Research in Social and Administrative Pharmacy j (2016) j–j

Original Research

Incremental impact of body mass status with modifiable unhealthy lifestyle behaviors on pharmaceutical expenditure Tae Hyun Kim, Ph.D.a, Eui-Kyung Lee, Ph.D.b, Euna Han, Ph.D.c,* a

Graduate School of Public Health and Institute of Health Services Research, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul, South Korea b School of Pharmacy, Sungkyunkwan University, 300 Cheonchoen-dong, Jangan-gu, Suwon, Gyeonggi-do 440-746, South Korea c College of Pharmacy and Yonsei Institute of Pharmaceutical Sciences, Yonsei University, 162-1 Songdo-Dong, Yeonsu-Gu, Incheon, South Korea

Abstract Background: Overweight/obesity is a growing health risk in Korea. The impact of overweight/obesity on pharmaceutical expenditure can be larger if individuals have multiple risk factors and multiple comorbidities. The current study estimated the combined effects of overweight/obesity and other unhealthy behaviors on pharmaceutical expenditure. Methods: An instrumental variable quantile regression model was estimated using Korea Health Panel Study data. The current study extracted data from 3 waves (2009, 2010, and 2011). Results: The final sample included 7148 person-year observations for adults aged 20 years or older. Overweight/obese individuals had higher pharmaceutical expenditure than their non-obese counterparts only at the upper quantiles of the conditional distribution of pharmaceutical expenditure (by 119% at the 90th quantile and 115% at the 95th). The current study found a stronger association at the upper quantiles among men (152%, 144%, and 150% at the 75th, 90th, and 95th quantiles, respectively) than among women (152%, 150%, and 148% at the 75th, 90th, and 95th quantiles, respectively). The association at the upper quantiles was stronger when combined with moderate to heavy drinking and no regular physical check-up, particularly among males. Conclusion: The current study confirms that the association of overweight/obesity with modifiable unhealthy behaviors on pharmaceutical expenditure is larger than with overweight/obesity alone. Assessing the effect of overweight/obesity with lifestyle risk factors can help target groups for public health intervention programs. Ó 2016 Elsevier Inc. All rights reserved. Keywords: Overweight/obesity; Pharmaceutical expenditure; Instrumental variable quantile regression model; Unhealthy behavior

Conflict of interest: All authors declare that they have no conflict of interest. * Corresponding author. Tel.: þ82 10 9334 7870, þ82 32 749 4511; fax: þ82 32 749 4105. E-mail address: [email protected] (E. Han). 1551-7411/$ - see front matter Ó 2016 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.sapharm.2015.12.009

2

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Introduction Obesity is a global public health concern, given that obesity prevalence has more than doubled since 1980. Of adults worldwide aged 18 years and older, 39% are obese and 13% are overweight in 2014.1 Korea has a similar public health problem, with 31.7% of adults being overweight or obese in 2007, up from 26.0% in 1998.2 Obesity is a risk factor for many illnesses, and thus it is not surprising that obese people are likely to use more health care services and have higher health expenditure than their non-obese counterparts.3–10 Obese people consume a disproportionate share of health care resources.11,12 Large health care expenditures attributable to overweight or obesity (OW/OB) are also reported in Korea.13,14 Medication is a key factor in managing diseases, particularly chronic diseases, although individual health behaviors and other health care services are important as well.15 Pharmaceutical expenses are even more important in Korea, where 29% of total health expenditure in the National Health Insurance Service (NHIS) is pharmaceutical spending and the annual increase in pharmaceutical spending for the past decade has been much higher than the increase in total health expenditure.16 However, except for the work of Cawley and Meyerhoefer,17 previous studies have estimated the association only at the mean of pharmaceutical expenditure, without addressing potential heterogeneity across the entire conditional distribution of the pharmaceutical expenditure. For example, obesity may impose a greater burden to pharmaceutical expenditure for those with higher drug expenses. If obesity is associated with pharmaceutical expenditure only at the bottom or top of the distribution of expenditure, mean estimation using ordinary least squares (OLS) can mask or underestimate potential statistical significance at the ends of the distribution.2 The impact of OW/OB on pharmaceutical expenditure can be larger if individuals have multiple risk factors and multiple comorbidities.18 For example, the effect of obesity on health expenditure is reported to increase when obesity is combined with physical inactivity.19–23 However, few studies have estimated the association of obesity with pharmaceutical expenditure when lifestyle risk factors other than physical inactivity simultaneously exist with obesity. A study by Alter et al24 is one exception, and they reported that obesity was not statistically significantly associated with

health care costs when it is a sole risk factor, whereas obesity in combination with other lifestyle risk factors such as smoking, physical inactivity, and/or social distress was associated with significantly higher health care cost. The current study estimates whether and how much obesity incrementally affects pharmaceutical expenditure when combined with other modifiable unhealthy lifestyle behaviors. A quantile regression model was used to explore the direction, magnitude, and statistical significance of the association of OW/OB with pharmaceutical expenditure when OW/OB co-existed with other modifiable lifestyle risk factors across the entire distribution of pharmaceutical expenditures. An instrumental variable (IV) approach was used to estimate the causal relationship of OW/OB with pharmaceutical expenditure. Assessing the comprehensive impact of OW/OB with concomitant lifestyle risk factors can be useful for targeting groups for public intervention programs.

