INTRODUCTION

Pharmacotherapies continue to be a central focus of clinical efforts toward successful tobacco control. Seminal reviews1,2,3,4,5 and recommendation statements6,7,8 consistently suggest that (1) all tobacco users should be offered brief advice to assist with quitting, and (2) unless medically contraindicated, all smokers should be offered pharmacotherapy. Unfortunately, uptake of cessation pharmacotherapies remains very low.9,10,11,12 Only about half of all smokers make a quit attempt in any given year,13 and less than 25% of these quit attempters use pharmacotherapy.9,14 Clearly, there is a strong need to identify new methods to get more smokers to use better treatments.

Medication sampling refers to providing short, starter packets of nicotine replacement therapy (NRT), either as a stand-alone regimen or added to other treatment protocols. The intent is to engage smokers in the process of quitting, regardless of their intention to quit, and without any requirement or expectation to quit immediately. Within the sampling exercise, instructions are kept minimal to increase scalable potential for provision within clinical settings. We previously tested medication sampling in four separate trials,15,16,17,18 one of which was implemented within primary care settings as a hybrid efficacy/effectiveness trial.16,19 Results from these trials were generally positive, yielding clinically and statistically significant improvements in cessation.

Methods19 and outcomes16,20,21 of these trials have been reported elsewhere. Briefly, within a cluster-RCT, we enrolled 1245 smokers across the motivational spectrum, across 22 primary care clinics throughout South Carolina between 2014 and 2018. All interventions were delivered directly by healthcare providers during routine clinic visits. In the “standard care” arm, providers gave a bag to all participants, including standard smoking cessation information and brochures for the state Quitline. For the “NRT” arm, a 2-week supply of nicotine patch (14 mg) and lozenge (4 mg) was added to the standard care package. Brief advice was given to all patients, and all providers were trained in the 5As (ask, advise, assess, assist, and arrange) at study initiation. There was no other intervention provided to either group. The trial showed that 6-month abstinence rates of NRT sampling and standard care are 12% and 8%, respectively.16

An enduring question about medication sampling is whether this strategy confers any long-term cost effectiveness. On one hand, the intervention is inexpensive and easy to deliver—especially in primary care, where 70% of all smokers visit at least once annually.22,23 NRT sampling does not require extensive primary care involvement, and explanation of rationale to both smoker and provider is face valid. On the other hand, absolute rates of quitting are low. Moreover, many smokers who are given medication samples will not use them at all (~ 40% in our trial), resulting in a waste of resources.

Thus, the “return on investment” of medication sampling is unclear. Simple estimations from our primary care trial suggest a cost per quit ~ $475, similar to or lower than Quitline-based medication give-away programs.24,25,26,27 However, any real-world application of medication sampling will depend on an in-depth assessment of cost effectiveness based on health outcomes that take years to materialize. Such an assessment should also provide information on the population-level effect of large-scale NRT sampling, as other interventions may have similar cost per quit attempt but may not be scalable. Simulation modeling, which we propose to undertake herein, is the only feasible method to calculate the cost effectiveness of an intervention with distal outcomes that take years for results to materialize.

METHODS

Markov Cohort Model Overview

Our approach is based on a Markov cohort model that simulates smoking behaviors over a lifetime. We took the structure of the model that was developed by Barnett et al.28 to accrue lifetime healthcare expenditures, quality-adjusted life years (QALYs), and life years (LYs) to conduct our cost-effectiveness analyses. The Markov cohort model consists of only three health states, “current smoker,” “former smoker,” and “dead” (Fig. 1). In each simulation cycle of 3 months, current smokers may quit, die, or continue to smoke, and former smokers may remain abstinent, relapse, or die. All transitions between health states are based on age-, sex-, and smoking status–specific mortality probabilities and smoking status–specific quit and relapse probabilities. Discounted costs, QALYs, and LYs over the lifetime are accrued based on projected lengths of time spent in each health state and the average healthcare expenditures associated with each health state. We obtained age, sex, and smoking status of the simulation cohorts and NRT sampling costs from our trial (Appendix Table 1), but all other model parameters were derived from the literature, as described in Barnett et al.28 For all literature-derived parameters, we adopted the underlying general population parameters unadjusted for a psychiatric patient population in the Barnett study. (See Appendix Tables 2–4 for key model parameters.)

