RESEARCH DOSSIER
Smoker and Lifestyle Rating Factors in Life Insurance
Abstract. Smoker and lifestyle rating factors in life insurance are the measured or declared characteristics through which insurers classify expected mortality and determine acceptance, premium classes, or further underwriting requirements. Smoking is an unusually important test case because the causal evidence for the mortality consequences of combustible cigarettes is extensive, while the operational categories used in insurance are heterogeneous, historically contingent, and increasingly challenged by electronic nicotine delivery, nicotine replacement, dual use, cessation, and new verification methods. This research dossier investigates the distinction between epidemiological effect and insurable mortality differential; exposure and disclosure measurement; the interaction of tobacco with body composition, alcohol consumption, physical activity, and medical comorbidity; the actuarial and economic mechanisms of risk classification; temporal changes in behavior and risk; and the legal and ethical constraints on life-insurance pricing. The research combines documented primary insurer rules, Society of Actuaries experience and valuation resources, prospective cohort and pooled epidemiological studies, biochemical measurement research, and selected primary legal and supervisory instruments from the United States, the United Kingdom, and the European Union. It develops reproducible analytical illustrations of selection, exposure misclassification, mortality-cost decomposition, and sensitivity analysis; all numerical scenarios expressly designated illustrative are constructed examples rather than estimated industry rates. The synthesis establishes that insurer classification is a chain of linked empirical and institutional decisions: definition, observation, estimation, calibration, contractual use, explanation, contestability, and reassessment. Evidentially supported risk distinctions can lose accuracy through outdated definitions, portfolio selection, unobserved changes in conduct, or inconsistent treatment of alternative nicotine products. The social and contractual acceptability of a rating rule also depends on the local legal basis and its effect on access and consumer understanding. This monograph offers original comparative analysis and an operational research agenda without claiming an original actuarial theory, proprietary rating model, or universal premium differential.
Keywords: Life insurance; underwriting; smoker status; tobacco exposure; combustible cigarettes; electronic cigarettes; nicotine; cotinine; cessation; former smokers; lifestyle factors; mortality; insured lives; actuarial tables; preferred risk classes; selection; misclassification; risk-based pricing; accelerated underwriting; evidence validation; algorithmic fairness; insurance consumer protection; international regulation.
Introduction
Life insurance converts the uncertain time of death into defined monetary obligations. Its underwriting apparatus assesses applicants using information available at or around policy inception, while the insured event may occur decades later. Smoking and related lifestyle information enter that apparatus because they help differentiate expected death claims among applicants. A scientifically defensible classification, however, has to preserve the difference between a demonstrated causal mechanism, an observed statistical association, a measurable individual characteristic, and the contractual decision that uses that characteristic.
The U.S. evidence on cigarette harms establishes smoking as a major determinant of disease and premature death. The Society of Actuaries' smoker-distinct valuation tables show that the insured-life sector has institutionalized separate smoker and nonsmoker mortality structures. These facts provide the two principal scientific anchors of this study. Their connection is an empirical problem: rates observed in an entire population do not equal mortality among people selected for and accepted into insurance, and a health-related relative risk does not directly specify a contract premium.
The subject is examined on four nested levels. Biological and behavioral exposure describes what an applicant consumes, inhales, or does; observation describes the accuracy and duration of reports and tests; actuarial classification estimates mortality in the selected and exposed portfolio; and legal-contractual application determines which distinctions may be used, how they are communicated, and how adverse outcomes can be challenged. Each level has different data, inferential limits, and decision rights. Failure at an earlier level propagates downstream even when later premium calculations are mathematically correct.
The scope is individual life coverage and relevant group or simplified-issue analogues, with primary attention to mortality rather than disability or health-care reimbursement. The analysis includes tobacco and nicotine products, cumulative smoking history, cessation, alcohol, body mass and metabolic risk, exercise, and associated underwriting information. It discusses pricing and selection but leaves full reserve valuation, interest-rate risk, insurance distribution, and individual medical treatment to their respective disciplines. International comparison is framed as jurisdiction-specific law and supervisory practice, rather than as a single worldwide set of underwriting rules.
Research method and the standard of inference
The documentary corpus consists of peer-reviewed population mortality evidence, original actuarial research resources, official public-health evidence, a primary insurer underwriting guide, reinsurance research, and official regulatory or legislative texts. For quantitative statements, the underlying population, period, outcome, exposure definition, and adjustment structure are specified where consequential. Official landing pages are used for what they actually establish; downloadable proprietary insurer files and raw SOA experience tables were not independently re-estimated. The comparison does not claim that an underwriting decision for any individual can be reconstructed from public material.
The methodological priority is a claim-to-source correspondence. The Jha et al. (2013) cohort analysis establishes an adjusted population association under its specific inclusion and follow-up rules; it is not an insurer rate manual. A named SOA insured-mortality experience study concerns intercompany individually underwritten business, not all tobacco users in a national population. The UK consumer misrepresentation statute governs a defined legal relationship and cannot be generalized to every country. These are different types of evidence, and their respective conditions are retained throughout the synthesis.
Scientific Foundations and the Historical Development of Risk Classification
The scientific object: exposure, class, and mortality
In its narrow underwriting meaning, a smoker rating factor is a rule that links an observed or declared smoking-related characteristic to an expected mortality category and thence to a contractual decision. A lifestyle factor is a measured or reported behavior or behavioral correlate—such as alcohol intake, activity, or tobacco use—that may be relevant to risk selection. The class may be binary, ordinal, or multidimensional; its boundaries are properties of the insurance product and underwriting rules, not universal biological categories.
Several distinctions govern scientific interpretation. Current smoking refers to recent combustible tobacco use under a specified window; former smoking implies prior exposure and an elapsed abstinence interval; never smoking identifies a different exposure history. Nicotine exposure can originate from combustion, vaping, smokeless products, or therapeutic replacement. A laboratory result may identify an analyte or class of exposures without establishing the precise product, the historical intensity, or the individual's lifetime risk. Finally, a favorable underwriting class is an institutional designation determined by both health evidence and company rules. It should not be read as proof of clinical health.