Methods Data The current study used data from the Korea Health Panel Study, a nationally representative annual panel survey started in 2009. It has a vast amount of information on health, health service use, and expenditure, as well as standard demographic and socioeconomic information at both the household and individual levels. The merit of this survey is that it extracts health service use and health expenditure (including pharmaceutical expenditure) information from NHIS, the sole public health insurer in Korea, with almost all Koreans included as compulsory beneficiaries. The survey sample is based on a two-stage stratified cluster sampling. Regional clusters were randomly sampled and households in the sampled clusters were systematically sampled. Approximately 350 clusters were sampled and approximately 8000 households in those clusters were surveyed in every wave. In each survey round, additional households were sampled to make up for attrition.24 The current study extracted data from 3 waves (2009, 2010, and 2011). The final sample for the current study included 7148 person-year observations for adults aged 20 years or older after dropping observations with missing information for any variables used in the estimation.

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Variables The outcome variable is defined as total annual pharmaceutical expenditure for each sample person. The Korea Health Panel Study data draws information on expenditure from administrative claims in NHIS, and thus, expenditure in the current study accounts for what NHIS reimbursed and patients paid out-of-pocket for listed drugs. Pharmaceutical expenditure was log-transformed, allowing estimation of the relative change associated with body mass status and normalizing skewed distribution of pharmaceutical expenditure to improve efficiency of estimation. A dummy indicator was generated for any positive pharmaceutical expenditure with the reference of zero pharmaceutical expenditure. The predictor variable of interest is body mass status. A dummy indicator for OW/OB was generated, which was defined as body mass index (BMI; weight in kilograms divided by height in meters squared) of 25 or larger, with BMI less than 25 as the reference group. It is worth noting that the World Health Organization (WHO) recommends defining overweight as a BMI between 25 and 30 and obese as a BMI 30 or larger,25 whereas the World Health Organization Regional Office for the Western Pacific (WPRO) recommends a BMI of 25 or larger as obesity for Asians.26 The current study focuses on the combined category of overweight or obesity to align with the WPRO guideline for Asians. Biological family members’ OW/OB status was generated as the just-identifying instrumental variable for the respondent’s OW/OB based on the average BMI of all biological family members included in the survey. Both height and weight information were self-reported. Other key independent variables included indicators representing each of the following 4 lifestyle unhealthy behaviors: current smoking; moderate to heavy drinking; being physically inactive; and no regular physical check-up. Moderate to heavy drinking was defined as having five, 0.355 L cans of beer for men (with total alcohol content of 62.5 g) or three cans for women (with total content of alcohol 37.5 g) at one time at least once in a given month. Current smokers were defined as respondents who answered “everyday” or “sometimes” to a survey question asking, “Do you currently smoke cigarettes?” Respondents were classified as physically inactive if they answered that they never engaged in active walk or any other form of physical activity in a given

3

week. No regular physical check-up was defined as non-participation in the biennial general physical check-up or twice-a-life cancer screening, both of which are provided free by the NHIS in Korea. Other individual characteristics were controlled for as covariates in all estimations. These included age, in a linear measure of the median in categories by 5-year span in the original data, as well as a linear measure of total annual household income in Korean Won (KW). The Charlson Comorbidity Index27 was also controlled for as a linear variable to account for the total number of comorbidities weighted by the severity of the respective comorbidity. Other covariates included in the estimations as dummy indicators are residential area (living in metropolitan areas or medium to small cities, with living in rural areas as the reference), marital status (married, with not being married as the reference), education level (college or more or high school graduate, with less than college as the reference), number of children, employment status (self-employed, full-time employees, or part-time employees with being unemployed as the reference), and survey years 2010 and 2011 with 2009 as the reference.

Statistical analysis First, a two-part model28 was estimated to obtain the overall pharmaceutical expenditure adjusting for the probability of any use of pharmaceuticals. The probability of any pharmaceutical use was estimated for the entire sample in the first part, and actual pharmaceutical expenditure was estimated conditional on any use among a subset with positive values of pharmaceutical expenditure in the second part. The association of being overweight or obese with the overall pharmaceutical expenditure was calculated as shown in Equation (1): vE½Y vðUðbXÞ  E½yjyO0Þ ¼ vX vX   vE½yjyO0 ¼ Pr½yO0  vX   vPr½yO0 þ E½yjyO0  vX

ð1Þ

The variables Y and X, respectively, represent pharmaceutical expenditure and the set of explanatory variables, including OW/OB. The b’s are parameters to be estimated.