Figure 1
figure 1

The Markov cohort model.

Full size image

Cost of NRT Sampling

The 2-week cost of NRT sampling was estimated to be $75. While the intervention consisted of both provider time and costs for cessation brochures, these added costs are not included because only the NRT sample differed across treatment groups. However, in subsequent one-way and probabilistic sensitivity analyses, we vary the intervention cost between $37.50 and $112.50, where the higher ranges can cover additional labor costs, or more costly or longer-duration medications.

Mortality Rates

We derived sex- and age-specific mortality rates for the general population directly from the 2017 Social Security Administration’s Actuarial Life Table.29 We then applied excess death hazards to estimate the mortality rates for smokers based on age and sex. These smoking-related excess mortality hazards vary by sex and three broad age brackets (24–54 years, 55–74 years, and 75+ years),30,31,32,33,34,35,36,37 but the underlying mortality rates differ by sex and 5-year age increments. Excess death hazards fall among former smokers after 5 years of abstinence.

Healthcare Expenditures

We obtained age- and sex-specific average annual healthcare expenditures from the Medical Care Expenditures Panel Survey (MEPS).38 We updated the latest available figures (from 2014) to 2019 expenditures using the cost inflation index from the Centers for Medicare and Medicaid Services’ National Health Expenditures Fact Sheet.39 We then multiplied these estimated age- and sex-specific expenditures by ratios representing how much more smokers and former smokers spend on healthcare versus non-smokers. These ratios were derived from healthcare charges in large employer’s health plans,40 comparing healthcare spending among current smokers, recent quitters (< 5 years), and long-term quitters (5+ years). Following patterns observed in the literature,41,42,43 we assume that former smokers have elevated healthcare spending around the time of cessation, but then experience a sustained reduction in spending after 5 years of abstinence.

QALY Weights, Natural Quit Rates, and Relapse Rates

We used age-, sex-, and smoking status–specific QALY weights derived from the literature to adjust total life years lived based on each cohort’s distribution of sex, age, and smoking status.44 We made the simplifying assumption that quality of life among former smokers does not depend on length of time since quitting.28 Smoking status varies over time according to the natural background successful quit attempt rates and relapse rates, which are determined by the length of abstinence.45,46,47,48 We used an annual natural quit rate of 4.3%.49,50 We adopted relapse rates of 60% within 1 year of quitting, which falls to 47% in year 2, 4% in years 3–5, 2% in years 6–9, and 1% after 10 or more years.48,51 We specifically chose high initial relapse rates that fall sharply after two full years of abstinence,48 given that a one-time, non-intensive intervention may not hold in the short run.

Discounting

All costs and outcomes were discounted at 3% per year. Costs were denominated in 2019 US Dollars. We used $100,000 per QALY as the threshold for willingness to pay.52

Cost-Effectiveness Analyses

We conducted the base-case cost-effectiveness analysis by running the simulation with one branch for NRT sampling and one for standard care. All model parameters were identical between the two cohorts except the initial distribution of former versus current smokers. Each cohort was 30.1% male, with everyone aged 50.72 years based on the mean age from our trial results. The model projected the cost and health outcome outputs for the two arms over the lifetime, from which we calculated the incremental cost (NRT samples) and the incremental benefit (savings in healthcare expenditures, QALY, and LY) of NRT sampling versus standard care.

One-Way Sensitivity Analyses and Probabilistic Sensitivity Analyses

We first conducted one-way sensitivity analyses, varying one parameter value at a time to assess model robustness to parameter changes. We then conducted a probabilistic sensitivity analysis (varying all parameters simultaneously), using 1000 independent random draws from each parameter’s distributions. We drew costs from a gamma distribution, mortality hazard ratios from a triangular distribution, and QALYs and all probabilities from beta distributions. These distributions were set to reflect ranges or confidence intervals of the respective parameters. (See Appendix Tables 2, 3 and 4). Using these 1000 incremental cost-effectiveness pairs between NRT sampling and standard care, we generated the “confidence interval” for our baseline cost-effectiveness results, taking into account the uncertainty of parameter values.

