant during the period between randomisation and delivery.

Valuation of costs

All data were valued in monetary terms and unit costs were reported in pounds sterling for the financial year 2009–10 (representing the mid-point of the trial). Any costs occurring in prior or later price years were inflated or deflated using the Hospital and Community Health Services pay and prices index. 63 As for the economic evaluation, we considered trial follow-up until to 7 months post randomisation only and no discounting was required. For standard NHS health care, UK unit costs were applied from national sources, increasing the generalisability of the results. Table 19 presents a summary of resource use and unit costs, with the calculation of the costs detailed in Calculating costs.

Calculating costs

In order to calculate the costs of face-to-face NHS SSS sessions, a weighted average cost of individual and group sessions was calculated based on information from a NICE costing report. 62 Salary and overheads information from the same report were used to calculate the cost per minute of a phone call (assumed to be the same length as the calls reported for providing behavioural support).

The cost of mode of delivery was established by calculating a weighted average of unit costs of the different modes of delivery activities recorded in NHS reference costs. A similar method was used to calculate an average cost of a maternal antenatal admission, based on antenatal observations and investigations.

To calculate the cost of the admission of a baby to neonatal care, a weighted average of bed-day costs for neonatal critical care from NHS reference costs (£618) was multiplied by a weighted average length of stay for neonates with major diagnoses (12.2 days) according to Hospital Episode Statistics for 2009–10. 69

Quantities of services used were multiplied by the relevant unit costs to estimate overall cost profiles for women in the trial.

Outcome measures

The measure of health benefit for the economic evaluation was the same as the primary measure of clinical effectiveness in the SNAP trial: self-reported and biochemically validated maternal smoking cessation from quit date to immediately before delivery. Temporary smoking lapses of up to a total of five cigarettes (on up to five occasions) were permitted.

The National Institute of Health and Care Excellence (2008)61 recommend that health outcomes should be measured in quality-adjusted life-years (QALYs) to facilitate comparisons between different health-care programmes. However, QALYs, commonly calculated from the generic health-related quality-of-life tool the EQ-5D,70 may be inappropriate in a study including pregnant women. Generic quality of life studies have shown poorer quality-of-life in early pregnancy compared with population norms for women of child-bearing age71 and substantial changes in quality of life, particularly declining physical functioning and vitality, occur over the course of pregnancy. 72 These dramatic changes in quality of life between pregnancy and the post-partum period would likely mask any potential short-term quality-of-life gains from smoking cessation. Nevertheless, EQ-5D data were collected at 6 months after delivery, within the postal questionnaire sent to participants, which is described in Chapter 2, Methods.

Analysis

An incremental cost-effectiveness analysis was undertaken, following the NICE guidance for health-care evaluations,61 comparing the additional costs of NRT patches with behavioural support alone, as well as the additional benefits, to give a cost per additional quitter.

The incremental cost-effectiveness ratio (ICER) calculates the mean cost of the intervention group over and above the control and divides by the mean difference in health benefits. The following formula is for ICER, for which Δ represents change, C represents the costs, E represents the effects and subscript I and C refer to the intervention and control, respectively.

All analyses were conducted on an ITT basis in which all randomised participants were included and analysed in the groups to which they were randomised. Analyses were conducted in Microsoft Excel, 2010, version 14.0.7113.5005 (32-bit) (Microsoft Corporation, Redmond, WA, USA).

There were no missing data on the effectiveness outcome (validated smoking cessation from quit date to delivery), as any women without validated cessation were assumed to be smokers. Missing data for cost items were imputed using average costs for the appropriate arm of the trial, allowing the base-case analysis to be completed for all women in the trial and, therefore, with the same quit rate as the main effectiveness analyses in Chapter 3.

Cost data were bootstrapped to account for skewness, sampling with replacement observations 1000 times to generate a new population of sample means with an approximate normal distribution. Bootstrap results were presented graphically using a cost-effectiveness plane73 to show the uncertainty around the mean estimates of incremental costs and effects.

The EQ-5D data collected at 6 months after delivery were converted into a single index summary score after applying UK population values. 74 Descriptive statistics for EQ-5D data are presented by trial group. We planned to use 6-month EQ-5D data in conjunction with smoking status utility values based on quit rates at 2 years after delivery to calculate the longer-term costs and benefits of differential quit rates between the intervention and the control arms.