4

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Given that both body mass status and pharmaceutical expenditure were measured at the same time, t, it is plausible that pharmaceutical expenditure affected body mass status via adverse drug effects or other unobserved medical effects. To establish causality in the association of body mass status with pharmaceutical expenditure, an instrumental variables (IV) approach was used, which takes only the variation of the predictor variable of interest predicted by the IV method.29 Biological family members’ OW/OB status was used as an IV for the respondent’s OW/OB status in this study’s estimations. The validity of the IV approach is based on two assumptions: the IV strongly predicts the predictor variable of interest; and the IV is associated with the outcome variable only through the predictor variable of interest, of which variation is predicted by the IV.30 The strength of the IV for the current study was supported with F ¼ 50.30 for the null hypothesis of no association of the IV with the predictor variable of interest in the first stage, which far exceeded the recommended minimum F-value for strong IV.31 The second condition for a valid IV indicates that OW/OB status of biological family members should be associated with the respondent’s pharmaceutical expenditure only through the OW/OB status of the respondent. This condition would be satisfied if it can be assumed that any unobserved shared family environment predicting the respondent’s pharmaceutical expenditure does not predict body mass status of either the respondents or their biological family members. In fact, a large body of research in behavioral genetics has detected no effect of shared family environment on body mass status,32,33 and several studies in health economics adopted this strategy of using family members’ body mass status as IV for self’s body mass status.17,34–37 A multivariate quantile regression model was then estimated on the linear measure of pharmaceutical expenditure among those with positive pharmaceutical expenditure to assess heterogeneity in the association of OW/OB status with pharmaceutical expenditure in different quantiles of pharmaceutical expenditure. The quantile regression estimation as follows: qt ðDRUG EXPi Þ ¼ d0t þ dOWOB OWOBi t þ dXt Xi þ εit

ð2Þ

In Equation (2), DRUG_EXP denotes a linear measure of pharmaceutical expenditure and

OWOB is a dummy indicator for the clinical classification of OW/OB status based on BMI. The variable t represents the tth quantile of pharmaceutical expenditure, and i represents the individual. The unit of analysis is individual-year. Variations were explored in the parameter estimates dOWOB and dX I I for OW/OB and each covariate in quantiles of pharmaceutical expenditure. Frolich and Melly’s38 conditional quantile treatment effect was used to estimate a quantile IV model, which requires discrete IV for discrete treatment variables. Subgroup analyses were run by gender (male vs. females) and age group (less than 65 vs. 65 or older), since the elderly have a weaker association of OW/OB with pharmaceutical expenditure than adults younger than 65 years old, while they are likely to use more health services with greater expenditures than their younger counterparts39 and have other risk factors incurring pharmaceutical expenditure.40 To estimate moderation of the impact of OW/OB on pharmaceutical expenditure by health behaviors, analyses were run for subgroups that reported health behaviors of interest. All statistical analyses were performed using Stata 13.1 (StataCorp, College Station, TX, USA).

Results Table 1 shows pharmaceutical expenditures for each independent variable. Average drug cost was 111,506 kW (approximately 111 US. dollars [USD]) with 2500 kW at the 5th quantile and 487,030 kW at the 95th. Slightly less than a quarter (23.61%) of the present study’s sample was overweight or obese. Females comprised more than half (62.20%) of the sample, and the sample persons were aged 49 years old on average. Most study subjects were moderate to heavy drinkers (70.03%) or did not undergo regular physical check-ups (86.24%), whereas current smokers (16.74%) or physically inactive subjects (18.35%) were less than one fifth of the sample. The average Charlson Comorbidity Index was 0.88. More than one-half of the sample persons were married, lived in large cities, and had a college or higher education. The annual total pharmaceutical expenditure for the elderly was 218,490 kW (approximately 218 USD) on average, approximately 3 times the expenditure for the non-elderly (84,610 kW). Male annual total pharmaceutical expenditure

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

5

Table 1 Study subject characteristics Variable Pharmaceutical expenditure Average At 5th quantile At 10th quantile At 25th quantile At 50th quantile At 75th quantile At 90th quantile At 95th quantile Body mass status Overweight or obese (BMI R 25) Underweight or normal weight (BMI ! 25, reference) Biological family members’ body mass status Overweight or obese (BMI R 25) Underweight or normal weight (BMI ! 25, reference) Gender Male Female (reference) Health behaviors Current smokinga Yes No (reference) Moderate to heavy drinkingb Yes No (reference) Physically inactivec Yes No (reference) Lacking regular physical check-upd Yes No (reference) Comorbidity Charlson Comorbidity Index Residential area Large city Small city or rural (reference) Marital status Married Not married (reference) Education level College or more High school graduate Less than high school (reference) Work status Self-employed Employees in permanent job Employees in temporary job Not employed (reference) Survey year Year 2011 Year 2010 Year 2009 (reference)

Mean (standard deviation), N (%) 111,506 (177,457) [300, 1,800,300] 2500 3700 9540 34,645 136,545 322,480 487,030 1687 (23.61) 5461 (76.39) 1407 (19.68) 5741 (80.32) 2702 (37.80) 4446 (62.20)

1197 (16.74) 5951 (83.26) 5006 (70.03) 2142 (29.97) 1515 (21.19) 7156 (78.81) 6164 (86.24) 984 (13.76) 0.8801 (1.1048) [0, 7] 3580 (50.09) 3568 (49.91) 4142 (57.96) 3006 (42.04) 2511 (35.14) 3072 (42.99) 1565 (21.87) 1356 1327 1402 3063