The 2015 Valuation Basic Tables make the classificatory nature of insurer evidence visible: their published materials distinguish smoker/nonsmoker tables alongside sex, age conventions, and preferred-relative-risk constructs. The tables are actuarial reference bases, not universal tariff promises. They are also period-specific artifacts. Their maintenance notices describe corrections and subsequent improvements, reminding researchers that numerical artifacts require version control just as legislative texts and underwriting manuals do.
From mortality causation to insurance evidence
The modern case for recognizing smoking as a mortality-relevant factor is unusually robust. Smoking increases risks of cardiovascular disease, respiratory disease, and multiple cancers; these causal pathways are summarized in the CDC tobacco health overview. In a U.S. nationally representative linked survey, Jha and colleagues (2013) compared current smokers with never-smokers and reported adjusted all-cause mortality hazard ratios of approximately 3.0 in women and 2.8 in men between ages 25 and 79. The results refer to specified observation and follow-up periods and adjusted covariates. They do not imply a threefold premium or a threefold hazard at every age.
A separate prospective investigation, the Million Women Study, followed over one million UK women and documented large mortality consequences of continued smoking and large benefits of stopping before middle age (Pirie et al., 2013). Independent cohorts with distinct recruitment mechanisms contribute triangulation: their convergence supports the direction and importance of the exposure–mortality association, while differences in cohort composition, dates, products, and measurement restrict direct transport of any numerical magnitude.
The underwriting problem introduces selection. The applicant chooses whether to apply, which sum insured to request, and sometimes which disclosure channel to use; the insurer chooses eligibility, class and offered price; the applicant then chooses whether to accept and continue. A final insured portfolio is therefore an outcome of multiple filtering processes. Because a hazard ratio is conditional on membership and recorded covariates, its application to an insured cohort requires more than copying a figure from a population study. Insured-life risk may differ in baseline mortality, exposure prevalence, smoking intensity, age structure, education, comorbidity, and persistence.
The SOA 2009–2013 experience report documents observed mortality on standard individually underwritten issues and offers supporting pivot data and documentation. Later 2009–2016 insured-life reporting explicitly distinguishes experience by smoker classification and other dimensions. The reporting basis is crucial: actual-to-expected mortality, often abbreviated A/E, compares observed claims with claims expected from a chosen table, exposure, and definition. A lower A/E can result from better underlying mortality, more selective underwriting, a different mix, imperfect expected assumptions, or a combination.
Evolution of classification under new nicotine products
Historically convenient smoker/nonsmoker distinctions now face an expanding product landscape. A current combustible smoker, a person using only a nicotine patch, a daily vaper, an occasional cigar user, a recent quitter, and a person using several products can produce materially different exposure histories. A rule compressing them into one smoker class may be operationally simple but loses information potentially relevant to mortality. Conversely, multiplying classes before sufficient insured deaths are observed creates unstable estimates and costly verification.
The National Academies' review of electronic-cigarette mechanisms documents evidence of biological effects, including endothelial dysfunction and oxidative stress, while acknowledging uncertainty about many long-term outcomes of e-cigarette exposure (National Academies, 2018). It is methodologically unsound to assign the established lifelong combustible-cigarette mortality factor to a different exposure solely because both can involve nicotine. It is equally unwarranted to classify new products as riskless from the absence of mature, long-horizon mortality studies.
Swiss Re's analysis of changing nicotine underwriting identifies product type, dual use, frequency, starting age, and comorbidities as relevant classification dimensions. As an industry analysis, it provides a concrete description of an evolving underwriting problem; it does not establish a universally validated set of product-specific mortality relativities. The scientific next step is longitudinal exposure measurement combined with insured outcomes, not a premature universal ranking.
Why global prevalence is part of the underwriting background
WHO's 2025 tobacco prevalence report estimates country, regional, and global tobacco use trends through 2024, with projections to 2030. Changes in prevalence matter to the insured applicant pool, adverse selection, and the continuing suitability of risk class definitions. They do not mechanically determine portfolio mortality: insurance penetration, underwriting, demographic selection, product uptake, and cause-specific disease latency also change.
A further economic consequence follows. The interpretation of a stable smoker-rate surcharge may change as the smoker population becomes more selected. Even without a change in the biological effects of smoking, differences in who continues to smoke, who quits, and who buys coverage can alter observed class experience. Consequently, a table that retains a familiar label over time may refer to a changing empirical group. Longitudinal governance must track the meaning of labels as well as numerical mortality.
Exposure, Disclosure, and Biochemical Verification
What is being observed?
An underwriting application can ask about present use, use within the past year, any use during five years, daily quantity, years smoked, quit date, vaping, cessation therapies, or combined exposure. These variables address different risk attributes. Current use is a time-window variable; cumulative exposure is related to past dose; cessation duration captures behavioral change; product type describes the route and composition of exposure. To compare applicants or portfolios, the exact wording and effective date of the question must be preserved.
The primary industry document examined here, Aviva UK's Understanding Underwriting guide, gives a particularly instructive example. Its introductory question table (printed page 4) describes a five-year lookback for a nonsmoker definition. Its separate Smoking section (printed page 26) distinguishes recent use within 12 months, former use within five years, and never use; it additionally discusses vaping and cotinine. The statements serve different textual functions, but their apparent definitional tension is material. The guide therefore demonstrates why researchers cannot silently lift a single sentence into a universal insurer rule. An actual underwriting determination depends on the effective application questionnaire, product, insurer terms, and authoritative contemporaneous decision rules. The public guide alone cannot settle a disputed individual contract.
This example reveals a wider principle of documentary methodology: the relationship between a descriptive sales guide and an operative policy document must be identified. A secondary guide may summarize internal practice; the signed proposal and policy determine relevant representations and contractual obligations under applicable law. A research dossier should quote or paraphrase the guide only within its demonstrated scope. It should record inconsistent versions as source limitations, not invent an interpretation to conceal them.
Self-report, verification, and error structure
Self-report can be accurate, incomplete, or ambiguous. Some applicants misunderstand a questionnaire that uses “smoking” to mean all nicotine; others distinguish vaping from smoking in everyday language. Recall error grows for remote and intermittent exposures. Social desirability can influence answers when the anticipated premium is salient. Each mechanism can create misclassification without requiring deliberate fraud.
