scripts/03_model_engine/01_clinical_pathway.R puts each year’s patients through the care pathway, once with the service and once without it (usual care in 2019). It counts patients and admissions separately: quitting happens to patients, while screening, assessments and NRT happen at admissions.
The steps follow the NHS Inpatient Tobacco Dependence Treatment Care Bundle in the NHS Standard Treatment Plan (Papadakis et al. 2023):
| Care bundle (Papadakis et al. 2023) | What it asks for | In the model |
|---|---|---|
| Admission | Ask every patient whether they smoke; advise those who do; offer NRT; refer to the trust’s tobacco dependence team (opt out) | Screening (pr_screened), very brief advice (pr_VBA) and NRT from ward staff (pr_pharma_vba) |
| Inpatient | A specialist assessment and treatment plan within 24 hours, with follow-up visits for longer stays | Assessment by an advisor, in person for patients staying a night or more and by phone for shorter stays (pr_assessment_ge1, pr_assessment_lt1, frac_inperson_*), and NRT (pr_pharma_inperson, pr_pharma_phone) |
| Discharge | At least two weeks of stop smoking aids; a referral for a 10 to 12 week course of specialist support after discharge | A quit attempt (pr_quit_attempt_seen) and registration for specialist support after discharge (pr_register_after_discharge_inperson, _phone) |
| After discharge | A call from the trust team at 7 to 14 days and a follow-up at 28 days, ideally in person with a carbon monoxide test, 10 minutes each | Follow-up for patients assessed in person (pr_call_seen), within the advisors’ costed time (below) |
The rates with the service come from services in England: screening and very brief advice from the CURE project (Evison et al. 2020), assessment and NRT from the London service (Robins et al. 2024), quit attempts from the national data (Agrawal et al. 2026), and registration for support after discharge and its quit success from South Yorkshire (Cherodian et al. 2025). All of them are in user_inputs/scenarios_control_panel.csv with their sources.
The comparator is usual care in 2019, when the NHS Long Term Plan commitment was made, from the British Thoracic Society’s audit that year (Mangera and Devani 2020): 76.8% of patients asked whether they smoke (10,484 of 13,647), 43.7% of smokers given very brief advice (asked whether they would like help to quit; 1,105 of 2,528), and NRT offered to 30.7% (777 of 2,528; the audit records offers, not receipt). One in eight smokers was referred to a stop smoking service, all among those advised, so pr_refer_vba is 0.125 x 2,528 / 1,105 = 0.286 of those advised. The 2019 audit doesn’t report how many referred patients used the service; in the 2021 audit 71 of the 359 smokers referred (19.8%) attended a follow-up service (Devani et al. 2023), which is pr_register_given_referral_vba. Advised patients who are not referred, or don’t register, make an attempt on their own at pr_quit_attempt_vba (6%, an assumption). The 2019 audit describes hospitals without a dedicated team, where ward staff did the referring.
With the service, ward staff refer smokers to the trust’s tobacco dependence team instead, as the Standard Treatment Plan specifies, and the advisors arrange support after discharge for the patients they see (Papadakis et al. 2023). Patients the team doesn’t see are not referred on by ward staff, so pr_refer_vba is 0 in the treatment arm. They can still try to quit on their own after very brief advice.
The benefit of the service comes from what its advisors add to ward care, not from ward care itself. Compared with usual care:
So a smoker seen by an advisor is about four times as likely to be quit at 4 weeks as a smoker given no help, and one given very brief advice alone about 1.24 times (below).
Each of these is a row of user_inputs/scenarios_control_panel.csv with a value for each arm: treatment is with the service and control is usual care. To test a different view of what the service adds:
| To test | Change | Already run as |
|---|---|---|
| Usual care that screens, or screens and advises, as many patients as the service | pr_screened, pr_VBA (control rows) |
Sens_Screening_Equal, Sens_Screening_Advice_Equal |
| A service whose ward staff also refer the patients the team doesn’t see | pr_refer_vba (treatment row), for example 0.286 as in usual care |
– |
| How many patients the team sees | pr_assessment_ge1, pr_assessment_lt1 (treatment rows) |
ASSESSMENT_BEST |
| How many register for support after discharge | pr_register_after_discharge_inperson, _phone (treatment rows) |
REFERRAL_BEST |
| Quit attempts after very brief advice | pr_quit_attempt_vba, quit_rate_4w_self_vba |
– |
| Quitting with no help | pr_quit_4w_background |
Sens_Background_Low, Sens_Background_High |
To test one of these as a scenario, add a column (chapter 8, “Adding a scenario”) and run main.R. To change the basecase, run the three calibrations first: the quit rates are rescaled to the national data, and coverage and follow-up admissions depend on the quits.