(18.98) (18.57) (19.62) (42.83)

2482 (34.73) 2218 (31.04) 2448 (34.23) (Continued)

6

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Table 1 (Continued ) Variable

Mean (standard deviation), N (%)

Age in years (mean, SD)

49.17 (17.76) [22.50, 92.50] 1228 (835) [0, 16,861] 7148

Yearly total household income in 10,000 Korean Won N

a Current smokers were defined as respondents who answered “everyday” or “sometime” to a survey question asking, “Do you currently smoke cigarettes?”. b Moderate to heavy drinking was defined as having 5 0.355 L cans of beer for men (with total alcohol content of 62.5 g) or 3 cans for women (with total alcohol content of 37.5 g) at one time at least once in a given month. c Respondents were classified as physically inactive if they answered that they never engaged in active walking or any other physical activity in a given week. d No regular physical check-ups was defined as participation in no biennial general physical check-ups or twice-a-life cancer screening, both of which are provided biennially for free by the National Health Insurance Service in Korea, or cancer screening with partial or full out-of-pocket payment.

(128,928 kW) was higher than that of females (101,301 kW). Being physically inactive was associated with the largest pharmaceutical expenditure for the whole sample and across 4 subgroups of gender and age (except for non-elderly), ranging between 90,139 kW (non-elderly) and 229,472 kW (elderly). Among females, current smokers spent the most on pharmaceuticals (111,393 kW), while among males the highest spending occurred by those not engaging in regular physical check-ups (124,516 kW) and the elderly (216,059 kW) (Table 2). Table 3 shows results from a two-part model of mean estimation to compare results of pharmaceutical expenditure only among users of pharmaceuticals and results for overall expenditure, adjusting for the likelihood of any use of pharmaceuticals. The estimated association of being overweight or obese with pharmaceutical expenditure was enlarged by adjusting the likelihood of any pharmaceutical usage in a two-part model compared to results based on only users of pharmaceuticals. This is expected given that the direction of the association of being overweight or obese with the likelihood of positive pharmaceutical use was positive although it was not statistically significant across all subgroups. For the overall expenditure, overweight or obesity was associated with 27.5% higher pharmaceutical expenditure, the amount of which increased to 35.4% after adjusting for the likelihood of any use. Table 4 shows quantile regression results for IV models for the subjects and for subgroups by gender and age group. The difference in annual total pharmaceutical expenditure between OW/OB and non-OW/OB groups varied across the

conditional distribution of pharmaceutical expenditure. Overweight or obese individuals spent more on total pharmaceutical expenditure annually than their under- or normal-weight counterparts only at the upper quantiles of pharmaceutical expenditure (by 119% at the 90th and 115% at the 95th quantiles). Further analyses by gender and age group also revealed a positive impact of OW/OB on pharmaceutical expenditure only at the upper quantiles of the conditional distribution of pharmaceutical expenditure. There were gender differences in the association of OW/OB with pharmaceutical expenditure, with a stronger association of OW/OB at the upper quantiles among men (by 151%, 144%, and 150% at the 75th, 90th, and 95th quantiles, respectively) than among women (by 152%, 150%, and 148% at the 75th, 90th, and 95th quantiles, respectively). Moreover, age differences were found such that OW/OB increased pharmaceutical expenditure only at the 75th quantile (by 101%) among adults, whereas no statistically significant impact was found among the elderly. Table 5 shows differences in total annual pharmaceutical expenditure for subgroups with a specific unhealthy behavior of interest, examined to assess whether the impact of OW/OB on pharmaceutical expenditure is moderated by each behavior. Similar to the entire sample, difference in incremental total annual pharmaceutical expenditure by OW/OB status was statistically significant only at the upper quantiles of pharmaceutical expenditure among moderate to heavy drinkers, those not undergoing any regular physical checkups, and subgroups with at least one unhealthy behavior of interest. For moderate to heavy

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

7

Table 2 Distribution of pharmaceutical expenditure by subgroup Pharmaceutical expenditure

Male

Whole group

128,928 (199,659) [400, 1,800,300] (N ¼ 2702) 111,050 (180,180) [400, 1,800,300] (N ¼ 2168) 110,474 (182,219) [500, 1,433,480] (N ¼ 1103) 145,208 (195,113) [500, 992,600] (N ¼ 947) 124,516 (197,654) [400, 1,800,300] (N ¼ 2398)

Female

Adults

Elderly

Mean (standard deviation), [minimum, maximum]

Heavy drinkersb

Current smokersa

Being physically inactivec

Not undergoing any physical check-upsd

101,301 (161,664) [300, 1,646,100] (N ¼ 4446) 79,074 (37,609) [500, 1,646,100] (N ¼ 2838) 111,393 (200,168) [1200, 1,371,600] (N ¼ 94) 118,316 (171,917) [500, 1,646,100] (N ¼ 947) 96,067 (158,510) [300, 1,646,100] (N ¼ 3767)

84,610 (148,649) [300, 1,800,300] (N ¼ 5712) 75,323 (139,737) [500, 1,800,300] (N ¼ 4299) 90,403 (160,007) [500, 1,069,260] (N ¼ 952) 90,139 (149,023) [500, 1,646,100] (N ¼ 1099) 78,647 (142,705) (300, 1,800,300) (N ¼ 4887)