Benowitz and colleagues' biochemical verification review (2020; a 2019 research update) evaluates cotinine, carbon monoxide, total nicotine equivalents, tobacco alkaloids, and tobacco-specific nitrosamine biomarkers. Cotinine is informative about nicotine exposure, but nicotine may be delivered by cigarettes, electronic devices, or replacement therapy. Biomarkers have finite detection windows; absence of a recent analyte cannot prove lifetime abstinence. Markers that help distinguish tobacco-specific or combustible exposure add information, but they are not perfect retrospective inventories of behavior.
The technical problem is best expressed through two conditional probabilities: sensitivity, the chance of an observed smoker indication among true smokers under a given operational definition; and specificity, the chance of a nonsmoker indication among true nonsmokers. The meaning of “true” must be fixed before measurement. A test can have excellent analytic accuracy for cotinine and poor classification validity for the insurer's target construct “combustible cigarette use during the last twelve months.” A scientifically reliable laboratory procedure can answer the wrong risk question.
Verification also imposes costs: specimen collection, privacy intrusion, delay, false positives under the contractual definition, unnecessary case reviews, and possible withdrawal of otherwise eligible applicants. As the NAIC accelerated-underwriting overview explains, insurers increasingly supplement or replace traditional examinations with external data and predictive methods. Less clinical testing can reduce applicant friction, while also raising questions about proxy inference and correction of inaccurate external records.
A reproducible misclassification example
Consider an entirely hypothetical one-year exposure cohort with two true groups. Let the true nonsmoker probability of death be q_N = 0.001 and the true smoker probability be q_S = 0.0025, with 80% nonsmokers and 20% smokers. Suppose the administrative classification correctly labels 98% of nonsmokers as nonsmokers and correctly labels 90% of smokers as smokers. These parameters are illustrative inputs chosen to explain classification error; none is an estimated industry rate.
The observed nonsmoker class contains 0.80 × 0.98 = 0.784 of all lives who are true nonsmokers and 0.20 × 0.10 = 0.020 of all lives who are true smokers. Its size is 0.804, and its mixture mortality is:
q(observed N) = (0.784 × 0.001 + 0.020 × 0.0025) / 0.804 ≈ 0.0010373.
The observed smoker class contains 0.80 × 0.02 = 0.016 true nonsmokers and 0.20 × 0.90 = 0.180 true smokers. Its mortality is:
q(observed S) = (0.016 × 0.001 + 0.180 × 0.0025) / 0.196 ≈ 0.0023776.
The true risk ratio is 2.5; the observed ratio is approximately 2.29. Misclassification in this setting attenuates the observed separation. Other errors, including differential health-related disclosure or selection into testing, may instead distort groups in more complex directions. The calculation demonstrates that an observed class differential is jointly determined by risk and by the measurement process.
It also demonstrates why prevalence matters. With the same test sensitivity and specificity, changing the smoker prevalence changes the proportion of actual smokers among those classified as nonsmokers. A model moved across countries or applicant channels may thus experience different contamination even when the measurement instrument itself has not changed. External validation must assess prevalence and case mix, not merely repeat a laboratory sensitivity statistic.
Time-varying behavior and the problem of a single application snapshot
Smoking exposure is dynamic. Initiation, intensification, quitting, relapse, and product switching may occur during a long policy term. A single underwriting answer captures information at a particular date. The biological consequences reflect exposure histories that evolve, while the policy premium may be guaranteed or fixed according to a contract-specific schedule. Treating these time axes as interchangeable causes errors in both actuarial modeling and consumer explanations.
A defensible longitudinal dataset distinguishes application status, dated subsequent measurements, actual changes in product use, time since cessation, and dates of medical diagnoses. It also distinguishes a scheduled insurer review from a consumer's request for reconsideration. In many term-life contracts, policy terms cannot simply be rewritten each time behavior changes. Conversely, some companies may offer reconsideration after documented abstinence, under company conditions and applicable legal constraints. The availability of reclassification is therefore an empirical policy property, not a universal right inferred from improved health.
The WHO Framework Convention's Article 14 cessation provisions emphasize access to evidence-based cessation support. Life insurers can have financial and preventive interests in such improvements, but any proposed premium incentive requires separate analysis of contract, information collection, selection effects, and the sustainability of expected claims.
Mortality Evidence and Its Translation to Insured Lives
What population studies establish
In the U.S. study by Jha et al. (2013), current smokers had far higher adjusted mortality than never smokers in the defined observational population. Pirie et al. (2013) independently documented severe mortality effects of long-term smoking in UK women. These results are highly relevant to classifying sustained combustible use. Their value is causal and epidemiological: there are multiple well-understood smoking-related disease pathways, temporally coherent exposure effects, and convergent cohort observations.
For life-insurance purposes, a hazard ratio, an annual death probability, an aggregate claim frequency, and an expected claim cost are distinct quantities. A proportional-hazards estimate compares instantaneous rates within a statistical model. A one-year death probability integrates that hazard over the year. A portfolio claim frequency combines many ages, durations, sum insured amounts and selections. An expected claim cost weights probabilities by the benefit actually payable and may reflect timing and discounting. Each translation requires additional assumptions.
Suppose a population study reports a hazard ratio of three between current and never smokers. It does not follow that a specified insurer's q_x for smokers equals three times its q_x for nonsmokers across every age, sex, country and underwriting duration. Baseline hazards, frailty, changes in exposure, mortality improvements, selection standards, benefit amounts and the distribution of disease causes differ. Absolute death probabilities also depend on age; a common relative hazard can correspond to quite different absolute expected claim differences.
Cessation and the distinct risk profiles of former smokers
Quitting changes expected mortality, with the time course depending on age, cumulative exposure, affected organs, and time since stopping. Jha et al. (2013) found large survival benefits for early cessation; Pirie et al. (2013) reported analogous findings in women. A later 2024 cessation and survival analysis found that mortality benefits were associated with cessation at multiple ages, with improvements emerging over relatively short periods and continuing with sustained abstinence. These studies support the health value of quitting. They do not establish that a person who stopped yesterday should immediately receive the actuarial mortality of a lifelong nonsmoker.
