Smoking Increases Health Economic Burden; A Multivariate Analysis of Payor Expenses

1. Introduction

Personal healthcare expenditure in the United States is characterized by continuous growth and wide individual-level disparity. Broad economic evaluations demonstrate that healthcare expenditures are shaped by a complex interplay of non-modifiable demographic traits (e.g., age and biological sex), modifiable lifestyle parameters (e.g., body mass index (BMI) and tobacco use), and structural or geographic variation. Understanding the distinct marginal contributions of these factors is essential for health insurance payors, actuarial modeling, public health policymakers, and clinical translation strategies targeting cost mitigation.

While bivariate analyses including direct group comparisons or standard scatterplots, provide initial descriptive utility, they are inherently limited by confounding. For instance, tobacco smokers (1,2), may exhibit distinct age distributions or elevated mean BMIs (3,4) relative to non-smokers. Consequently, observed differences in crude, unadjusted expenditure comparisons fail to isolate the independent effect of individual health determinants. Multivariable statistical frameworks are required to control for mutual confounding and isolate the unique coefficient attributable to each risk factor.

In this study, we evaluate an individual-level health insurance cohort (n = 1338) to model the determinants of personal medical reimbursement claims. Using both primary descriptive exploratory plots and an adjusted Ordinary Least Squares (OLS) multivariate linear regression model, we quantify the independent financial burdens imposed by smoking, elevated BMI, aging, and dependency count. Furthermore, we utilize a geometric dual-plane regression surface model to map how behavioral risk profiles elevate baseline expenditure planes across continuous demographic parameters.

2. Methods

Cohort and Variable Definitions

The dataset evaluated in this study comprises 1338 complete individual records extracted from a public health insurance administrative billing repository (5). The primary continuous dependent variable, Yi (charges), represents total annual individual medical claims reimbursement in US dollars ($). The dataset incorporates 6 explanatory variables (Xk):

• Age (X1): Continuous variable measured in years.

• Body Mass Index (BMI) (X2): Continuous variable calculated as weight in kilograms divided by height in meters squared (kg/m²).

• Children (X3): Discrete numerical variable indicating the total number of dependent children covered under the health policy.

• Smoking Status (X4): Categorical binary variable indicating current tobacco smoking status (yes vs. no).

• Sex (X5): Categorical binary variable indicating biological sex (male vs. female).

• Geographic Region (X6): Categorical factor indicating primary US residential region (northeast, northwest, southeast, southwest).

Statistical Modeling

Bivariate descriptive analyses were conducted to examine unadjusted relationships. Scatterplots with continuous smoothing were generated for numerical predictors (age, BMI, and children) against annual charges. Categorical variables (smoking status, sex, and region) were visualized using box-and-whisker plots depicting medians, interquartile ranges (IQR), and empirical outliers.

To account for multivariable confounding, an Ordinary Least Squares (OLS) multiple linear regression model was specified:

Yi = β0 + β1(Age_i) + β2(BMI_i) + β3(Children_i) + β4(Smoker_i) + β5(Sex_i) + Σ γm(Region_m,i) + εi

where εi ~ N(0, σ²) represents i.i.d. random error. Categorical predictor variables were dummy-encoded using explicit baseline reference categories:

• Smoking Status: Reference group = nonsmoker (no).

• Biological Sex: Reference group = female.

• Geographic Region: Reference group = northeast.

Model fit was assessed using the coefficient of determination (R²) and adjusted R². Parameter estimates (β_k), standard errors (SE), 95% confidence intervals (95% CI), and two-sided p-values were calculated. Statistical significance was defined at α = 0.05. Dual 3D regression planes were rendered across Age and BMI axes to visualize behavioral shift vectors.

3. Discussion

Multivariable modeling of healthcare claims datasets provides crucial insights for health economists, actuarial underwriters, and public health strategists. In this analysis, an OLS regression framework accounted for three-quarters of total claim variance (R² = 0.751), demonstrating that individual medical expenses are strongly structured by a small panel of demographic and behavioral risk factors.

The primary finding of this investigation is the major financial burden imposed by tobacco smoking. Holding age, BMI, family size, sex, and geographic region constant, active tobacco use adds nearly $24,000 in annual claims expenditure. This finding aligns with clinical literature documenting the broad multi-organ pathology, chronic systemic inflammation, and elevated oncologic and cardiovascular risks associated with chronic smoking.

Continuous non-behavioral determinants, age, BMI, and dependent count, demonstrated statistically significant, linear contributions to overall expenses. The observed age expansion coefficient (≈$257/year) aligns with expected lifetime disease accumulation and chronic illness management burdens in older populations. Similarly, the positive coefficient for BMI (≈$339 per kg/m²) reflects elevated clinical utilization associated with metabolic dysfunction, osteoarthritis, and cardiovascular comorbidities.

Importantly, unadjusted comparisons can mislead policy formulation. While crude visualizations suggested potential variance in charges across biological sexes, multivariable adjustment demonstrated that sex exerts no independent effect on charges (β = −$131, p = 0.693) when accounting for age, smoking status, and BMI. Geographic differences were similarly minor relative to lifestyle choices.

Several limitations should be noted when interpreting these results. First, health insurance charges are intrinsically right-skewed and non-negative, which can violate the homoscedasticity assumption of basic OLS linear regression. Future analyses could employ log-transformed dependent variables or Generalized Linear Models (GLM) with Gamma families and log link functions to better account for heavy-tailed cost distributions. Second, while the additive OLS model provides clear main-effect estimates, evidence suggests that behavioral risk factors interact non-linearly. Specifically, previous studies indicate a strong positive interaction between high BMI and smoking status (BMI × Smoker), where obesity accelerates healthcare costs at a steeper rate among active smokers than among non-smokers. Incorporating explicit multiplicative interaction terms into larger administrative claims cohorts represents a natural extension of this work. Finally, these cross-sectional statistical associations do not directly establish causal mechanisms or guarantee immediate claims reduction following behavioral modification. Unobserved confounding including pre-existing genetic liabilities, environmental exposures, or occupation, remains a factor in observational administrative data.

4. Conclusion

This analysis quantifies the determinants of personal health insurance charges using a multivariable regression framework. Tobacco smoking represents the dominant modifiable driver of annual medical expenditure, inducing an adjusted surcharge of approximately $23,849 per individual annually. Continuous risk factors, including age and BMI, contribute steady incremental increases, whereas biological sex and geographic region play minor roles after adjustment. These findings highlight the importance of targeted smoking cessation and metabolic health programs in managing healthcare expenditure.