218,490 (233,599) [400, 1,433,480] (N ¼ 1436) 199,934 (212,823) [400, 1,433,480] (N ¼ 707) 188,817 (240,235) [1000, 1,433,480] (N ¼ 425) 229,472 (217,092) [500, 1,075,500] (N ¼ 416) 216,059 (235,497) [400, 1,433,480] (N ¼ 1278)

a Current smokers were defined as respondents who answered “everyday” or “sometime” to a survey question asking, “Do you currently smoke cigarettes?” b Moderate to heavy drinking was defined as having 5 0.355-L cans of beer for men (with total alcohol content of 62.5 g) or 3 cans for women (with total alcohol content of 37.5 g) at one time at least once in a given month. c Respondents were classified as physically inactive if they answered that they never engaged in active walking or any other physical activity in a given week. d No regular physical check-ups was defined as participation in none of biennial general physical check-ups or twicea-life cancer screening, both of which are provided biennially for free by the National Health Insurance Service in Korea, or cancer screening with partial or full out-of-pocket payment.

drinkers, overweight or obese adults spent 153%, 127%, and 143% more on pharmaceutical expenditure at the 75th, 90th, and 95th quantiles of pharmaceutical expenditure than their counterparts. Overweight or obese people lacking regular physical check-ups on the 90th and 95th quantiles of pharmaceutical expenditure had larger pharmaceutical expenditure by 124% and 131%, respectively, than their counterparts. Those overweight or obese individuals who had at least one unhealthy behavior (moderate to heavy drinking, current smoking, being physically inactive, or no regular physical check-ups) similarly showed larger pharmaceutical expenditure (by 138%, 133%, and 132% at the 75th, 90th, and 95th quantiles, respectively) than their counterparts. Lastly, in Table 6 each subgroup is grouped with an unhealthy behavior by age and gender.

Statistically significantly higher pharmaceutical expenditure by OW/OB was found only at the upper quantiles among moderate to heavy drinkers in only the male and adult subgroups, and among those without any regular physical check-ups in both the male and female subgroups. The impact of OW/OB on pharmaceutical expenditure was moderated upward among males (at the 75th, 90th, and 95th quantile, by 173% and 153%, and 156%) and non-elderly moderate to heavy drinkers (at both 75th and 90th quantiles, by 168% and 143%, respectively) compared to all males or all non-elderly adults. Males without any regular physical check-ups had higher pharmaceutical expenditure by OW/OB (by 199% at the 75th quantile and 161% at the 90th quantile), both of which were larger than the expenditure for all males. For females without regular physical

8

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Table 3 Mean estimation of the association of overweight or obesitya with annual total pharmaceutical expenditure with and without adjustment for the probability of any pharmaceutical use Dependent variable: Log of pharmaceutical expenditure

Coefficient (standard error) Before adjustment for the probability of use

After adjustment for the probability of use with twopart model

Whole (N ¼ 7148) Subgroup by gender Male (N ¼ 2702) Female (N ¼ 4446) Subgroup by age group Adults (N ¼ 5712) Elderly (N ¼ 1436) Subgroup by unhealthy behaviors Current smokers (N ¼ 1197) Moderate to heavy drinkers (N ¼ 5006) Being physically inactive (N ¼ 1515) No regular physical check-up (N ¼ 6165) One of four behaviorsb (N ¼ 6946)

0.2759*** (0.0388)

0.3546*** (0.0899)

0.2017*** (0.0624) 0.3464*** (0.0501)

0.2968** (0.1424) 0.3994*** (0.1249)

0.2674*** (0.0434) 0.1985** (0.0893)

0.2823** (0.1049) 0.4550** (0.1943)

0.1153 (0.0941)

0.0753 (0.2149)

0.2522*** (0.0451)

0.3498*** (0.1117)

0.2297 (0.1851)

0.2677 (0.1965)

0.2521*** (0.0422)

0.3346*** (0.1000)

0.2643*** (0.0393)

0.3430*** (0.0931)

*P ! 0.1, **P ! 0.05, ***P ! 0.01. a A dummy indicator for overweight or obesity was defined as body mass index, weight in kilograms divided by height in meters squared, of 25 or larger. b Other individual characteristics controlled for as covariates in all estimations were the following: age, in a linear measure of the median in categories by 5-year span; a linear measure of total annual household income in Korean Won (KW); Charlson Comorbidity Index as a linear variable. Additionally, dummy indicators include residential area (living in metropolitan areas or medium-to-small cities, with living in rural areas as the reference); marital status (married, with not being married as the reference); education level (college or more or high school graduate, with less than college as the reference); number of children; employment status (self-employed, full-time employees, or part-time employees, with being unemployed as the reference); and survey years 2010 and 2011, with 2009 as the reference.

check-ups, OW/OB was associated with incrementally higher pharmaceutical expenditure by 133%, 138%, and 137% at the 75th, 90th, and 95th quantiles, respectively, which were all lower than the expenditure for the entire female sample. No statistically significant impact of OW/OB on pharmaceutical expenditure was found for current smokers or those who were physically inactive across gender and age subgroups.