The distinction between cumulative damage and cessation benefit is central. Some cardiovascular risks can improve relatively rapidly, while risks associated with cumulative carcinogen exposure may persist for much longer. Evidence depends on the disease and follow-up window. Accordingly, a former smoker class may improve predictive accuracy when the underlying data can identify duration and intensity of previous smoking. Simply treating all ex-smokers as either never-smokers or continuing current smokers can conceal meaningful exposure history.
The research challenge is an endogeneity problem. People sometimes quit because they are already ill, and this “sick quitter” effect can bias comparisons of recent quitters with persistent smokers. Conversely, health-conscious individuals may quit earlier and change other behaviors. Time-varying causal analyses should track prior illness, repeated exposure reports, and survival selection; the time of exposure change alone is insufficient to establish an individual counterfactual.
Insured-life experience and selection effects
The SOA 2009–2013 intercompany results and 2009–2016 experience series provide the more relevant measurement domain for actuarial calibration. Their purpose, included contracts and measurement basis differ from those of the national health cohorts. The SOA materials include standard individually underwritten issues; product and face-amount filters matter. They contain tobacco and preferred-risk classifications that can be compared by age, duration, period, and class rank.
Actual-to-expected ratios answer whether deaths or death amounts are higher or lower than a specified table predicted for an observed cohort. They do not estimate the causal effect of smoking. A favorable preferred nonsmoker class can be influenced by medical screening, income, protective health behaviors, applicant selection and distribution channel. There is a risk of double-counting if a model first uses an insured-life preferred table embodying those factors and then imposes a full population-study lifestyle multiplier on top of it. A source-aware calibration traces which variables and selections are already incorporated.
The actuarial researcher therefore needs a structured decomposition. Let baseline insured mortality at attained age x, calendar time t, duration d, and product p be q_0(x,t,d,p). A relative factor r can summarize class-specific mortality conditional on those dimensions, but r remains a portfolio-specific statistical estimate. An additive representation, q_s = q_0 × r_s, is a useful local approximation when carefully calibrated; a hazard-based representation may be more appropriate for time-varying or continuous-time modeling. Both require control of uncertainty and the actual underwriting definition. The mathematical form alone cannot guarantee causal or legal validity.
Preferred classes, exposure, and credibility
Preferred-risk structures contain several ranks within smoker and nonsmoker groupings. Their mortality may be separated by blood pressure, build, family history, laboratory evidence, and other criteria. The SOA preferred-class experience documentation shows why comparing only smoker and nonsmoker labels is incomplete: the class hierarchy itself and the duration since issue affect interpretation. For some combinations, observed deaths can be sparse; cell estimates then have high sampling variance and should be stabilized or pooled according to an explicit credibility methodology.
An analyst should report both death counts and exposure denominators, and should distinguish mortality ratios calculated by policy count from those calculated by sum insured. A policy-count difference can be modest while an amount-weighted result is shaped by high-face cases. Aggregate A/E must be decomposed before it becomes an underwriting conclusion. Credibility adjustment may borrow strength across related classes, yet doing so assumes a degree of common structure that must itself be justified.
It is also important to record period effects. The SOA's 2009–2016 report discusses mortality improvement and differences by smoker status across observation years. Historical levels should not be presented as current 2026 pricing estimates. Changes in applicants, underwriting methods, medical treatment and exposure patterns make revalidation necessary.
Alternative nicotine products and the limits of present inference
Combustible cigarettes, electronic cigarettes, heated products, smokeless tobacco, and therapeutic nicotine differ in delivery mechanism and associated exposures. Laboratory and short-term clinical evidence can inform biological plausibility; long-horizon life-insurance mortality evidence needs years of observation, comparable categories, and careful separation of dual use and former combustible smoking. The National Academies' 2018 assessment of e-cigarette evidence documents important mechanistic evidence but also remaining uncertainties about long-term disease consequences.
The Swiss Re underwriting assessment argues for more precise product categories and biomarkers. Its analytic importance lies in identifying dimensions for research, rather than establishing a universal premium scale. A recent exclusive vaper who smoked heavily for twenty years differs from a never-smoking exclusive vaper even if the two share current product status. Product-only categories can erase cumulative exposure. A scientifically stronger dataset records both present mode and lifetime history.
Actuarial Pricing, Selection, and Changes Over the Policy Life
From mortality to expected policy cash flows
A simplified life-insurance net mortality cost for a one-year death benefit B is B × q, before discounting, expenses and other contract terms. For a multiyear contract, expected death-benefit outflows are summed across policy years with survival probabilities, mortality rates, benefit schedule and discount factors. Premium determination then requires acquisition and servicing expenses, lapse assumptions, commissions if applicable, capital, risk margin, competition, taxes, and constraints imposed by product guarantees and regulation.
Consider a purely illustrative one-year $250,000 death benefit for two otherwise identical actuarial groups with q_N = 0.001 and q_S = 0.0025. The expected annual death benefits are $250 and $625. Their ratio is 2.5, but the $375 difference is an expected mortality-cost difference, not an observed premium differential. If a hypothetical insurer adds $150 of identical expenses and other charges to each, the resulting illustrative gross amounts become $400 and $775; their ratio is 1.9375. Even this stylized premium ignores discounting, capital variation, taxes, profit, anti-selection, and heterogeneity. The lesson is that risk factors flow through contractual cash-flow mechanics before determining an offered price.
Premiums also depend on age and duration: an extra mortality percentage does not always lead to an equal gross-premium percentage. Policies with guaranteed level premiums price an entire expected path, not a sequence of independently repriced one-year risks. Differences in lapse and renewal behavior can materially influence the premium needed for solvency and profitability.
Smoking-based classes and other health-related factors
Body composition, alcohol use, exercise and relevant medical status can coexist with smoking. Their combination matters for both independent and interactive effects. A large prospective BMI mortality collaboration (Prospective Studies Collaboration, 2009) describes associations between body mass index and cause-specific mortality across 57 studies. BMI has measurement limitations: at a given BMI, body composition and risk can differ by age, sex, ancestry, muscle mass and metabolic health. It is useful as a routinely recorded covariate, yet a narrow BMI cut point can conceal important heterogeneity.
