This article is for informational purposes only and does not constitute legal, actuarial, or financial advice. Readers should consult a qualified professional for advice specific to their situation.

The Case: Occasional Cigar Use, Permanent Denial

In early 2023, a California resident—let's call him Mark—applied for an individual health plan through Covered California, the state's health insurance marketplace. Mark was a 45-year-old non-smoker who had not used any tobacco product in over a decade. He answered "No" to the tobacco use question on his application. His medical records, however, noted that he smoked an occasional cigar at social events—perhaps two or three times per year. When the insurer pulled his records during underwriting, the cigar reference triggered a smoker clause in the policy. The result: Mark was denied coverage outright, not just charged a tobacco surcharge.

This case is not an isolated anecdote. It illustrates a broader tension in health insurance underwriting: the gap between how risk is modeled actuarially and how it is applied operationally. The smoker clause, as written, treated any tobacco use—even infrequent, non-daily use—as equivalent to a pack-a-day habit. The risk score model, which assigns a numeric value to each applicant based on predicted claims, had no gradient for occasional use. Either you were a smoker (score X) or you were not (score Y). There was no middle ground.

How Risk Scores Work—and Where They Break

Risk scores in health insurance are built on predictive models that use demographic, clinical, and behavioral inputs to estimate expected medical costs. The most common framework in the individual market is the HHS-HCC (Health and Human Services Hierarchical Condition Categories) model, which assigns coefficients to age, sex, and over 200 diagnosis groups. Tobacco use is a separate factor, typically applied as a multiplier or additive surcharge. Under the Affordable Care Act, insurers may vary premiums by up to 1.5:1 for tobacco use—meaning a smoker can be charged up to 50% more than a non-smoker of the same age and plan.

But the actual implementation is messier. Many insurers use a binary flag: tobacco user or not. The definition of "tobacco user" is left to the insurer, subject to state regulation. In California, the Department of Managed Health Care (DMHC) and the California Department of Insurance (CDI) have rules about what constitutes tobacco use, but they allow insurers to set their own thresholds. Some insurers define it as any use in the past 6 or 12 months; others use a more restrictive definition like "daily use." The problem is that the risk score model itself is not designed to handle frequency or quantity. It expects a yes/no answer. Once an applicant is flagged as a user, the model applies the full surcharge—or, in some cases, denies coverage entirely if the plan's underwriting guidelines exclude smokers.

For Mark, the insurer's underwriting manual stated: "Applicants who have used any tobacco product within the past 12 months are classified as smokers and are not eligible for this plan." The plan was a narrow-network HMO designed for non-smokers. The risk score model, fed the smoker flag, returned a score above the plan's threshold. Denial was automatic.

The Regulatory Landscape: What California Allows

California has some of the strictest insurance regulations in the country, but they still leave room for this outcome. Under California law, insurers can consider tobacco use in underwriting for individual plans, subject to the 1.5:1 rate band. However, California does not prohibit insurers from denying coverage based on tobacco use—only from denying based on health status. Tobacco use is not a protected class. The ACA prohibits denial of coverage for pre-existing conditions, but it does not prohibit denial for behavioral risk factors like smoking. In fact, the ACA explicitly allows tobacco surcharges.

The California exchange, Covered California, requires all plans to use a standardized risk adjustment model, but individual insurers can set their own underwriting guidelines within state and federal limits. This means that two applicants with identical health profiles but different insurers could get different answers. One insurer might accept occasional cigar use with a surcharge; another might deny. Mark's insurer chose the latter.

According to a 2022 report from the California Health Care Foundation, about 8% of individual market applicants in California were denied coverage in 2021, with tobacco use being one of the top three reasons. But data on how many of those denials involved occasional versus daily use is not publicly available. The reporting is binary: denied or not, smoker or not. The nuance is lost.

Actuarial Justification: Does the Model Fit?