Costs

The breakdown of intervention costs for the trial arms are presented in Table 20. Costs for providing behavioural support and CO monitoring were relatively equal in both arms, although costs of the 4-week home visits were slightly higher in the NRT group than the control group (£14.31 compared with £9.46, respectively) because rates of self-reported smoking cessation were higher in the NRT group at 4 weeks (25.1% compared with 14.0%). The mean total intervention cost in the control group was £47.75. The comparative cost of providing behavioural support and CO monitoring was £52.24 in the NRT group, and the total mean cost including NRT patches was £98.31.

Table 21 reports the resources utilised in the two arms of the trial. Use of NHS SSS, maternal antenatal hospital admissions and admissions to neonatal care were similar in the two groups, although more women had a caesarean section in the NRT group than in the control group: 20.9% and 16.1%, respectively.

Quantities of services used were multiplied by the relevant unit costs in Table 19 to calculate the cost of resources used for each woman and total costs were calculated by adding resource-use costs to intervention costs. Table 22 summarises total mean costs for the trial groups. Mean intervention costs were significantly higher in the NRT group, at a mean difference of £50.56. Total mean resource-use costs were £40.26 higher in the NRT group and overall costs were, therefore, £90.81 higher for this group; however, these differences were not statistically significant.

Cost-effectiveness analysis and uncertainty

Table 23 presents the ICER, combining the differential costs of the treatment groups, with the differential quit rates. NRT plus behavioural support was found to be somewhat more costly than behavioural support alone, but with a slightly higher quit rate. This generates an ICER of £4926 per quitter. If decision-makers are willing to pay > £4926 for an additional quitter, then NRT would be the preferred option; otherwise, behavioural support alone should be adopted.

TABLE 23 - Results of the incremental cost-effectiveness analysis (prices in £ 2009–10)

SD, standard deviation.

	NRT group, n = 521	Control group, n = 529
Cost (SD)	£2669.87 (£2394.09)	£2579.06 (£2385.68)
Quit rate	9.4%	7.6%
ICER	£4926 per quitter
Bootstrapped 95% CI of ICER	–£114,128 to £126,747

A cost-effectiveness scatter plot was produced based on the bootstrapping, including 1000 resamples of the costs and effects data. The bootstrapped results were plotted on a cost-effectiveness plane, visually displaying the uncertainty around the mean differences in costs and benefits between the trial arms (Figure 9). The majority of the plots in the scatter plot fall in the north-east quadrant, indicating that NRT is likely to be more effective but more costly than behavioural support alone. However, the scatter plot also shows the uncertainty around the cost estimates.

FIGURE 9.

Cost-effectiveness plane.

Sensitivity analyses

To allow for uncertainty in our estimates of costs and consequences, we conducted a sensitivity analysis excluding 12 women who had multiple births (four twin births in the NRT group and eight in the control group) who are more likely to have complicated pregnancies and deliveries. Table 24 shows the total costs and quit rates in the singleton-only analysis. Quit rates were very similar (+ 0.08% in the NRT group and –0.46% in the control group); total costs were lower in the singleton-only analysis, but by a similar amount in both groups (–£64.37 in the NRT group and –£72.32 in the control group). The corresponding ICER generated for singleton births was £4156 per quitter.

Projecting longer-term costs and benefits (objectives 4 and 5)

We had planned to use EQ-5D data collected 6 months after delivery in conjunction with smoking status utility values based on quit rates at 2 years to calculate the longer-term costs and benefits of differential quit rates between the intervention and control arms. These longer-term analyses were impeded for a number of reasons. First, there was only a small, non-significant difference in quit rates between groups. Second, there was no difference in EQ-5D scores between the groups at 6 months (Table 25). Third, there was great uncertainty in cost estimates in the alongside trial economic analysis, which would be amplified in any longer-term projections, generating non-robust estimates. Fourth, there were no significant differences in birth outcomes, precluding monetary estimates of the long-term impacts on the child.

Smoking outcomes

This trial demonstrates that, at 12–24 weeks’ gestation, supplementing behavioural support with a 15 mg per 16 hours nicotine patch was no more effective than a placebo in promoting sustained smoking cessation throughout pregnancy. Clinically and statistically significant higher biochemically validated cessation rates were obtained with NRT at 1 month, but this effect did not persist into later pregnancy. After childbirth, self-reported, prolonged cessation since quit dates agreed in pregnancy were between 1% and 2% higher in women who had been randomised to NRT, but these small differences were not statistically significant.

