Evaluation of a complex healthcare intervention to increase smoking cessation in pregnant women: interrupted time series analysis with economic evaluation

Objectives To evaluate the effectiveness of a complex intervention to improve referral and treatment of pregnant smokers in routine practice, and to assess the incremental costs to the National Health Service (NHS) per additional woman quitting smoking. Design Interrupted time series analysis of routine data before and after introducing the intervention, within-study economic evaluation. Setting Eight acute NHS hospital trusts and 12 local authority areas in North East England. Participants 37 726 records of singleton delivery including 10 594 to mothers classified as smoking during pregnancy. Interventions A package of measures implemented in trusts and smoking cessation services, aimed at increasing the proportion of pregnant smokers quitting during pregnancy, comprising skills training for healthcare and smoking cessation staff; universal carbon monoxide monitoring with routine opt-out referral for smoking cessation support; provision of carbon monoxide monitors and supporting materials; and an explicit referral pathway and follow-up protocol. Main outcome measures Referrals to smoking cessation services; probability of quitting smoking during pregnancy; additional costs to health services; incremental cost per additional woman quitting. Results After introduction of the intervention, the referral rate increased more than twofold (incidence rate ratio=2.47, 95% CI 2.16 to 2.81) and the probability of quitting by delivery increased (adjusted OR=1.81, 95% CI 1.54 to 2.12). The additional cost per delivery was £31 and the incremental cost per additional quit was £952; 31 pregnant women needed to be treated for each additional quitter. Conclusions The implementation of a system-wide complex healthcare intervention was associated with significant increase in rates of quitting by delivery.


Appendix 2
Additional data relating to variable definitions, models 1-3 and missing values From status of smokers at booking, we were able to identify smokers at delivery and those who had quit at delivery: 3-Smokers at delivery (N=6882) includes: -Recorded smokers at delivery (maternity) 4-Quitters by delivery (N= 3712) includes: -Recorded smokers at booking that are not recorded as smokers at delivery 5-Unknown smoking status (N= 82) includes: -Smokers at booking with unknown smoking status at delivery (cannot determine whether they quit or not) Engagement with smoking cessation services Setting a quit date (N=881) includes records of having set a quit date(s), one or more quit date(s) recorded, SCS outcome as having set a quit date.
Referrals to smoking cessation services N=5613 referrals, include all observations with any record with SCS of: -Whether a quit date was set -Whether appointment(s) were attended or booked -Ever being referred to SCS, any referral date -Being sent information or contacted by SCS -Any record of smoking status for 'quit at 4 weeks' Estimated first appointment date First appointment reflects the first potential contact within babyClear©; booking dates were not included as a variable in the dataset.
Estimated conception date = Delivery date -gestational age at delivery Estimated first appointment date = Estimated conception date + 11 weeks Before or after intervention A response variable is needed that accounts for whether the first booking occurred before or after the start of the intervention. In some cases of first referrals occurring prior to the intervention, further referrals and appointments may have taken place after intervention introduction. This classification considers these pregnancies as "prior" which is to be interpreted as not having received the full intervention protocol.
Strategy used for initial contact and referral with smokers Developed from information gathered during process evaluation (SJ, personal communication) regarding different approaches to arranging contact between women and smoking cessation services Categories: None: SCS contacted woman after first appointment with midwife Appointment: Referred directly SCS by midwives during the first appointment Early contact: A specific enhanced strategy for early contact with smokers before their first appointment with a midwife We used an area based measure of deprivation based on lower super output area (LSOA). The LSOA was linked directly to maternal postcode and the overall score for Indices for Multiple Deprivation (IMD) was obtained. IMD scores for each mother were then partitioned into deprivation quintiles by dividing the range of scores into five equal groups. We partitioned the range of IMD scores based on LSOA for the northeast region as a means of generating a measure of deprivation that may be more representative of local area as opposed to national statistics where it has been suggested that the distribution of deprivation may unequal between geographical areas. The IMD quintiles were ranked and reclassified as ordinal variables according to the magnitude of deprivation. Quintiles were reclassified 1-5 from the least deprived to the most deprived scores.   Generalized linear mixed-effects model analysing the probability of quitting by delivery. The model assessed the smoking status of each individual mother at delivery (smoker or quitter)

Models
following assessment of smoking status at booking. We used a binomial error distribution. A random intercept for trust which allows for heterogeneity in the baseline probability of quitting by delivery between observations recorded at different trusts.  A linear mixed-effects model of the log value of weight at birth for each child. A random intercept for trusts allows for heterogeneity in the baseline weight between observations recorded at different trusts. A quadratic term was used to describe the relationship between gestational age at birth and birthweight. The results show that up to 40 weeks gestation, birthweight increases significantly until a threshold is reached. Beyond this 40-week threshold, birthweight significantly decreases. The linear term in the model, described by each additional week beyond 40 weeks gestation provides an assessment of the increase in birthweight. In contrast, the squared term, which is used additively in the model, describes the downward inflection of the curve and is therefore negative.

Model of variation in monthly count of referrals
Data were aggregated by month for this analysis, resulting in a total of 132 observation points through time across all trusts 8 trusts. The response variable was concerned with a subset of the population (count of smokers referred), however the model structure included an offset for the monthly sum of observations per trust. In the initial (non-aggregated data), from the total of N=37726 observations, N=36907 were included in the analysis.
From the initial dataset, a total of N=819 records of women were omitted due to either missing records (gestational age) or because information was unavailable to allow initial booking dates to be estimated.
Missing information on gestational age at delivery (N=672, information required to estimate relation to intervention in time, see table 2.7 and 2.8) Observations during the initial months of data collection in each trust were omitted when information for estimated first booking dates was incomplete (N=41 Table 2

Model of probability of quitting by delivery
The number of smokers at booking was N= 10676. This cohort was reduced to N= 9967 observations due to N=709 missing records (

Model of birthweight
The initial dataset consisted of N=37726 deliveries. This was reduced to N=22826 deliveries due to N=14900 missing records (Table 2.10).