Counterfactual Analyses

As currently conceived, patients only receive a one-time sample of medication. Counterfactual scenarios allow us to model options for recurrent sampling. We conducted two counterfactual analyses, in which 50% of the participants who remained smokers could randomly receive another NRT sample in each quarter of the first 6 months (counterfactual A) or first 12 months (counterfactual B) after model initiation. We assumed, as a base-case scenario, that multiple NRT samples would have the same quit rates as the initial sample.

We constructed the model using TreeAge Pro 2021 (Williamstown, MA). Additional details on the model can be found in the Supplemental Appendix.

RESULTS

Baseline Characteristics

The clinic19 and participant16 characteristics have been described elsewhere; baseline participant characteristics for the Markov model are presented in Appendix Table 1. Six-month quit rates, within both NRT sampling (12%) and standard care arms (8%), represented the primary model inputs for our simulation model.

Cost-Effectiveness Findings

The results of the base-case model are presented in Table 1. Discounted lifetime follow-up healthcare expenditures for NRT sampling were $322,939, versus $324,079 for standard care. NRT sampling cost an additional $75 per person to implement, but resulted in a net savings of $1065 per person over a lifetime. The cohort receiving NRT sampling was projected to live on average 16.609 life years, or 0.01 life year more than the 16.599 life years realized under standard care. NRT sampling yielded 12.861 quality-adjusted life years (QALYs), or 0.01 QALY more than the 12.851 QALYs under standard care. Therefore, NRT sampling is a dominant strategy because it yields higher lifetime QALYs at a lower lifetime cost.

Table 1 Lifetime Cost-Effectiveness Model—Base-Case and Probabilistic Sensitivity Results
Full size table

One-Way Sensitivity Analyses

We considered many parameters, including effect sizes between 0 and 8% and time-dependent relapse rates ranging from 50 to 150% of the baseline relapse rates.48,51 These results, presented in Table 2, show that NRT sampling is the dominant strategy in all but one scenario in which NRT sampling provides no additional successful quits. We also conducted a threshold analysis and found that NRT sampling need only achieve an increase of 0.262% in abstinence relative to standard care to justify the NRT costs.

Table 2 One-Way Sensitivity Analysis of Select Model Parameters
Full size table

Probabilistic Sensitivity Analysis

Table 1 (right panel) provides the confidence intervals around the simulated costs and outcomes, as well as around the difference in costs and outcomes between NRT sampling and standard care. The results show that relative to standard care, NRT sampling decreases costs by $1061 (95% CI $1017–$1106) and increases discounted life years by 0.008 (95% CI 0.0081–0.0089) and discounted QALYs by 0.008 (95% CI 0.0085–0.0093). In Appendix Figure 2, we present an incremental cost-effectiveness scatterplot, showing that > 94.1% of these 1000 pairs are cost saving at the willingness-to-pay threshold of $100,000 per QALY gained.

Counterfactual Analyses

In Table 3, we show that from $1065 in net savings for a one-time administration of NRT sampling, the extended model projected a net savings of $1275 and $1412 in the counterfactual analyses, extending the NRT sampling period to 6 months and 12 months, respectively. QALYs gained also increased from 0.008 in the original one-shot model, to 0.013 and 0.015 in the two extended counterfactual models.

Table 3 Hypothetical Counterfactual Analyses
Full size table

DISCUSSION

Our model shows that NRT sampling increases lifetime QALYs and decreases lifetime costs relative to standard care in the great majority of scenarios. If we ignore the lifetime reduction in healthcare expenditures associated with sustained quitting and consider only the cost of implementation and QALY gains, the incremental cost per QALY was approximately $7500, which is considered cost effective. Given the totality of findings here, NRT sampling achieves robust results in cost savings and improves quality of life, even when overall quit rates are low.