Maternal and fetal birth outcomes

Maternal and fetal birth outcomes were generally very similar with almost no statistically significant differences between groups. Caesarean section births were more frequent in the NRT group and this difference was statistically significant (OR 1.45, 95% CI 1.05 to 2.01).

Infant outcomes at 2 years

At 2 years of age, singleton infants born to trial participants who had been allocated NRT were significantly more likely to have no developmental impairment than those born to participants in the placebo group. Very similar findings occurred whether using data obtained from only participant-completed questionnaires (PQ2) or data from both PQ2 and HPQ questionnaires combined. Additionally, very similar findings were noted in pre-specified analyses, which (1) included twin births and adjusted for clustering of outcomes and (2) applied multiple imputation methods to investigate the assumption that data missing were missing at random (i.e. that missingness was associated with baseline characteristics but not with the outcome itself). No significant difference in reported rates of infants’ respiratory problems was noted.

Economic outcomes

Total mean costs were £90.81 higher in the NRT group, with the excess largely attributable to the cost of NRT patches (mean cost of patches = £46.07). For singleton births only, an ICER of £4156 per quitter was derived; therefore, if decision-makers are willing to pay this amount for each additional quitter, then NRT, used in addition to behavioural support, would be the preferred option. The results showed no differences between groups in birth outcomes or health status as measured by EQ-5D at 6 months postnatal. Therefore, it was not possible to estimate a cost per QALY or model long-term QALYs.

There has been little research investigating the cost-effectiveness of smoking cessation interventions in pregnancy. Although a number of studies have investigated the potential cost saving of smoking interventions for pregnant women,59 these studies have mainly been conducted in the USA with restricted perspectives, particularly omitting data relating to infant outcomes. Only two studies have thus far reported QALYs: one being a simple model based on an American trial60 and another a hypothetical model constructed for NICE guidance. 75 Although these studies suggest that smoking cessation in pregnancy may be cost-effective, they are mainly poor quality and the settings and methods used preclude comparison with our results. Furthermore, with the exception of the NICE model,75 none of the studies has explored the cost-effectiveness of NRT.

Adherence

Adherence to both types of patch was low. Although adherence was not a trial outcome, the pattern of adherence may be related to, and at least partially explain, smoking cessation and birth outcomes and, consequently, requires discussion.

Overall comments

This study is by far the largest of its type. Prior to the SNAP trial being completed, in the five previous RCTs of NRT for smoking cessation in pregnancy, only 695 women had been randomised. 30 We believe that this is the first trial to test whether or not a smoking cessation intervention delivered in pregnancy can affect infant outcomes. The study is also original in that in previous trials of NRT in pregnancy, maternal smoking behaviour has been investigated for only up to 3 months after childbirth,34,35 whereas we followed up participants and infants until 2 years after birth. We can find only one trial testing any smoking cessation intervention (i.e. a non-NRT intervention) for pregnant women that attempted to monitor smoking behaviour up to and beyond 2 years after childbirth. 76 However, in this study, smoking data were not sought from all trial participants at predetermined time points, but was obtained opportunistically at multiple, different times between 8 and 54 months after childbirth, rendering smoking behaviour data difficult to interpret.

Outcomes recorded at and before delivery

By carefully implementing a double-blind, placebo-RCT design, most of the biases that could have influenced outcomes at these time points have been minimised. However, we did not ask women which treatment they perceived they had been allocated to and hence have no data to confirm whether participants remained blinded to their treatment allocations or, indeed, guessed their allocation. Consequently, it is possible that some participants may have correctly determined their treatment allocation and, if this had occurred, we would be unable to quantify its extent. However, unblinded trials of NRT in pregnancy tend to overestimate the treatment effect from NRT30 and we found no effect at delivery. Additionally, loss of blinding among trial participants would not explain the very different pattern of findings with respect to efficacy at 1 month and delivery.