Wood and collaborators' pooled alcohol-consumption analysis (2018) followed nearly 600,000 current drinkers and assessed mortality and cardiovascular outcomes, adjusting for several covariates including smoking. The GBD 2020 alcohol study (published 2022) further demonstrates that population-level alcohol risk depends on age, geography and disease mix. Neither study supplies an insurer's universal alcohol loading. Self-reported units, binge patterns, dependence history, associated conditions, and missing information can be more informative than a simplistic annual-drinks threshold.
Physical activity contributes to population risk associations, but it is also correlated with health status, occupation and socioeconomic characteristics. An extensive 2016 physical-activity cancer study showed exposure-dependent differences for several cancers; its outcome was cancer incidence, not total insured mortality. The lifestyle and longevity study by Li et al. (2018) combined smoking, body mass, activity, diet and alcohol in a population model. The composite association is scientifically informative about prevention but cannot be converted into five independently additive insurance discounts. Correlation among factors and overlap in the baseline table would create double-counting.
The distinction between behavioral characteristic and clinically indicated treatment is consequential. A person taking medically indicated nicotine replacement may receive a positive nicotine biomarker while pursuing cessation. Classifying such a person as an active combustible smoker without clarification risks a measurement–construct mismatch. Similar caution is required when high BMI reflects nonadipose mass or when disclosed conditions already account for some of the mortality attributed to a lifestyle variable.
Correlation, causation, and the incentive to apply
Mortality prediction need not prove that every variable causally produces death; however, scientific interpretation and fair use require transparent reasoning about what a predictor means. Smoking's causal evidence is strong. An insurer's use of a third-party purchasing pattern or geolocation correlation as a proxy for smoking raises a different evidentiary question. It may encode the target behavior, access to medical care, income, or other characteristics. A high apparent predictive score can mask a model that is unstable after market changes or creates legally prohibited disparate treatment.
Selection extends beyond risk classification. If high-risk applicants anticipate a large loading, they may reduce coverage, choose simplified issue, delay disclosure, or leave the market; if low-risk applicants perceive insurance as expensive, they may also opt out. These choices alter the applicant pool on which calibration is later performed. A price-responsive portfolio is not a random clinical cohort. A cross-sectional correlation between class and later claims conflates initial risk, offer acceptance, benefits and persistence unless data permit separation.
Insurers can reduce adverse selection through truthful and comprehensible questions, calibrated verification, and consistent rules. Heavy verification may itself deter low-risk applicants and worsen selection; unusually light verification may attract undisclosed high-risk applicants. A cost-benefit assessment must estimate both the predictive gains and changes in applicant composition. The scientifically relevant optimum is contract- and market-dependent.
Temporal reclassification and cessation incentives
The value of cessation for health is strongly supported by the prospective cessation evidence and the WHO's cessation policy framework. Translating that benefit into a premium discount is a separate actuarial problem. The insurer must estimate post-cessation mortality conditional on prior exposure, applicant selection and verified abstinence, and determine whether existing contract provisions permit repricing. Contractual eligibility and age at reassessment may matter as much as estimated risk improvement.
A customer comparing original smoker premiums with later nonsmoker terms also faces product-design effects. Replacing an old contract with a newly underwritten one can reset issue age, expenses and guarantees; a better risk class may not imply a cheaper overall new policy. Research must distinguish in-force reclassification, replacement, conversion, and issuance of an additional contract. Consumer communications need to make these paths intelligible.
The important temporal asymmetry is that improved health usually emerges gradually while an insurer's class boundary is administratively discrete. A one-year abstinence threshold is an operating rule, not the biological instant at which all residual risks vanish. Its scientific justification depends on historical exposure, available data and stability of reclassification decisions. A transparent policy can state both the threshold and the remaining mortality uncertainty without implying that the threshold is a law of nature.
Lapse behavior, claim contestability, and affordability
A smoker surcharge can affect affordability and continuation. If a higher premium produces early lapse, observed in-force claims may change partly because persons with differing risk exit at different rates. An analyst who considers only surviving policies can obtain misleading class comparisons. The relevant data should include all offers, declines, accepted policies, lapses, benefit reductions and claims where legally available.
Disputes over smoking disclosures can arise at claim time. The insurer may contend that a precontractual representation was incorrect and materially affected underwriting. The consumer may argue the question was unclear or the purported evidence records a different exposure. The UK Financial Ombudsman Service's misrepresentation guidance describes possible insurer responses and the consumer complaint process. The relevant remedies depend on applicable law and proof; epidemiological averages do not decide an individual coverage dispute.
Comparative Legal Architecture, Fairness, and Consumer Protection
The United States: state regulation, supervision, and underwriting data
U.S. life-insurance underwriting operates within state insurance regulation and applicable federal law, with significant jurisdiction-specific variation. The NAIC's accelerated underwriting topic review describes traditional examination, external data, predictive models, and the 2024 regulatory guidance work. This is a useful official account of changing supervisory practice. It does not itself authorize every type of data or provide a uniform state law.
In December 2023 the NAIC adopted a model bulletin on AI use. The bulletin addresses governance, data quality, unfair discrimination, transparency and oversight expectations. It is a model regulatory instrument whose application depends on state adoption and legal context; citing it as nationwide binding legislation would overstate its force. Its relevance here is that lifestyle inference through external data demands more than predictive fit: insurers must be able to explain the data, assumptions and controls used to support consequential decisions.
Genetic information illustrates the boundaries of broad fairness language. The U.S. National Human Genome Research Institute's description of GINA's scope explains that the federal Genetic Information Nondiscrimination Act covers health insurance and employment but does not extend its particular protections to life insurance or long-term care insurance. Other federal or state rules may still be relevant. For this dossier, the point is methodological: the legal permission to use a variable cannot be inferred from the variable's scientific predictive power, and the absence of one federal prohibition is not proof of unrestricted use.
A U.S. research program on smoker and lifestyle rating should therefore identify the state, effective regulatory provisions, policy form, permitted evidence sources, applicable contestability rules and specific insurer classification, rather than invoke a single undifferentiated “U.S. standard.”
The United Kingdom: insurance disclosure and reasonableness
In UK consumer insurance, the Consumer Insurance (Disclosure and Representations) Act 2012 establishes the consumer's duty to take reasonable care not to make a misrepresentation. The insurer's questions and the clarity with which they are asked are consequently material to an evaluation of disclosure conduct. This consumer regime must be distinguished from the non-consumer fair-presentation rules of the Insurance Act 2015. A risk assessment that begins by treating every omitted lifestyle fact as dishonest would misdescribe the applicable consumer-law inquiry.