Discussion Findings of the current study confirmed the heterogeneity of the association of OW/OB with pharmaceutical expenditure across the entire conditional distribution of pharmaceutical expenditure. Overweight/obese people had higher pharmaceutical expenditures than their

counterparts only at the upper quantiles of pharmaceutical expenditure. The findings of the current study revealed that the association at the upper quantiles was stronger when OW/OB was combined with unhealthy behaviors, particularly moderate to heavy drinking and lack of regular physical check-ups. The association was also stronger among male moderate to heavy drinkers, males lacking regular physical check-ups, and non-elderly moderate to heavy drinking adults compared to all males or non-elderly at the upper quantiles. Pharmaceutical expenditure is an important public health and policy issue, but often challenged by a lack of quality data. A few studies linked large individual survey datasets including socio-demographic information and comprehensive insurance claims data with detailed information on health service use to estimate the impact of OW/OB on pharmaceutical expenditure.41,42

Group

Whole (N ¼ 7148) Male (N ¼ 2702) Female (N ¼ 4446) Adults (N ¼ 5712) Elderly (N ¼ 1436)

Quantiles in the conditional distribution of log of pharmaceutical expenditurea Coefficient for overweight/obesityb (standard error) 5th

10th

25th

50th

75th

90th

95th

0.8751 (3.0473) 0.2231 (1.0421)

0.9337 (2.0467) 0.5906 (1.0066)

0.9257 (1.3625) 0.8512 (1.0282)

0.3457 (1.1086) 0.6485 (1.4729)

1.1070 (0.7594) 1.5184*** (0.5657)

1.1962** (0.4878) 1.4474** (0.7273)

1.1512** (0.5498) 1.5019* (0.8527)

1.0054 (2.5573)

1.0641 (2.2159)

1.2468 (3.3756)

1.8090 (1.3373)

1.5267* (0.7470)

1.5070* (0.6646)

1.4856* (0.8177)

0.2191 (2.1326)

0.2657 (1.7325)

0.0128 (1.1321)

0.3962 (1.0878)

1.0127** (0.4676)

0.8138 (0.8180)

0.6192 (0.6486)

4.1743 (4.3129)

3.9025 (3.5537)

3.1582 (3.1349)

1.6080 (3.2744)

1.2584 (1.6130)

1.2584 (1.8944)

1.1929 (2.1358)

*P ! 0.1, **P ! 0.05, ***P ! 0.01. a Other individual characteristics controlled for as covariates in all estimations were the following: age, in a linear measure of the median in categories by 5-year span; a linear measure of total annual household income in Korean Won (KW); Charlson’s Comorbidity Index as a linear variable. Additionally, dummy indicators are residential area (living in metropolitan areas or medium-to-small cities, with living in rural areas as the reference); marital status (married, with not being married as the reference); education level (college or more or high school graduate, with less than college as the reference); number of children; employment status (self-employed, full-time employees, or part-time employees, with being unemployed as the reference); and survey years 2010 and 2011, with 2009 as the reference. b A dummy indicator for overweight or obesity was defined as body mass index, weight in kilograms divided by height in meters squared, of 25 or larger.

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Table 4 Heterogeneous association of overweight or obesity with annual total pharmaceutical expenditure: quantile IV regression results

9

10

Table 5 Quantile instrumental variable regression results for subgroup by unhealthy behaviors Quantiles in the conditional distribution of log of pharmaceutical expenditureg Coefficient for overweight/obesityf (standard error) 5th

10th

25th

50th

75th

90th

95th

Moderate to heavy drinkera (N ¼ 5006) No regular physical checkupsb (N ¼ 6165) Current smokersc (N ¼ 1197) Being physically inactived (N ¼ 1515) One of four behaviorse (N ¼ 6946)

0.4997 (1.8335)

0.0548 (1.0790)

0.6020 (1.1787)

0.7052 (0.8337)

1.5363** (0.6308)

1.2741* (0.7486)

1.4374* (0.7584)

0.3794 (1.7859)

0.2998 (1.8156)

0.4207 (1.5414)

0.5257 (1.1054)

1.2179 (0.7230)

1.2438** (0.6118)

1.3101* (0.6973)

0.6104 (1.6341)

0.8231 (1.4213)

0.3053 (1.9167)

0.1135 (1.3869)

0.6095 (2.5905)

1.4716 (1.8271)

1.2931 (2.8990)

0.7302 (5.8587)

0.6849 (5.0724)

0.8325 (8.4250)

0.5603 (2.7603)

0.4126 (1.3897)

0.5922 (1.5072)

0.6518 (1.7152)

0.0789 (1.7166)

0.0175 (1.3544)

0.5911 (1.0582)

1.2796 (1.0809)

1.3873* (0.6687)

1.3324** (0.5671)

1.3214** (0.6164)