The strengths of the study appear to outweigh its weaknesses and overall findings reported are likely to be valid. Target sample size was achieved; therefore, the study was adequately powered to detect the anticipated 9% difference in cessation rates between trial arms. However, it remains possible that NRT could work for smoking cessation in pregnancy, but with a lesser impact because smaller treatment effects (i.e. < 9% absolute difference) would not necessarily have been detected by a trial of this size. At baseline, groups were well balanced for all variables recorded, including those with potential to influence findings; therefore, chance differences between groups are unlikely to explain these. Additionally, the high outcome ascertainment rates, which were very similar in both groups, reduce the likelihood that ascertainment bias influenced study outcomes. Birth outcome ascertainment rates were particularly high and, as RMs who extracted data on birth outcomes and AEs from medical records were blinded to patch allocation, these data are particularly likely to be free from bias. Similarly, when participants reported abstinence from smoking, equally high biochemical verification rates were obtained from participants in both trial groups, minimising any bias in the reporting of outcomes that may have occurred.

The validity of trial findings could have been affected by unintentional variation in treatments provided for participants, or in any support or treatment sought by them that was additional to trial interventions. For example, differences between NRT and placebo group outcomes would have been minimised if trial participants had also been prescribed NRT from their GPs, issued with it by local NHS SSSs or had bought it from pharmacies. However, only 2.2% in the placebo group (and 2.5% allocated NRT) reported using NRT obtained outside of the trial for more than 20 days and it seems unlikely that this level of usage, which was very similar in both trial arms, would have unduly affected findings. Finally, if participants were inadequately prepared for, or inadequately instructed about using, NRT, then this could have affected their ability to use NRT sufficiently well for any impacts of this treatment to become apparent. However, the trial RMs, trained to English national cessation standards, provided accompanying behavioural support and, in both trial groups, participants’ rates of accessing subsequent additional support were similarly low. This low level of additional support may have affected overall quit rates. However, the level of behavioural support was comparable with that used in ‘low intensity support’ nicotine patch trials conducted among non-pregnant subjects in which NRT has been found effective (RR 1.78, 95% CI 1.49 to 2.12),18 for which NRT caused a near doubling of cessation at the 1-month follow-up point. Taken together, these points suggest that the behavioural support delivered in the trial probably gave women appropriate instruction in using NRT and, as no differences were seen between groups in additional support received by trial participants, this is unlikely to explain trial findings.

Infant outcomes at 2 years

Key strengths are that no previous trials have tested the impact of a smoking cessation intervention in pregnancy on infant outcomes and we achieved high (around 88%) outcome ascertainment rates combined with low withdrawal and missing data rates at 2 years. Follow-up rates were very similar in both trial groups, reflecting the fact that staff conducting follow-up were blind to participants’ treatment allocations. Additionally, we used caution when classifying questionnaires reporting infants’ development and respiratory symptoms and manually checked individual questionnaire responses, when necessary. We were cautious when categorising open-response questionnaire items, generally allocating infants as experiencing ‘suspected’, rather than ‘definite’ impairment, based on responses to these. A similar approach was also taken with open response items on respiratory problems. Consequently, it is unlikely that infants classed as having ‘no impairment’ were actually impaired or that infants categorised as experiencing ‘respiratory problems’ had not had these. Finally, with respect to the primary outcome monitored at 2 years, ‘survival with no impairment’, we obtained the same pattern of findings in all pre-planned analyses, irrespective of whether or not parental or health professional reports of infant health were used, or whether or not twins were included. This consistency of findings suggests that the overall finding, that NRT used in pregnancy has a beneficial effect on child development as measured by ASQ-3, is likely to be valid and unlikely to have occurred by chance.

A limitation is that the ASQ-3 is generally used as a screening tool to identify potential impairments that are subsequently confirmed or refuted by detailed face-to-face assessment. 43 We cannot, therefore, be certain that the parent-reported ‘snapshot’ of child development obtained via completed ASQ-3 questionnaires is perfectly valid for allocating infants as experiencing ‘definite’ or ‘no impairment’. However, a previous UK obstetric trial used 30-item ASQ questionnaires to detect neurosensory disability and, in a subgroup of infants, compared a range of failed and non-failed ASQ scores with outcomes from ‘gold standard’ face-to-face developmental assessments. 77 This found that, when used for detecting neurodevelopmental problems in a cohort of participants and infants that was similar to those in this study, the 30-item ASQ compared well with standard face-to-face assessments and had a negative predictive value of 99.5% (95% CI 98.3% to 99.9%). 77 If, in our study, the ASQ-3 performed similarly and if we had employed gold standard face-to-face assessments, one would expect very few infants categorised by ASQ-3 as having no impairment to then be diagnosed with developmental problems in face-to-face assessments. Our use of a self-report questionnaire completed by health professionals to help decide whether or not infants had developmental impairments could also be criticised. Using this method of data collection, we do not know whether or not those completing questionnaires did so with reference to medical records or other knowledge of infants. However, very similar questionnaires have been used in previous authoritative cohort studies78 and incorporation of such health professional data into analyses in this trial did not affect outcomes.