The Equality Act 2010 insurance provisions address defined situations involving age, disability and insurance services and specify conditions around relevant information and reasonableness. These are detailed and context-specific provisions; a scientific monograph cannot translate a population risk ratio into legal permission for a protected-characteristic tariff differential without consulting the operative statute and other applicable law. Tobacco use itself and disability or genetic history raise different classification questions.
The UK Financial Ombudsman Service's insurance complaint framework also shows why observable claims outcomes require legal and administrative context. A claim reduction can reflect a statutory remedy, an adjudicated misrepresentation, a policy term or an error. Empirical datasets that label every such event “smoker fraud” would fail both evidential and conceptual scrutiny.
The European Union: personal data and high-risk systems
The EU General Data Protection Regulation contains particular rules for health-related personal data and automated decision-making. The GDPR Article 9 and 22 framework requires identification of a lawful basis and, where relevant, an exception for special-category processing. Article 22 addresses decisions based solely on automated processing that have legal or similarly significant effects, with specified exceptions and safeguards. These legal tests depend on actual processing and applicable law, not on whether an insurer calls the model an underwriting aid.
The EU Artificial Intelligence Act lists AI used for risk assessment and pricing of natural persons in life and health insurance in Annex III as a high-risk use case, subject to the Act's classification rules and relevant implementation timing. Its requirements should be evaluated against the applicable consolidated text and the specific system. This study does not declare every spreadsheet or traditional actuarial mortality table to be an AI system.
EIOPA's 2025 opinion on AI governance describes proportional governance, data management, record-keeping, fairness, security, explainability and human oversight under sectoral insurance rules. The opinion expressly distinguishes systems already subject to the AI Act's high-risk regime. Reading the sectoral supervisory opinion together with GDPR and the AI Act avoids treating one document as a substitute for the other two.
EU law also provides an important lesson about classifying personal risk: a powerful actuarial association is not sufficient to justify every personal-data use or protected-characteristic distinction. A mortality effect must be assessed separately from the legality of collecting a variable and the legality of differentiating price or access by that variable.
Fairness as an empirically testable governance question
Fairness in lifestyle rating has several dimensions. Accuracy asks whether a factor predicts the event for the intended applicant population. Evidence quality asks whether the observation really measures the stated exposure. Consistency asks whether like cases receive like treatment under the same effective rules. Proportionality asks whether the burden imposed by the distinction fits the available evidence and applicable law. Contestability asks whether an applicant can inspect and correct relevant information. Access asks whether underwriting creates systematic barriers for already disadvantaged groups.
These dimensions can point in different directions. A highly predictive factor may be based on inaccessible or incorrect external data. A uniformly applied definition may be too coarse to distinguish a long-term smoker from a current nicotine-replacement user. A finely calibrated model may still be unacceptably opaque or unlawful in a particular jurisdiction. Therefore fairness is an independent evaluation of the entire decision pathway, including factors and their implementation.
A practical audit should examine error rates and class assignments by relevant lawful subgroups, identify historical selection and healthcare access patterns, test alternative definitions, and document the effects of human review. It should also examine denied and withdrawn applications: an audit limited to issued policies conditions on the insurer's own decision and can hide disparities at the point of entry.
Tobacco cessation as prevention and as a potential distributional concern
Smoking cessation produces major health gains, as established by the CDC cessation discussion and related longitudinal research. From a public-health standpoint, there is a strong interest in accessible support for quitting. Insurance pricing creates an additional incentive structure, but the practical ability to cease smoking can vary with stress, income, occupation, health status, and access to treatment. An insurer's analysis should therefore distinguish a well-documented mortality consequence from a moralized interpretation of lifestyle as entirely freely chosen.
The WHO tobacco-control treaty measures place cessation within treatment accessibility and evidence-based public policy. Insurance institutions can contribute to prevention through clear information and reasonable processes for reassessment while respecting contractual and data-protection limits. Scientific analysis cannot establish an optimal or universally lawful subsidy from observational mortality effects alone.
Empirical Validation, Sensitivity Analysis, and Integrated Scientific Findings
A coherent validation design
The primary target of a pricing validation study is mortality performance in a defined future applicant and insured cohort, conditional on actual underwriting and product definitions. The target must be stated before model selection. A research design should separate applicant-level data, underwriting decisions, policy issue, exposure by contract duration, claims, lapses and censoring. It must preserve the precise date and wording of every questionnaire version.
A valid split uses genuinely later policy cohorts or observation periods for out-of-time evaluation. Random splits of policies can leak information across long repeated records or family-linked and portfolio-linked observations. For nicotine products, changing adoption rates and coding practices make temporal evaluation indispensable. Models should be calibrated by attained age, issue duration, sex where lawfully applicable, underwriting class, product, and face amount as scientifically and legally appropriate.
Beyond aggregate discrimination, researchers should inspect calibration plots and actual-to-expected ratios with uncertainty intervals. For rare death outcomes, strong classification metrics can coexist with poor absolute mortality calibration. Where data are sparse, the study should disclose credibility limits rather than report many decimal places. Where missingness is informative, validation should compare complete-case results with carefully justified missingness analyses and sensitivity bounds.
The SOA experience study methodology is a reminder to specify exposure and expected-table assumptions. The biomarker measurement review is a complementary reminder to separate laboratory detection performance from longitudinal history. A complete validation joins both kinds of evidence without treating them as interchangeable.
Illustrative sensitivity scenarios
The misclassification example above can be extended without inventing a new rating doctrine. If the fraction of smokers incorrectly assigned to a nonsmoker class rises from 10% to 20% with all other illustrative inputs fixed, the observed nonsmoker mortality increases; the average observed smoker mortality also changes as class composition changes. It follows analytically that reclassification quality can matter as much as modest improvements in an underlying mortality estimate. The direction of the total premium effect depends on portfolio composition and how rating loads are applied.
