Moriarty, James P. MSc; Branda, Megan E. MS; Olsen, Kerry D. MD; Shah, Nilay D. PhD; Borah, Bijan J. PhD; Wagie, Amy E. BS; Egginton, Jason S. MPH; Naessens, James M. ScD
The economic burden of smoking and obesity is an increasing strain on the already financially stretched American health care system. A 2009 article estimated that annual medical expenditures for obesity could exceed $147 billion,1 whereas the annual medical expenditures from smoking-related diseases are estimated at $96 billion.2 These costs will be far higher when indirect costs are taken into account. A 2008 systematic review reported that costs of absenteeism associated with obesity ranged from $3.38 billion to as high as $6.38 billion per year.3 The Centers for Disease Control and Prevention reported that the annual cost of smoking-related productivity losses during 2000 to 2004 was $96.8 billion.2
One method used in the health economics literature to compare costs of an existing treatment with that of a new treatment is the examination of incremental, or additional, costs of the new treatment. A new treatment that is more expensive would have positive incremental costs compared with the existing treatment.4 This same concept of incremental cost is replicated when comparing individuals with and without certain characteristics or medical conditions. Previous literature has looked at the incremental costs due to smoking and obesity. It has been estimated that obesity can lead to an increase in medical costs as high as 38% compared with normal-weight individuals.5,6 Warner and colleagues7 performed a systematic review and found that incremental costs of smoking ranged between 6% and 8% of medical expenditures compared with nonsmokers.
Given the significant cost implications of smoking and obesity, including both direct medical costs and indirect costs such as absenteeism and productivity losses, employers have a high incentive to encourage healthy behaviors. An annual survey conducted by the Kaiser Family Foundation found that of the firms providing health benefits and wellness programs, 20% reported the primary reason for offering the wellness program was for improvement of employee health and reduction of absenteeism.8 This same survey found that among firms with health benefits, 30% provided gym membership discounts or on-site exercise facilities and 24% provided smoking cessation programs. Evidence suggests that workplace wellness programs can have a positive return on investments for both medical costs and absenteeism. A recent meta-analysis of 22 studies reporting on workplace wellness programs found an average return on investment of 3.27 for medical costs and 2.73 for absenteeism.9
Most previous reports of the incremental costs of smoking and obesity are based on single-year, cross-sectional studies. This study reports the simultaneous, 7-year estimates of incremental costs of smoking and obesity among a population of employees and their dependents having continuous insurance coverage in a large health care delivery system. This approach allows for the capture of long-term costs and determination of the persistence of costs over time. Furthermore, we determined body mass index (BMI) from clinical records rather than self-report, and assessed the effect of comorbidities. More specifically, we were interested in the impact on incremental costs when controlling for comorbidities.
All employees and dependents of the Mayo Clinic in Rochester, Minnesota aged 18 years and older who had continuous benefit coverage between 2001 and 2007 were eligible for this study. Enrollees who died during the time frame were included in order to avoid potential bias. The study population was stratified by retirement status. To adjust for possible differences in benefit payments for Medicare coverage, all enrollees who were 65 years of age at any time in the study period were classified as retired. Enrollees declining research authorization were excluded from analyses. This study was approved by our institutional review board.
Baseline characteristics of the study population were gathered from administrative and clinical databases spanning the years 1999 to 2002. Demographic data (age, gender, race, and marital status) were obtained from patient registration. Self-reported smoking status at office visit was based on patient provided information databases within our institutional medical records. Enrollees' smoking status could be reported as “current smoker,” “former,” or “never.” In the event of multiple responses, the report closest to January 1, 2001, was used. The group of enrollees not reporting any smoking status between 1999 and 2002 was categorized as unknown.
To address the large proportion of enrollees with unknown smoking status, two alternative smoking status definitions were used. Option 2 used the earliest reported smoking status between 2001 and 2007. Option 3 used smoking status data between 1999 and 2007. To generate an “upper bound” for smoking prevalence, if at any point an enrollee reported a status of “current smoker,” that enrollee was categorized as a smoker. Nonsmokers had to have at least one report of “former” or “never” without any reports of “current smoker.” In both alternatives, the enrollee was still categorized as unknown if there were no reports during the time frame.
Body mass index (BMI) was based on measured heights and weights collected during health care visits and stored in clinical notes databases. Because adult heights rarely change, values for height from any visit throughout the time frame 1999 to 2007 were allowed while still using the measurement closest to January 1, 2001. Given a minimum study age of an individual included in the study population being 18 years, heights should be consistent across the time frame. Weights were based on the closest observation to January 1, 2001, from observations within 1999 to 2002. To avoid capturing increased BMI due to pregnancy, weights of pregnant women 6 months prior to and 6 weeks after delivery date were excluded. Pregnant women and delivery dates were identified using delivery/prenatal service Current Procedural Terminology, 4th Edition codes (59400 to 59857). The BMI categories were classified as underweight (≤18), normal (>18 to 25), overweight (>25 to 30), obese (>30 to 35), morbidly obese I (>35 to 40), and morbidly obese II (>40). Individuals without any height and/or weight value were classified as “unknown.”