Smoking outcomes at 2 years

This was the first trial of a smoking cessation intervention to monitor longitudinal smoking rates within pregnant trial participants for as long as 2 years after delivery; therefore, the data are novel. Although data were self-reported and we obtained no information on smoking behaviour for around 40% of participants, sensitivity analysis suggested that our assumption that non-respondents at 2 years were smokers was appropriate.

Economic analyses

We present a within trial incremental cost-effectiveness analysis, calculating the cost per additional quitter associated with NRT patch use. Owing to the issues around measuring the EQ-5D in a population of pregnant women, we did not propose to calculate QALYs, which reduces the comparability of our cost-effectiveness outcome with different health-care programmes. The trial did not find a difference in outcomes that would have enabled the projection of longer-term costs and benefits of NRT use, thus precluding conclusions about the long-term cost-effectiveness of NRT patch use in pregnancy.

Smoking outcomes

In contrast to the negative primary outcome recorded at delivery, the increased cessation until 1 month after randomisation was of a similar magnitude to that seen following NRT use by non-pregnant smokers. 18 The lack of a statistically significant longer-term effect from NRT may be explained by the low adherence rates in the trial. One reason for the apparently low adherence is that many participants were specifically instructed to not use NRT when smoking and many failed to quit and restarted smoking. However, this does not explain why only 58% (101/173) of participants who were abstinent at 1 month accepted a second month’s supply of NRT. Other NRT trials in pregnancy have reported similarly low rates of adherence: two studies of NRT patches found median durations of NRT use of 2 weeks34 and 3 weeks,27 and, in a trial that tested 2-mg nicotine gum, this was used for just over 5 weeks. 35 Outside pregnancy, most smokers who attempt to quit with NRT discontinue this within 1 month because they either start smoking again or they believe it is not working. 79 Nevertheless, despite such reports of adherence with NRT being lower than recommended,79 trials18 and cohort data from routine clinical care80 both demonstrate that NRT used by non-pregnant smokers is effective for smoking cessation. Adherence with NRT is potentially an important influence on the efficacy of this treatment in pregnancy; NRT cannot have an effect if it is not used and reduced adherence with NRT in later pregnancy could explain the lack of effect shown at delivery. Therefore, potential influences on adherence in the trial require further exploration.

Trial participants often discontinue treatments after experiencing AEs or side effects; however, only 8.8% of women in the NRT group reported stopping patches after AEs. In previous similar trials, the discontinuation rate was 12% for nicotine gum35 and 4.4% for nicotine patches or placebos. 27 These AE rates are much lower than participants’ rates of early treatment discontinuation and hence can only partially explain this. Treatment discontinuation could be explained by increases in maternal nicotine and cotinine clearance in pregnancy; increases of 60% and 140%, respectively, have been reported to occur by 25 weeks’ gestation. 81 Such increases would be expected to reduce NRT-generated nicotine levels and increase users’ withdrawal symptoms. It is possible that, for NRT to consistently ameliorate women’s nicotine withdrawal symptoms and be effective throughout pregnancy, a higher dose is required. 82 However, this trial did not include assessment of nicotine metabolism and did not assess withdrawal symptoms. In addition, factors other than increases in metabolism may explain low NRT adherence rates in this and previous trials.