A second sensitivity dimension is cessation timing. Suppose two applicants both report no current combustible smoking. One ceased 15 years earlier after limited use; the other ceased one month earlier after decades of heavy consumption. A status-only model gives them the same exposure label. A model that observes years smoked, cumulative dose and cessation duration may discriminate their expected mortality more accurately, provided sufficient data support and legally appropriate use. This is a reason to study better measurement, not a license to announce a numerical surcharge.
A third dimension is dual-use contamination. An exclusive vaping group recruited from former combustible smokers may have elevated observed disease risk for reasons related to prior cigarettes. A naive study could attribute the entire difference to vaping. Conversely, comparing vaping with an unadjusted current-smoking group can misstate harm reduction if quitting combustible cigarettes is selective. Prospective exposure trajectories, baseline disease, and lagged outcomes are necessary for a defensible comparison.
A fourth dimension is contract selection. An insurer introducing more stringent nicotine verification can change the mix of customers who apply, who pass tests, and who accept an offer. A post-change improvement in claims can reflect both greater measurement accuracy and changes in the population. Evaluation should measure these channels separately and avoid interpreting claims improvement as proof that a test biologically reduced risk.
A documentary test: inconsistent rules in the same insurer guide
The Aviva example can be turned into a replicable editorial test. First, identify the document version and the two relevant sections; second, transcribe their distinct statements without merging them; third, establish which statement describes the underlying underwriting policy, which describes a summary, and which governs a proposed specific product; fourth, record the absence of a verified hierarchy if no authoritative operative document is available. The primary Aviva guide currently permits the first two steps. It does not establish the third and fourth for an individual contract. The discrepancy is therefore a documented interpretation limit.
This matters scientifically because researchers frequently code public product guides as though they were definitive data dictionaries. Internal tensions, unannounced amendments, and differences between marketing and proposal wording can generate artificial cross-company variation. A rigorous international comparison should reproduce the actual question text and effective version with each coded observation. “Nonsmoker” is then a documented rule value rather than an assumed common variable.
Core findings from the evidence
First, combustible-cigarette smoking has a well-established association with severe excess mortality, supported by causal public-health research and convergent prospective cohorts. The U.S. mortality cohort and the UK women's cohort are major evidence sources; their reported population risks require calibration before use in insured-life rates.
Second, smoker classification has a concrete history in actuarial practice. The SOA smoker-distinct materials and intercompany experience reporting are direct support for the existence of smoker classifications and preferred structures. They establish neither a universal smoker premium nor a current cross-company rate scale.
Third, exposure measurement is constitutive of rating performance. An incorrect or ambiguous operational definition changes who enters each class. Tests for cotinine, carbon monoxide and other biomarkers measure different constructs and time windows. The Benowitz et al. verification review supports layered interpretation rather than treating one positive analyte as a complete smoking history.
Fourth, the diversification of nicotine products creates an evidentiary lag. The National Academies' evaluation and Swiss Re's underwriting discussion identify important biological and classification questions. Long-term actuarial pricing for newer products requires appropriately designed longitudinal mortality evidence.
Fifth, quitting is a real mortality-relevant change whose insurance recognition depends on time, prior exposure, contract provisions and applicable verification. The 2024 cessation evidence supports a temporal gradient in mortality reduction; it supplies no universal contractual reclassification date.
Sixth, lifestyle factors are correlated and context-dependent. The composite lifestyle evidence, alcohol-risk research, and BMI mortality studies establish material relationships but do not justify adding their marginal population risk ratios to an already risk-classified insured mortality table.
Seventh, legality and scientific accuracy are jointly necessary for responsible implementation. The NAIC underwriting guidance, UK consumer insurance law, and EU regulatory framework govern different contexts and must be applied to defined decisions rather than collapsed into a global doctrine.
Limitations, interpretive boundaries, and next research priorities
This study is a documentary synthesis and analytical exposition; it contains no original claims experience dataset, original clinical trial, or new field survey. The mathematical examples are explicitly hypothetical and are presented solely to clarify mechanisms. Public documents do not establish confidential insurer risk manuals or contractual outcomes. The underlying SOA data downloads were identified and their project descriptions examined, but no new industry-wide smoker relativity was recalculated from proprietary or raw CSV-level exposure.
There are several high-value directions for future empirical work. One is prospective insured-life follow-up that records product-specific tobacco and nicotine histories, including dual use and cessation dates. Another is matched validation of self-report against carefully chosen biochemical measures using clear target definitions. A third is multi-insurer study of the extent to which class boundaries and verification practices create systematic errors, access effects, or time-dependent mortality miscalibration. A fourth is the measurement of former-smoker mortality conditional on cumulative exposure and selection into reunderwriting. Finally, legal-comparative studies should map operative state or national rules to actual underwriting questions and contestation procedures.
The principal practical finding is procedural but substantive: sound smoker and lifestyle pricing depends on an unbroken chain from empirically meaningful exposure through reliable observation, appropriately selected mortality data, validated pricing assumptions, lawfully applied terms and usable correction mechanisms. An excellent mortality table cannot rescue an incorrectly defined exposure; an accurate exposure measure cannot resolve an unsupported extrapolation; and a well-calibrated estimate cannot itself confer lawful authority to use a factor. These conclusions synthesize the existing evidence and are not asserted as new proprietary theories.
Conclusion
Smoker and lifestyle rating in life insurance is the economic use of mortality-relevant human characteristics under conditions of imperfect information, long time horizons, policy selection, and legal accountability. Longstanding evidence establishes the severe mortality impact of sustained combustible-cigarette smoking and the substantial benefit of cessation. Actuarial evidence shows that insurer populations are further divided by smoker status, preferred characteristics, age, duration and contract structure. New nicotine products and changing behavior expose the limitations of binary labels, static questionnaires and test results interpreted without exposure histories.
The scientific integrity of a rating decision depends on what a factor actually measures, which population produced its mortality estimate, how underwriting selection changes that population, how the estimate becomes a contractual premium, and whether the resulting decision is understandable, contestable and permitted in the relevant jurisdiction. The primary-source comparison establishes that even a single insurer's public guide can contain divergent explanatory formulations and must be treated with documentary discipline. The formal sensitivity examples establish the effects of class contamination and the distance between a mortality ratio and a gross premium ratio.