Patients with smoking-related conditions are often admitted more than once a year. Applying quit rates to admissions would count the same person quitting more than once, so quits are worked out for patients (unique_individuals, the patients in each group from HES), and the work in hospital for admissions.
Patients are screened at their first admission, and screening finds the smokers among them. Identified smokers are split by length of stay, using the share of each group’s admissions with a stay of a night or more (pr_los_ge1, from HES): those staying a night or more are assessed at pr_assessment_ge1 and the rest at pr_assessment_lt1. In the basecase both are 60.2%, the share of all identified smokers assessed in two London hospitals (Robins et al. 2024), whatever their length of stay; patients staying a night or more are seen in person and shorter stays by phone. Every assessed patient has first had very brief advice. The identified smokers not assessed are given very brief advice only, or nothing.
Patients seen by an advisor make a quit attempt at pr_quit_attempt_seen, and of those, pr_register_after_discharge_inperson (52.5% in the basecase (Cherodian et al. 2025)) register for specialist support after discharge, or pr_register_after_discharge_phone for patients assessed by phone. The two are equal in every scenario: a phone assessment may be a weaker intervention, but there is no evidence on registration after one, so this is an assumption. Of those registered, pr_quit_support_after_discharge (78.8%) set a quit date with the support, and 61.1% of those attempts are quit at 4 weeks (quit_rate_4w_after_discharge, before quit_rate_scale). Every other attempt is made without that support, at a lower quit rate:
# Extract from 01_clinical_pathway.R: attempts without support after discharge
# are all attempts less those with it
pathway_dt[, total_attempts_inperson := group_seen_inperson_indiv * pr_quit_attempt_seen]
pathway_dt[, ssa_inperson := group_seen_inperson_indiv * pr_quit_attempt_seen * pr_register_after_discharge_inperson * pr_quit_support_after_discharge]
pathway_dt[, self_attempt_inperson := total_attempts_inperson - ssa_inperson]Attempts without that support succeed at quit_rate_4w_inperson_self (20.2%, before quit_rate_scale) for patients seen by an advisor, and at quit_rate_4w_self_vba (18.3%) for patients given very brief advice only. The second is set with pr_quit_attempt_vba (6%) so that very brief advice has about the effect found in trials. The 6% is close to the 6.6% that NICE’s economic modelling took for very brief advice (Filby and Taylor 2018), but that is a 12-month quit rate, not a rate of quit attempts, so it is only a starting point; what supports the pair is the check that follows. Cheng et al. pooled 13 trials of advice lasting 3 minutes or less against no advice: self-reported abstinence under 6 months was 1.22 times as common (95% CI 1.01 to 1.47), and quit attempts barely changed (1.03, very low certainty) (Cheng et al. 2024). In the model a smoker given very brief advice alone is 1.24 times as likely to be quit at 4 weeks as one given none; R/calibrate_pathway.R prints this after each calibration, beside Cheng et al.’s estimate (CHENG_VBA_RR in 00_config.R). The model gets there through more attempts rather than more successful ones, which changes the count of quit attempts but not of quits. In usual care, ward staff also refer some advised patients to a stop smoking service: 5.7% of those advised register and quit at the supported rate, and the ratio rises to about 2, above Cheng et al.’s interval, although several of their trials offered referral too. That makes usual care look better, not worse, so it makes the service’s extra quits smaller.
The Standard Treatment Plan gives five routes for specialist support after discharge (Papadakis et al. 2023): the local authority stop smoking service; the NHS Community Pharmacy Smoking Cessation Service; the trust’s own team, for 3 to 12 months; the community mental health team; and a digital service. Areas use different mixes, and some trusts give the support themselves. The model has one step for support after discharge, whatever the route: the registration and quit rates are from a local authority service (South Yorkshire), and the cost is that of local authority services (chapter 6). The community mental health route is for mental health patients, outside the model’s scope. The quit success of the other routes is not published on the same basis, so they are taken to be like the local authority service.