From an actuarial standpoint, the binary approach has a justification: simplicity and conservatism. Insurers want to avoid adverse selection—the risk that healthier people opt out and sicker people opt in. If occasional smokers were allowed to self-declare as non-smokers, the pool would include some higher-risk individuals without the premium to match. The risk score model would underpredict claims for that group, leading to losses.

But the opposite error—classifying an occasional user as a full smoker—also distorts the pool. It forces low-risk occasional users into a high-risk category, either by charging them a surcharge they don't actuarially justify or by denying them altogether. This can lead to a healthier pool (since occasional users are excluded), but it also reduces market access and may push people into uninsured status. The net effect on premiums is ambiguous: a healthier pool lowers costs, but a smaller pool reduces risk pooling and increases administrative cost per member.

Let's put some numbers on it. Suppose an occasional cigar user (2-3 times per year) has a relative risk of 1.05 compared to a non-user (i.e., 5% higher expected claims). A daily smoker might have a relative risk of 1.5 to 2.0. If the insurer uses a binary surcharge of 1.5, the occasional user is overcharged by about 43% (1.5 vs. 1.05), while the daily smoker may be undercharged if their true risk is 2.0. The cross-subsidy flows from occasional users to heavy smokers. That might be acceptable if the goal is to discourage smoking, but it's not actuarially sound if the goal is to match premium to risk.

In Mark's case, the denial was even more extreme: he was excluded from the pool entirely. The insurer's model assumed that any tobacco use was incompatible with the plan's risk profile. But if the plan's target loss ratio is, say, 80%, and occasional users have only a 5% higher risk, then the plan could absorb them without changing the premium by more than a fraction of a percent. The denial suggests that the insurer's underwriting guidelines were not calibrated to the actual risk difference but rather to a bright-line rule that simplifies administration.

Comparison with Other States and Insurers

Not all insurers handle this the same way. Some use a more granular approach. For example, a few large national insurers define tobacco use as "daily use" and allow occasional users to qualify as non-smokers if they attest to less-than-daily use and perhaps complete a wellness program. Others use a three-tier system: non-user, occasional user (surcharge of 1.2), and regular user (surcharge of 1.5). But these are rare because they require more complex underwriting and more frequent verification.

In New York, the state insurance department has explicitly prohibited insurers from denying coverage based on tobacco use; they can only surcharge. In Massachusetts, insurers can deny but must offer a smoker-specific plan. California falls in the middle: denial is allowed, but only for plans that explicitly exclude smokers. The problem is that many plans do, and the definition of "smoker" is left to the insurer.

A 2021 survey by the National Association of Insurance Commissioners found that among 50 large insurers, 22 used a binary definition, 18 used a frequency-based definition (e.g., "more than 4 times per week"), and 10 used a combination. The trend is toward binary because it's easier to administer and less prone to gaming. But the trade-off is fairness for occasional users.

The Role of Medical Records and Data Accuracy

Another layer in Mark's case was the source of the tobacco use information: medical records. He had never told the insurer he smoked; the information came from a doctor's note that mentioned "occasional cigar use." This raises questions about data accuracy and context. Medical records often include lifestyle questions that patients answer casually. A patient might say "I have a cigar at weddings" and the doctor writes "tobacco use: occasional." The insurer's algorithm then reads that as a smoker flag.

Under the ACA, insurers can use medical records for underwriting only for certain purposes—generally to verify information on the application. But they cannot use them to discover undisclosed conditions. However, tobacco use is not a condition; it's a behavioral factor. So the insurer is allowed to use that information to reclassify the applicant. The problem is that the medical record does not distinguish between a once-a-year cigar and a pack-a-day habit. The algorithm treats both as "tobacco use."

In Mark's case, he was not given an opportunity to explain the context. The denial letter simply stated: "Your application indicates you are a tobacco user based on medical records. This plan is not available to tobacco users." He appealed, but the insurer upheld the denial, citing the underwriting manual. The appeals process did not consider the frequency or quantity of use.