*P ! 0.1, **P ! 0.05, ***P ! 0.01. a Moderate to heavy drinking was defined as having 5 0.355-L cans of beer for men (with total alcohol content of 62.5 g) or 3 cans for women (with total alcohol content of 37.5 g) at one time at least once in a given month. b No regular physical check-ups was defined as participation in none of biennial general physical check-ups or twice-a-life cancer screening, both of which are provided biennially for free by the National Health Insurance Service in Korea, or cancer screening with partial or full out-of-pocket payment. c Current smokers were defined as respondents who answered “everyday” or “sometime” to a survey question asking, “Do you currently smoke cigarettes?” d Respondents were classified as physically inactive if they answered that they never engaged in active walking in a given week. e Respondents were classified as having 1 of 4 behaviors if they had any of the following 4 unhealthy lifestyle behaviors: current smoking, moderate to heavy drinking, being physically inactive, and no regular physical check-ups. f A dummy indicator for overweight or obesity was defined as body mass index, weight in kilograms divided by height in meters squared, of 25 or larger. g Other individual characteristics controlled for as covariates in all estimations were the following: age, in a linear measure of the median in categories by 5-year span; a linear measure of total annual household income in Korean Won (KW); and Charlson Comorbidity Index as a linear variable. Additionally, dummy indicators are residential area (living in metropolitan areas or medium to small cities, with living in rural areas as the reference); marital status (married, with not being married as the reference); education level (college or more or high school graduate, with less than college as the reference); number of children; employment status (self-employed, full-time employees, or part-time employees, with being unemployed as the reference); and survey years 2010 and 2011, with 2009 as the reference.

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Group

Table 6 Quantile instrumental variable regression results for those with at least one unhealthy behavior (moderate to heavy drinking, current smoking, being physically inactive, or lack of regular physical check-ups) by gender and age group Group

Quantiles in the conditional distribution of log of pharmaceutical expendituref Coefficient for overweight/obesitye (standard error) 5th

10th

25th

50th

75th

90th

95th

0.7102 1.5930 0.2456 2.1695

(1.0487) (1.9699) (1.4887) (2.8965)

1.2301 1.6099 0.6424 2.0845

(0.7670) (1.1167) (1.1240) (2.1403)

1.4295 (1.9644) 2.4540* (1.2948) 1.0303 (0.8370) 1.7507 (1.3488)

1.7331*** (0.5891) 2.5069* (1.4458) 1.6851** (0.6748) 1.9548 (1.5991)

1.5382* (0.8587) 2.2002 (2.4166) 1.4307* (0.7397) 1.4488 (1.7413)

1.5611* (0.8214) 1.9372 (1.8993) 1.3797 (0.9262) 1.2854 (2.0312)

0.1880 2.2202 0.7265 4.2205

(0.8720) (3.2268) (1.1556) (5.0930)

1.1184 1.4164 0.4586 3.3132

(0.7965) (2.6614) (1.2571) (4.6612)

1.5517 0.4373 0.9214 2.0462

1.9988** (0.5853) 1.3343* (0.7937) 1.1114** (0.5409) 1.6410 (1.5364)

1.6167* (0.7831) 1.3823* (0.8747) 1.0556* (0.5796) 1.4581 (1.8281)

1.7389 (1.4527) 1.3734* (0.7903) 1.0008 (0.8418) 1.5501 (2.2818)

(1.2254) (2.2176) (1.0703) (2.4841)

0.2844 (1.5749) 1.4017 (9.5406) 0.6476 (0.9353) 0.5415 (2.8002)

0.1631 (1.2395) 0.8671 (8.4419) 0.4776 (1.5529) 0.7180 (1.9255)

0.0146 (0.8413) 0.7363 (24.9871) 0.3184 (1.0838) 0.44236 (2.3863)

0.8083 (1.8568) 0.2752 (18.6027) 0.0935 (1.1626) 0.5890 (0.8429)

0.5870 (2.4458) 0.7777 (16.3930) 1.2459 (1.2719) 0.7540 (0.8983)

0.2802 (1.9001) 0.7739 (16.8039) 1.3393 (1.3799) 0.7541 (0.8991)

0.4985 (2.9306) 0.4407 (2.3624) 2.0871 (3.9547) 0.5088 (2.8029)

0.2964 (1.6199) 0.3553 (1.6963) 1.9361 (2.1556) 0.3165 (2.1818)

1.0314 (3.3027) 0.1754 (1.4826) 2.2689 (1.9813) 0.3692 (2.0745)

2.2001 (1.1528) 0.0238 (1.8699) 2.1733 (2.7329) 0.4900 (1.7046)

2.1247 0.2775 1.8279 0.6546

2.1004 0.3237 1.6735 0.7288

(2.2664) (3.5914) (2.4153) (2.0462)

(1.5062) (4.8298) (2.7936) (2.5104)