Chronic comorbid conditions were based on claims databases from 1999 to 2002 using International Classification of Diseases, Ninth Revision, diagnostic codes to classify into categories.10 A total of 55 categories including 6 mental disorders were used. Enrollees were classified as having 0, 1, or 2 or more comorbidities.
Annual medical care costs per enrollee were the primary outcomes of interest. Medical care costs were defined as the sum of patient and health plan paid amounts and were identified from medical claims data including pharmacy claims. Costs related to pregnancy were excluded. All costs are reported in 2007 constant dollars with inflationary adjustments based on the Consumer Price Index.11 Indirect costs were not available.
All factors for enrollees are provided using descriptive statistics by the stratification factor of retired versus non-retired. The incremental costs associated with BMI and smoking status were assessed for each year from 2001 to 2007. We used multivariate regression approaches to adjust for potential confounding at baseline. A generalized linear regression model was performed where each year's costs for each person were summed and analyzed. A general estimating equation with a gamma distribution and log link using an unstructured correlation structure was used to model cost.12–14 This structure was chosen based off the “quasi-likelihood under the independence model criterion.”14 Incremental mean costs were calculated using the method of recycled predictions along with 95% confidence limits (CL).15,16 This method uses the fitted regression model to predict incremental costs of smoking (or obesity) by subtracting the mean predicted costs assuming all enrollees are nonsmokers (or normal weight) from the mean predicted costs assuming all enrollees are smokers (or obese). Interactions between smoking and BMI were considered, as were the alternative smoking measures. All calculated P values were two-sided and P < 0.05 was considered statistically significant. Statistical analyses were performed using SAS (SAS Institute, Cary, NC) and Stata (StataCorp, College Station, TX) software.
To investigate the effects of comorbidities on model results, a comparative analysis was performed with models first excluding comorbidities and then including comorbidities. This approach was taken because obesity and smoking are known risk factors of several of the identified comorbidities. Incorporating these conditions into the model may underestimate the total incremental costs of obesity and smoking.
Baseline characteristics of the study population are depicted in Table 1. A total of 25,022 non-retirees and 5507 retirees were included in the analyses. Among the non-retirees, 3333 (13.3%) enrollees were classified as smokers at baseline, whereas 8209 (32.8%) had an unknown smoking status. Similarly, for the retirees, 963 (17.5%) were smokers and 1205 (21.9%) had an unknown smoking status. The retired population tended to have a greater percentage in the overweight, obese, and morbidly obese I categories (P < 0.001). There were 235 (0.94%) deaths for the non-retirees and 674 (12.24%) deaths for the retirees.
Mean, unadjusted annual costs, percentage of cases with zero costs, interquartile range and maximum costs stratified by retirement status are reported in Table 2. For non-retirees, the mean annual costs for the entire time frame was $7022 compared with $11,443 for retirees. The mean, unadjusted annual costs by retirement stratification were further classified by smoking status and BMI category (Fig. 1A). In both stratifications, the unknown group for both smoking status and BMI had the lowest mean, unadjusted annual costs. The mean, adjusted annual costs (without adjustments for comorbidities) are depicted in Fig. 1B.
The annual incremental mean costs of non-retirees by smoking status and BMI category, along with associated 95% CL, are depicted in Fig. 2A. Smokers had significantly higher costs ($1274; 95% CL: $746 to $1801) than nonsmokers, whereas those with an unknown smoking status had significantly lower costs than both smokers and nonsmokers(–$2544; 95% CL: –$2863 to –$2225). In terms of BMI status, significantly higher costs than those with normal BMI were found for the overweight ($382; 95% CL: $23 to $741), obese ($1850; 95% CL: $1284 to $2415), morbidly obese I ($3086; 95% CL: $2458 to $3713), and morbidly obese II ($5530; 95% CL: $4329 to $6731). The unknown group had significantly lower costs (–$2192; 95% CL: –$2633 to –$1751).
Similar results were found among the retirees. Smokers had significantly higher costs ($1401; 95% CL: $371 to $2430) than nonsmokers. Significantly higher costs were found for the morbidly obese I ($2907; 95% CL: $1262 to $4562) and morbidly obese II ($5467; 95% CL: $2427 to $8507) groups than the normal BMI group. Those with an unknown BMI status had significantly lower costs (–$3663; 95% CL: –$5034 to –$2291) than the normal weight group. Unlike the non-retirees, retirees with an unknown smoking status or BMI categories of overweight and obese did not have significantly different costs compared with the reference groups.