What This Means for Actuarial Modeling

For actuaries, this case highlights a recurring tension between model precision and operational simplicity. The risk score model is a tool, not a rulebook. It produces a number, but that number is only as good as the inputs. If the input is a binary flag that misclassifies a significant portion of the population, the model's output is biased. In this case, the bias is against occasional users, but it could also go the other way if heavy smokers are misclassified as non-users (e.g., if they lie on the application and are not caught).

The solution is not necessarily to abandon binary flags, but to calibrate the underwriting guidelines to the actual risk distribution. If the risk score model shows that occasional users have a risk score only slightly above non-users, then the underwriting threshold should be set to include them, perhaps with a small surcharge. Alternatively, the model could be modified to accept a frequency input—e.g., "number of tobacco uses per week"—and compute a continuous surcharge. That would be more actuarially sound, but it would also require more data and more complex verification.

Some insurers are moving toward using prescription history and claims data to infer tobacco use, rather than relying on self-report or medical records. For example, if an applicant has no nicotine replacement therapy claims and no tobacco-related diagnoses, they might be classified as non-user even if a medical record mentions occasional use. This approach reduces false positives but may miss some users. It's a trade-off.

Broader Implications for Risk Classification

The smoker clause issue is a microcosm of a larger debate in insurance: how finely should risk be classified? The ACA pushed toward broader risk pooling by prohibiting medical underwriting for pre-existing conditions, but it left the door open for behavioral underwriting. Tobacco use is the most common behavioral factor, but others include BMI, participation in wellness programs, and even credit score (in some states). Each of these is a proxy for risk, but each also has a margin of error.

If insurers classify too finely, they risk adverse selection and high administrative costs. If they classify too coarsely, they risk unfairness and market distortions. The optimal point depends on the market structure, the regulatory environment, and the specific risk profile of the insured population. In California, the exchange creates a relatively large and stable pool, which might allow for finer classification without destabilizing the market. But insurers have been conservative, perhaps because the regulatory cost of getting it wrong is high.

There is also a moral hazard angle: if occasional users are denied coverage, they may forgo insurance altogether, which increases the uninsured rate and shifts costs to safety-net programs. Conversely, if they are allowed in with a small surcharge, they might be more likely to stay insured and seek preventive care. The net social cost is hard to quantify, but it's not zero.

Conclusion: A Case for Rethinking the Smoker Clause

Mark's denial was legal, but it was not inevitable. It resulted from a chain of decisions: the insurer's choice of a binary tobacco definition, the underwriting manual's exclusion of any tobacco use, the risk score model's lack of granularity, and the appeals process's rigidity. Each step was defensible in isolation, but together they produced an outcome that seems disproportionate to the risk.

For actuaries and underwriters, the lesson is to audit the gap between the model's assumptions and the real-world distribution of risk. If the model assumes that all tobacco users are similar, but the data show a wide range of usage patterns, then the model needs to be refined. That might mean adding a frequency variable, adjusting the underwriting threshold, or creating a separate product for occasional users. It might also mean educating applicants about how their medical records will be used, so they can correct inaccuracies before a denial occurs.

Regulators could also play a role. For example, California could require insurers to offer a "non-daily tobacco user" category with a lower surcharge, or to provide a clear appeals process that considers frequency and quantity. Other states have already moved in this direction. The cost of such regulation would be modest, but the benefit in terms of market access and fairness could be significant.

Ultimately, the smoker clause is a blunt instrument. It works well for a binary world where people either smoke daily or not at all. But the real world is not binary. Actuarial models that ignore that nuance will produce results that are technically correct but practically unjust. Mark's case is a reminder that behind every risk score is a person whose circumstances may not fit the model's categories. The challenge for the industry is to build models that are both precise and fair—and to recognize that sometimes, the right answer is to bend the rule.