The QALY improvements and cost savings may at first glance appear to be modest. It is therefore important to place these figures in context. First, our study results appear favorable in terms of cost per quit for tobacco cessation interventions. A recent meta-analysis53 suggests that the average cost per quit for pharmacological interventions was approximately $19,510.54,55 Behavioral interventions cost on average $11,416 per quit,27,56,57,58,59,60 and combined behavioral and pharmacological interventions were on average $14,662 per quit.53 NRT sampling, however, costs approximately $475 per quit. Moreover, a comprehensive assessment of 380 public health interventions between 2005 and 2018 by the UK’s National Institute for Health and Care Excellence revealed that the median cost per QALY gained was £1986 ($2641 US Dollars), with 21% of interventions found to be cost saving.61 In other words, our cost-saving NRT sampling is in the top 20% of all public health interventions, not just tobacco cessation interventions.

Moreover, NRT sampling is feasible and scalable because of its ease of implementation. Other interventions may need to recruit participants, provide a longer-term cessation program, and/or add a high-cost cognitive behavioral component.56 Numerous studies have evaluated the effectiveness of pharmacological interventions around the world in multiple settings,27,54,55,60,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77 ranging from $426 per QALY (telemedicine)59,78 to $340,000 per QALY (school-based education).79 Many of these studies were based on more intensive interventions and achieved higher quit rates than did NRT sampling herein. However, as a brief (~ 1 min) and pragmatic intervention, medication sampling offers significant scalability, particularly within primary care. Thus, low quit rates may be offset by high reach.

The results of our simulation are insensitive to manipulation of model parameters across plausible ranges. What drove our model results was simply differences in the initial distribution of current and former smokers. Any intervention that is sufficiently low cost, increases the quit rate, and implementable on a large scale can generate positive savings over a lifetime horizon. Based on our simulation results, a health plan that chooses to spend $75,000 ($75 * 1000) in NRT sampling on 1000 current smokers can on average expect to recover $1,140,000 in averted healthcare spending over the lifetime, if enrollees remain in the same health plan.

We also show that if subsequent reissues of NRT sampling retain the same rate of cessation as our original trial, the cost savings will increase with each re-issue. We acknowledge, however, that it is unclear whether repeated administrations of NRT sampling will yield similar quit rates. Theoretically, an intervention that failed the first time may not succeed on a second attempt. On the other hand, tobacco cessation often requires multiple attempts to be successful.80 Future research should consider examining the effect of multiple or different NRT samples.

Finally, higher cost and higher effect strategies will continue to drive policy efforts to reduce the burden of smoking. However, approaches such as NRT sampling, representing low-cost, modest-effect, high-reach interventions that are feasible and scalable, also have a place in the policy armamentarium to encourage smoking cessation. Such programs can also be administered in conjunction with established evidence-based tobacco treatment interventions.

A key question is to understand how durable the abstinence rates remain after NRT sampling. As shown by our one-way sensitivity analyses, the higher the relapse rate, the less cost saving NRT sampling becomes. At 50% of the relapse rates, net savings from NRT sampling reach $2220 per person, but at 150% of the base-case analysis, net savings fall dramatically to only $201. In fact, relapse rates of over 90% in both year 1 and year 2 will drop savings in subsequent healthcare spending so low that they will be less than the $75 cost of NRT. Future research should consider developing and testing additional pragmatic interventions to analyze and, if necessary, reduce relapse rates after the initial NRT sampling.

LIMITATIONS

Our model structure focused on population averages, with very little accommodation for established clinical and sociodemographic differences that may affect outcomes. Future modeling efforts should include other patient characteristics to model that are important to tailor and design policy. Abstinence rates from our study were also not biochemically verified, although there are plans to do so in future studies. In addition, future modeling efforts may also consider disaggregating health states into major smoking–related illnesses. Finally, many of the model parameters are drawn from studies that may be dated today. As new evidence is accumulated, future modeling should incorporate the latest information on how smoking affects health.

CONCLUSION

NRT sampling is cost saving ($1065 in savings per person) when we take into consideration reduced health expenditures over a lifetime for those who quit. When we ignore these lifetime cost reductions, NRT sampling costs approximately $475 per quit and $7500 per QALY gained. These figures are well within the acceptable range of $100,000 per QALY gained. While interventions with higher effectiveness exist, they are costlier to implement, and many are difficult to scale to a large population. NRT sampling should therefore be considered an appropriate and useful addition to the existing policy toolkit to reduce the mortality and morbidity burden of smoking.