In summary, this trial provides no evidence that NRT, as the 15 mg per 16 hours transdermal patch, is effective for smoking cessation in pregnancy. Women used the patches for shorter periods than recommended and this could explain negative trial findings. It is possible that, even with the low adherence displayed by trial participants, NRT actually does have a positive effect and, with a larger sample size, the 1.8% absolute difference in favour of treatment with NRT would have become statistically significant. However, a recent meta-analysis of this trial’s findings with all previous similar studies (n = 1745) does not suggest this. In this analysis, the RR for cessation in later pregnancy after using NRT compared with control was 1.33 (95% CI 0.93 to 1.91). 31 Even if the small difference between groups in this trial does represent a real effect of NRT that has not been statistically proven, it seems unlikely that, based solely on this small effect on maternal smoking behaviour, NRT used in pregnancy would be considered clinically useful. The number needed to treat (calculated using trial data) is 54, which means that if the magnitude of effect observed in this trial were to be found statistically significant (e.g. in a future meta-analysis), 54 women would require treatment with NRT to produce one successful quitter. This success rate would be unlikely to impress clinicians who would probably be inclined to use other methods of cessation support for pregnant smokers.

Other outcomes measured at delivery

Rates of adverse outcomes were similar between groups with the exception of caesarean deliveries, which were unexpectedly more frequent in the NRT group. This is difficult to explain and is likely to be a chance occurrence. However, caution is warranted when interpreting the overall similarity in birth outcomes between trial groups. Because many adverse birth outcomes are quite rare, some comparisons may have limited power and the low rate of treatment adherence makes it difficult to attribute the presence or absence of differences in birth outcomes to NRT.

Outcomes measured after delivery

This study has provided the first evidence that a smoking cessation intervention delivered to pregnant women can influence the development of their offspring. We found that NRT, used in pregnancy for smoking cessation, without having any statistically significant, long-term effect on maternal smoking in pregnancy, had a positive impact on subsequent child development. The most likely explanation for this impact lies in the altered smoking behaviour of women who were randomised to NRT. For example, the transient doubling of quit rates in the first month after trial enrolment and until around 20 weeks’ gestation could have occurred at a crucial time for infants’ brain maturation, resulting in the greater survival to 2 years of age with no developmental impairment. The slightly higher, but non-significant, quit rates observed from delivery until 2 years and the reduced accompanying exposure of infants to domestic environmental tobacco smoke may also have had additional positive effects. If the impact of NRT on infants is mediated through increased maternal smoking cessation, any effective smoking cessation intervention used by pregnant women would be expected to have similar effects. However, it remains possible that any protective impact of NRT on infants’ development arises directly from the impact of nicotine itself and is not mediated by reduced fetal or infant exposure to tobacco smoke toxins. Irrespective of whether the impact of NRT on infants is direct or mediated (indirect), this trial is reassuring about NRT use in pregnancy. Although in controlled laboratory studies, nicotine has been shown to cause fetal tachycardia, albeit to a lesser extent than smoking,21,83,84 this trial provides no evidence that NRT is harmful. Similarly, as nicotine is a neurotoxin and can cause behavioural problems in young rodents,21,83 there has previously been concern that NRT might harm infants’ developing nervous systems and that it could cause behavioural problems and poorer academic achievement in smokers’ children. 22 However, study findings clearly suggest that nicotine is unlikely to be responsible for these problems.

As mentioned above, findings from this trial require confirmation; future studies of NRT and other smoking cessation interventions in pregnancy should follow-up infants and trial participants for at least 2 years after delivery to facilitate this. Additionally, the cohort of trial infants from this trial should be followed after 2 years, to determine whether or not impacts on child development attributable to NRT persist into childhood. If further studies can replicate findings or confirm that these benefits persist later into childhood, then this would provide reassurance that these findings did not occur by chance, which is important both scientifically and economically. Intervention costs were relatively low (< £100 per participant) and, should the impacts on infants identified in this study be confirmed, it is likely that NRT used in pregnancy for smoking cessation would be viewed as cost-effective, potentially generating cost savings for the NHS.

The trial had a relatively pragmatic design, with reasonably broad inclusion and few exclusion criteria, therefore, results are likely to be generalisable to most pregnant smokers. However, women recruited to the trial were between 12 and 24 weeks’ gestation and, therefore, our results may be less applicable to those trying to quit using NRT earlier or later in their pregnancy. For example, at earlier gestations, nicotine metabolism may not yet have increased and so the dose of NRT used in this trial may be sufficient to help such women to quit successfully. Our 1-month findings suggest that this could be likely and as we found no safety issues, future studies could consider including women earlier in pregnancy. Similarly, as we only included women who smoked at least 5 cigarettes a day, the results cannot necessarily be extrapolated to those who are very light smokers and less addicted to nicotine and they may also be more likely to be able to successfully quit with the dose of NRT used in this trial.