When comorbidities were included in multivariate analysis, many of the significant differences in incremental costs remain, but to a lesser degree (Fig. 2A). For non-retirees, the incremental costs of smoking decreased to $865 (95% CL: $471 to $1258), yet remained within the bounds of the 95% CL of the base-case analysis. Conversely, incremental costs of obesity, morbid obesity I, and morbid obesity II dropped well outside the bounds of the 95% CL of the base-case analysis while maintaining statistically significant differences compared with normal weight individuals (Fig. 2A). For retirees, the adjustment of comorbidities had a similar effect (Fig. 2B).
Alternative measures for smoking status reduced the unknown category to only 5% and 8% of the non-retirees and retirees, respectively. For non-retirees, the incremental costs of smoking with option 2 remained relatively unchanged at $1145 (95% CL: $733 to $1557), whereas option 3 resulted in higher incremental costs of $2058 (95% CL: $1704 to $2413). In this alternative approach, 30% of individuals were categorized as smokers. For retirees, the incremental costs of smoking with option 2 also remained relatively unchanged at $1531 (95% CL: $588 to $2473), whereas option 3 resulted in higher incremental costs of $24,507 (95% CL: $1734 to $3280). This approach resulted in 39% of retirees being deemed as smokers. No significant interactions between smoking and BMI were observed.
This study extends the knowledge about the impact of smoking and obesity on health care costs in several ways. Simultaneous estimates of incremental costs of smoking and obesity show that these factors seem to act as independent multiplicative factors. Smoking and obesity significantly increased the costs of the adult enrollee population of our institution. This was the case in both the non-retiree and retiree cohorts. Furthermore, we saw high cost persistence of these health behaviors over 7 years. Estimates of the effect of comorbidities and the relationship between smoking, BMI and chronic conditions were provided.
Our results are consistent with previous studies. An earlier study using a 7-year longitudinal analysis in a health maintenance organization found mean annual incremental costs of smoking to be $1428.17 A systematic review of the literature conducted in 2010 found incremental costs of obese individuals (BMI 30+) to be $1474 (in 2009 US dollars).18 In addition, Arterburn et al19 reported incremental costs of morbidly obese II individuals (referred to in their study as class III obese) of $2011 (in 2000 US dollars). Couple these increased long-term incremental costs of obesity with the reported increase in the proportion of obese people in the United States, and one quickly concludes that obesity will be a significant driver of total costs of this population.20
Data used in this study have distinct advantages over previous studies that relied on cross-sectional samples of the US population. In cross-sectional studies, the ability to track an individual's expenditures over a prolonged time is limited or impossible, depending on the data source. Longitudinal assessments, like the one used in this study, capture longer-term costs associated with smoking and obesity and assess the persistence of costs. Mean annual costs will be less likely to be affected by both high- and low-cost outliers. True cost differences due to smoking or obesity will be shown by higher costs consistently over time. Furthermore, previous studies of obesity relied on self-reported measurements, as detailed medical records containing BMI data were not obtainable.21 It has been found that self-reported BMI is lower than BMI measured by a technician due to underreporting of weight.22 In addition, our data allowed for the exclusion of BMI values gathered from women during pregnancy rather than excluding pregnant women entirely from the analysis. This was not possible for many previous analyses.1,4,19
This study also illustrates that adjustment for comorbidities substantially affects the incremental costs of obesity. Given that obesity is a risk factor of many comorbidities, controlling for these issues is likely to underestimate the true incremental costs of obesity. We were not able to disentangle conditions that preceded the individual's risk factor from those that resulted from obesity. Our results show that when comorbidities are not controlled for, there is a two- to threefold increase for the incremental costs of those with a BMI of 30 or greater compared with those with normal BMI. This could lead to significant policy implications.
It must also be noted that our base-case results included a large portion of enrollees with an unknown smoking status (33% non-retirees and 22% retirees). This is partly due to the self-report only being taken from those that seek care at our institution. Non-users of medical care would not have the opportunity to provide the information. In addition, those seeking care but choosing to do so at another health care facility would also not have the opportunity to provide the information. The high frequency of unknown smoking status is also partly due to definition. Our baseline assessment only considered the years 1999 to 2002. Both alternative approaches resulted in the unknown category decreasing to comprise 5% and 8% of the non-retirees and retirees, respectively. Option 2 had little impact on incremental costs, whereas option 3 resulted in slightly higher incremental costs. We would consider our base-case classification of smoking status to be a lower bound of the incremental costs of smoking, whereas option 3 might be an upper bound.