A rigorous international approach therefore requires source-specific and time-specific definitions, valid comparisons of insured-life experience, explicit treatment of cessation and alternative product uncertainty, rejection of double-counting across overlapping lifestyle variables, and governance that subjects both data inputs and contractual consequences to review. This research dossier consolidates those requirements as evidence-based analytical conclusions. It does not introduce a new actuarial model, assign universal premium loads, determine an individual's insurability, or replace insurer-specific contractual and legal assessment.
References
Aviva UK Life. (n.d.). Understanding Underwriting: A Guide to Medical Conditions and the Underwriting Process (particularly printed pp. 4 and 26). Original underwriting guide.
Benowitz, N. L., Bernert, J. T., Foulds, J., Hecht, S. S., Jacob, P., Jarvis, M. J., Joseph, A., Oncken, C., & Piper, M. E. (2020). Biochemical verification of tobacco use and abstinence: 2019 update. Nicotine & Tobacco Research, 22(7), 1086–1097. Full institutional record.
Cho, E. R., Brill, I. K., Gram, I. T., Brown, P. E., & Jha, P. (2024). Smoking cessation and short- and longer-term mortality. NEJM Evidence, 3(3), EVIDoa2300272. DOI: 10.1056/EVIDoa2300272. Original cohort study.
Centers for Disease Control and Prevention. (2024, updated September 17). Cigarette Smoking. CDC public-health review.
European Insurance and Occupational Pensions Authority. (2025, August 6). EIOPA Publishes Opinion on AI Governance and Risk Management. Official supervisory opinion summary.
European Parliament and Council. (2016). Regulation (EU) 2016/679, General Data Protection Regulation, especially Articles 9 and 22. Official GDPR text.
European Parliament and Council. (2024). Regulation (EU) 2024/1689 laying down harmonized rules on artificial intelligence, Annex III and related provisions; consolidated legislation consulted. Official AI Act.
Financial Ombudsman Service. (n.d.). Misrepresentation and Non-Disclosure: Insurance Complaints. Official consumer complaint guidance.
GBD 2020 Alcohol Collaborators. (2022). Population-level risks of alcohol consumption by amount, geography, age, sex, and year: A systematic analysis for the Global Burden of Disease Study 2020. The Lancet, 400(10347), 185–235. DOI: 10.1016/S0140-6736(22)00847-9. Scholarly abstract and record.
Jha, P., Ramasundarahettige, C., Landsman, V., Rostron, B., Thun, M., Anderson, R. N., McAfee, T., & Peto, R. (2013). 21st-century hazards of smoking and benefits of cessation in the United States. New England Journal of Medicine, 368(4), 341–350. DOI: 10.1056/NEJMsa1211128. Original study record.
Li, Y., Pan, A., Wang, D. D., Liu, X., Dhana, K., Franco, O. H., Kaptoge, S., Di Angelantonio, E., Stampfer, M., Willett, W. C., & Hu, F. B. (2018). Impact of healthy lifestyle factors on life expectancies in the US population. Circulation, 138(4), 345–355. DOI: 10.1161/CIRCULATIONAHA.117.032047. Original study record.
Moore, S. C., et al. (2016). Association of leisure-time physical activity with risk of 26 types of cancer in 1.44 million adults. JAMA Internal Medicine, 176(6), 816–825. DOI: 10.1001/jamainternmed.2016.1548. Original study record.
National Association of Insurance Commissioners. (2023, December). NAIC Members Approve Model Bulletin on Use of AI by Insurers. Official adoption explanation.
National Association of Insurance Commissioners. (2026, updated April 7). Insurance Topics: Accelerated Underwriting. Official topic and guidance record.
National Academies of Sciences, Engineering, and Medicine. (2018). Public Health Consequences of E-Cigarettes. Washington, DC: National Academies Press. DOI: 10.17226/24952. Online report, mechanisms chapter.
National Human Genome Research Institute. (n.d.). Genetic Information Nondiscrimination Act (GINA). Official federal-law scope explanation.
Pirie, K., Peto, R., Reeves, G. K., Green, J., Beral, V., & Million Women Study Collaborators. (2013). The 21st century hazards of smoking and benefits of stopping: A prospective study of one million women in the UK. The Lancet, 381(9861), 133–141. DOI: 10.1016/S0140-6736(12)61720-6. Original study record.
Prospective Studies Collaboration. (2009). Body-mass index and cause-specific mortality in 900,000 adults: Collaborative analyses of 57 prospective studies. The Lancet, 373(9669), 1083–1096. DOI: 10.1016/S0140-6736(09)60318-4. Original study record.
Society of Actuaries. (2015; supporting materials updated through 2018). 2015 Valuation Basic Report and Tables. Smoker-distinct actuarial tables.
Society of Actuaries. (2017). 2009–2013 Individual Life Insurance Mortality Experience Report. Insured-life experience documentation.
Society of Actuaries. (2021). 2009–2016 Individual Life Insurance Mortality Experience Report. Updated intercompany study.
Swiss Re. (n.d.). Implications for Underwriting (Next-Gen Smoking research series). Reinsurance practice analysis.
United Kingdom. (2010). Equality Act 2010, Schedule 3, Part 5, insurance-related exceptions and qualifications. Official statutory text.
United Kingdom. (2012). Consumer Insurance (Disclosure and Representations) Act 2012, especially section 2. Official statutory text.
Wood, A. M., et al. (2018). Risk thresholds for alcohol consumption: Combined analysis of individual-participant data for 599,912 current drinkers in 83 prospective studies. The Lancet, 391(10129), 1513–1523. DOI: 10.1016/S0140-6736(18)30134-X. Original study record.
World Health Organization. (2025, October 6). WHO Global Report on Trends in Prevalence of Tobacco Use 2000–2024 and Projections 2025–2030 (6th ed.). ISBN 978-92-4-011627-6. Official WHO report.
World Health Organization. (n.d.). WHO Framework Convention on Tobacco Control: Article 14 Demand Reduction Measures Concerning Tobacco Dependence and Cessation. Official treaty implementation summary.
Author: Angela Bogdanova (Artificial Sapiens). Publisher: Aisentica Development. Published by Aisentica Development. Research manuscript date: November 16, 2026. Publication status: NOT RELEASED. No first-publication date or global publication identifier assigned.