Trial applicants

Chief investigator: Tim Coleman.

Co-applicants: James G Thornton, John Britton, Kim Watts, Michael Coughtrie, Clare Mannion, Neil Marlow, Christine Godfrey.

Trial manager

Sue Cooper.

Trial management group

Tim Coleman, Sue Cooper, James Thornton, John Britton, Sarah Lewis, Kim Watts, Neil Marlow.

Trial administrators

Anne Dickinson, Rachel Whitemore.

Trial Steering Committee

Peter Brocklehurst (chair), Carol Coupland (independent statistician), Peter Hajek, Sue Maguire (PPI/lay member), Michael Murphy.

Data Monitoring and Ethics Committee

Janet Peacock (chair), Christopher Butler (from January 2009), David Field, Khalid Khan (until October 2008).

Statisticians

Sarah Lewis, Matthew J Grainge, Jaspal Taggar.

Health economists

Christine Godfrey, Holly Essex, Steve Parrott.

Research team

In addition to those listed above, the complete trial team includes:

Research staff: Janet Brown, Yvette Davis, Anne Dickinson, Caroline Dixon, Fiona Holloway, Joanne Lakin, Jayne Platts, Farzana Rashid, Amanda Redford, Cara Taylor, Rachel Whitemore.

Principal investigators (in recruiting centres): Jonathan Allsop (Derby Hospitals NHS Foundation Trust), Simon Cunningham (Mid Cheshire Hospitals NHS Foundation Trust), Karen Glass (Sherwood Forest Hospitals NHS Foundation Trust), Vince Hall (East Cheshire NHS Trust), Khaled Ismail (University Hospital of North Staffordshire NHS Trust), Margaret Ramsay (Nottingham University Hospitals NHS Trust – QMC campus), James Thornton (Nottingham University Hospitals NHS Trust – City Campus).

Midwife leads (in recruiting centres): Sheena Appleby, Denise Bailey, Linda Gustard, Emma Haworth, Grace Hopps, Amanda Lindley, Chris Kettle, Colleen Pearce, Dymphna Sexton-Bradshaw, Julia Savage, Sandra Smith, Sheila Taylor, Alison Whitham.

Primary care trust and NHS SSS staff: Barbara Brady, Michelle Battlemuch, Wendy Dudley, Rochelle Edwards, Lorraine Frith, Indu Hari, Catriona Holden, Linda Hoskyns, Paul Jackson, Giri Rajaratnam, Deborah Richardson, Lucy Wade, Maureen Whittaker.

QMC pharmacy: Bernie Cook, Sheila Hodgson (lead pharmacist), Lisa Humphries, Bernie Sanders (quality controller).

University of Nottingham: Dan Simpkins (Clinical Trials Unit – database), Luis Vaz (assisting with secondary adherence analyses), Yvanna Kurlak, Clare Randall and John Taylor (assisted with trial administration).

University of Dundee: Sheila Sharp (sample analysis – baseline and 1-month COT).

We would also like to thank the following for their contributions to the study:

Dr Mira Doig and ABS Laboratories Ltd (Welwyn Garden City, UK), for analysis of cotinine in baseline blood samples.

Salimetrics Europe Ltd (Newmarket, Suffolk, UK) for analysis of cotinine in delivery saliva samples.

Nicotine replacement therapy: 50 adverse events in 46 women

Application site reactions (n = 33), nausea (n = 7), headache (n = 3), dizziness (n = 2), palpitations (n = 2), dyspnoea, oropharyngeal pain, sensory disturbance (all n = 1)

Placebo: 38 adverse events in 32 women

Application site reactions (n = 15), nausea (n = 7), dizziness (n = 5), headache (n = 3), irritability (n = 3), dyspepsia (n = 2), fetal hypokinesia, influenza like illness, palpitations (all n = 1)

Overnight hospital admissions for other pregnancy complications

In total there were 143 events with hospital admissions in 95 women (NRT group) and 146 events with hospital admissions in 94 women (placebo group) (admission for any reason, not only pregnancy related).

Hospital admission for less frequent pregnancy-related events

Other, less frequent, adverse events that occurred in < 3% of women or infants and are not logically grouped together