Smokers quit without any help from the pathway at a background rate, pr_quit_4w_background (2%), the share quit at 4 weeks, covering both making an attempt and staying quit. The rate belongs to the smokers, not to usual care, so it doesn’t depend on screening or advice: it applies to every smoker who makes no quit attempt through the pathway, whether not screened, not advised, or advised or seen without going on to make an attempt. It is the same in both arms, is not scaled by quit_rate_scale and has no odds ratios by group. A pathway attempt replaces the background chance for the smoker who makes it, so the service is credited with each attempt’s success over the background rate, not the whole of it. Every pathway quit rate (after quit_rate_scale and the odds ratios) must be above the background rate, and the code checks this in every row, so advice or an assessment always leaves a smoker more likely to quit than no contact.
Economic models of stop smoking interventions assume background quitting of about 2% a year, net of relapse (EQUIPT (Coyle et al. 2018); NICE public health modelling uses 2%, with 1.2% in sensitivity analysis). Those are long-term annual rates. Here the rate is at 4 weeks, and 42% of 4-week quitters are still quit at 12 months, so 2% at 4 weeks is about 0.8% at 12 months, below those models’ assumption. Sens_Background_Low (1%) and Sens_Background_High (4.8%, which is 2% at 12 months) test the rate. Tab_19b shows the background quits on their own row.
The 4-week quits are followed to 6 months (54% still quit), 12 months (42%) and for life (65% of those quit at 12 months), from the curve of continuous abstinence in trials of NRT and from Stapleton and West (Stapleton and West 2012).
Three parts of the pathway are set to the national results of inpatient tobacco dependence services in England in 2024 (Agrawal et al. 2026), in which 41,258 of 185,195 recorded smokers (22.3%) commenced a quit attempt with the service and 8,321 of them (20.2%) were quit at four weeks, counting those lost to follow-up as smoking:
pr_quit_attempt_seen), so that 60.2% x 37.0% = 22.3% of identified smokers make a supported quit attempt, as nationally. The differences by age, sex and deprivation in the equity panel come from the same data and keep this national average.quit_rate_scale, so they keep their relative sizes but sit on the national, intention-to-treat basis. The factor is set by the pathway calibration (R/calibrate_pathway.R) so that the model’s supported quit attempts, by either route, succeed at 20.2%. Four-week quits are proportional to the factor, so one run finds it exactly.The calibration also checks that the basecase sees 60.2% of identified smokers and has 22.3% making a supported quit attempt, and stops if either is more than 0.5 percentage points out. Tab_MSF_Supported_Attempts shows the share seen, supported attempts and their quit success for every year of a run.
Each identified smoker’s first admission counts, and so do their readmissions in the year, fewer for those who quit: the readmissions (admissions per patient minus one) are reduced by the share of identified smokers who quit at 4 weeks.
# Extract from 01_clinical_pathway.R
pathway_dt[, identified_smokers_adms := identified_smokers_indiv +
identified_smokers_indiv * (get(ADMISSION_MULTIPLIERS) - 1) * (1 - effective_quit_rate)]The assessments, NRT and follow-up calls are then counted at these admissions, split by length of stay in the same way as patients. In-person and phone assessments are costed at their own times (45 and 20 minutes, chapter 6).
The model counts follow-up after discharge for 81.3% of patients assessed in person (pr_call_seen, the share followed up at 4 weeks in the CURE project (Evison et al. 2020)); for a patient assessed by phone, the assessment is the first call. The follow-ups are not costed separately: the advisors make them within the time costed for the service. The advisors needed are worked out from the assessments, at 5 a day (chapter 6). As a check, 5 assessments of 45 minutes take 225 minutes, and the two 10-minute follow-ups for 81.3% of those patients take about 80 minutes, about 300 minutes of a 450-minute working day, leaving time for inpatient follow-up visits, travel between wards and records. In the CURE pilot, whose 45-minute consultation these times follow, the nurses assessed 3.7 patients a day and made 2.1 follow-up contacts after discharge for each, at 2, 4 and 12 weeks (3,084 contacts for 1,450 patients) (Evison et al. 2020): about 250 minutes a day at the same 45 and 10 minutes, against about 300 in the model. Extended support from the trust team after these follow-ups, one of the routes of specialist support after discharge, is not in the trust’s costs: it is part of the cost of support after discharge (chapter 6).