Several limitations of this study should be noted. Because these results come from a single health care system, generalizability may be limited. In addition, because of the fact that the analyses are based on those with continuous insurance coverage, results may be sensitive to whether individuals with multiple chronic conditions vary in likelihood of retaining insurance coverage. Furthermore, although cost data were analyzed longitudinally, our multivariate analyses included only baseline covariates. Changes in smoking status or BMI category were also not taken into account. In addition, our findings of the incremental cost of smoking are limited on the basis of smoking status being determined from self-report, unlike our BMI classification, which was based on medical records.
Incremental costs of smoking are significant, and costs of individuals classified in obese (preretirement), morbid obesity I, and morbid obesity II categories are significantly higher than normal weight individuals with increasing cost increments at higher weight categories. In our study, the incremental costs of smoking seem to be about the size of the incremental costs of being obese and lower than in the morbid obesity categories. Controlling for comorbidities had a large impact on incremental costs of obesity and could lead to underestimation of the true incremental costs of obesity because obesity is a risk factor for the development of many chronic conditions. Additional research is warranted to address this issue.
1. Finkelstein EA, Trogdon JG, Cohen JW, Dietz W. Annual medical spending attributable to obesity: payer-and-service-specific estimates. Health Aff. 2009;28:w822–w831.
2. Centers for Disease Control and Prevention. Smoking-attributable mortality, years of potential life lost, and productivity losses—United States, 2000–2004. MMWR Morb Mortal Wkly Rep. 2008;57:1226–1228.
3. Trogdon J, Finkelstein EA, Reyes M, Dietz WH. Indirect costs of obesity: a review of the current literature. Obes Rev. 2008;9:489–500.
4. Drummond MF, Sculpher MJ, Torrance GW, O'Brien BJ, Stoddart GL. Methods for the Economic Evaluation of Health Care Programmes. 3rd ed. New York, NY: Oxford University Press. 2005:66.
5. Finklestein E, Flebelkorn I, Wang G. National medical spending attributable to overweight and obesity: how much, and who's paying? Health Aff. 2003;(suppl Web Exclusives):W3-219–W3-226.
6. Sturm R. The effects of obesity, smoking and drinking on medical problems and costs. Health Aff. 2002:21;245–253.
7. Warner K, Hodgson T, Carroll C. Medical costs of smoking in the United States. Tob Control. 1999;8:290–300.
9. Baicker K, Cutler D, Song Z. Workplace wellness programs can generate savings. Health Aff. 2010;29:304–311.
10. Naessens JM, Stroebel RJ, Finnie DM, et al. Effect of multiple chronic conditions among working-age adults. Am J Manag Care. 2011;17:118–122.
12. Liang K-Y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika.1986;73:13–22.
13. Hardin JW, Hilbe JM. Generalized Estimating Equations. Boca Raton, FL: Chapman and Hall/CRC; 2003
14. Pan W. Akaike's information criterion in generalized estimating equations. Biometrics. 2001;57:120–125.
15. Basu A, Rathouz PJ. Estimating incremental and marginal effects on health outcomes using flexible link and variance function models. Biostatistics. 2005;6:93–109.
16. Basu A, Arondekar BV, Rathouz PJ. Scale of interest versus scale of estimation: comparing alternative estimators for the incremental costs of a comorbidity. Health Econ. 2006;15:1091–1107.
17. Fishman PA, Khan ZM, Thompson EE, Curry SJ. Health care costs among smokers, former smokers, and never smokers in an HMO. Health Serv Res. 2003;38:733–749.
19. Arterburn DE, Maciejewski ML, Tsevat J. Impact of morbid obesity on medical expenditures in adults. Int J Obes. 2005;29:334–339.
21. Faeh D, Marques-Vidal P, Chiolero A, Bopp M. Obesity in Switzerland: do estimates depend on how body mass index has been assessed? Swiss Med Wkly. 2008;138:204–210.
22. McAdams MA, Van Dam RM, Hu FB. Comparison of self-reported and measured BMI as correlates of disease markers in U.S. adults. Obesity (Silver Spring). 2007:15:188–196.
This article has been cited 2 time(s).
Journal of Health Care for the Poor and Underserved
Equal and Universal Access? Water at Mealtimes, Inequalities, and the Challenge for Schools in Poor and Rural Communities
Journal of Health Care for the Poor and Underserved, 24(2):
Plos OnePrevalence of Overweight, Obesity, and Abdominal Obesity in a Representative Sample of Portuguese AdultsPlos One
©2012The American College of Occupational and Environmental Medicine