11

*P ! 0.1, **P ! 0.05, ***P ! 0.01. a Moderate to heavy drinking was defined as having 5 0.355-L cans of beer for men (with total alcohol content of 62.5 g) or 3 cans for women (with total alcohol content of 37.5 g) at one time at least once in a given month. b No regular physical check-ups was defined as participation in none of biennial general physical check-ups or twice-a-life cancer screening, both of which are provided biennially for free by the National Health Insurance Service in Korea, or cancer screening with partial or full out-of-pocket payment. c Current smokers were defined as respondents who answered “everyday” or “sometime” to a survey question asking, “Do you currently smoke cigarettes?” d Respondents were classified as physically inactive if they answered that they never engaged in active walking in a given week. e A dummy indicator for overweight or obesity was defined as body mass index, weight in kilograms divided by height in meters squared, of 25 or larger. f Other individual characteristics controlled for as covariates in all estimations were the following: age, in a linear measure of the median in categories by 5-year span; a linear measure of total annual household income in Korean Won (KW); Charlson Comorbidity Index as a linear variable. Additionally, dummy indicators are residential area (living in metropolitan areas or medium-to-small cities, with living in rural areas as the reference); marital status (married, with not being married as the reference); education level (college or more or high school graduate, with less than college as the reference); number of children; employment status (self-employed, full-time employees, or part-time employees, with being unemployed as the reference); and survey years 2010 and 2011, with 2009 as the reference.

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

Moderate to heavy drinkinga Male (N ¼ 2168) 0.0572 (0.8389) Female (N ¼ 2838) 1.3841 (3.1316) Adults (N ¼ 4299) 0.3001 (1.1593) Elderly (N ¼ 707) 2.3039 (3.2029) No regular physical check-upb Male (N ¼ 2398) 0.3629 (0.7794) Female (N ¼ 3767) 2.5038 (2.6207) Adults (N ¼ 4887) 0.8304 (0.9518) Elderly (N ¼ 1278) 4.5971 (5.3634) Current smokingc Male (N ¼ 1103) 0.2726 (1.8944) Female (N ¼ 94) 1.9176 (10.0167) Adults (N ¼ 952) 0.7355 (1.2589) Elderly (N ¼ 425) 0.6742 (3.6718) Physically inactived Male (N ¼ 947) 0.2088 (3.6696) Female (N ¼ 947) 0.4930 (3.5367) Adults (N ¼ 1099) 2.1613 (7.8213) Elderly (N ¼ 416) 0.5284 (3.3738)

12

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

The current study is consistent with those studies by using a national survey that extracted information on health service use and expenditure from claims to NHIS in Korea. Therefore, potential measurement issues with regard to selfreported health service use are of less concern in the current study. The current study did not distinguish out-ofpocket spending from insurers’ coverage in total pharmaceutical expenditures. Actual individual spending for health services use is out-of-pocket expenses. However, total expenditure, including both out-of-pocket and insurer’s, is important, particularly in the context of national health insurance systems such as that of Korea, because most of the expenditure is publicly financed (either via health insurance premiums or government subsidies). People cannot opt out of national health insurance, and their premiums are determined based on income or wealth rather than medical needs. Therefore, excess pharmaceutical expenditure due to modifiable lifestyle risk behaviors such as OW/OB can impose a negative externality not only to the overweight or obese people themselves but also their normal weight or underweight counterparts. However, it also should be noted that the current study focused on the heterogeneous impact of OW/OB on pharmaceutical expenditure among users of pharmaceuticals, and any implications from the current study are thus limited to those users. Numerous studies have estimated the association of OW/OB with health care expenditure, but only a handful of them have attempted to establish causality in the association.9,17,41 The current study controlled for the unobserved characteristics that hamper establishment of the causality by using an instrumental variable approach with biological family members’ body mass status as a just-identifying instrument for the individual’s body mass status. Although there is a piece of conventional wisdom that shared family environment affects individual health, importantly for OW/OB status and pharmaceutical use, the literature did not find for the effect of the shared environment on body mass status.17,42,43 However, the validity of the IV for the current study through formal specification tests remains unverified due to lack of overidentifying IVs, although a vast literature supports the IV of choice in the current study.29–35 The current study drew height and weight information from self-reports, which may attenuate the impact of OW/OB on pharmaceutical

expenditure. Other studies suggest that central adiposity or direct measurement of body fat can be a better predictor of OW/OB that predicts health service use or expenditure.41 Such information is lacking in the current data and should be addressed in further studies. Both pharmaceutical expenditure and body mass status were measured at 3 points over 3 years, which might underestimate the actual impact of OW/OB on pharmaceutical expenditure. Future studies are needed to address the cumulative impact of the accumulation of various lifestyle risk factors and OW/OB on pharmaceutical cost over the long term, beyond the 3 year span of the current study. Heterogeneity in the impact of OW/OB on pharmaceutical expenditure across various aspects, i.e., by disease characteristics, also needs to be further explored in future studies in order to gather more specific evidence for public health interventions. Most studies of the association of OW/OB with health expenditure that include pharmaceutical use are based in Europe, the USA, or Australia, but there seems to be only one study in Asia.44 The results of the current study add to global evidence on the impact of OW/OB on pharmaceutical expenditure, affirming OW/OB as a major public health concern in Korea as well. Given that the National Health Insurance is the backbone of the Korean health care system, the incremental increase in pharmaceutical expenditure stemming from OW/OB status might have some negative externality for the whole population in Korea.

Conclusion The association of overweight/obesity with pharmaceutical expenditure is considerable. The current study confirms that the association of OW/OB with modifiable unhealthy behaviors on pharmaceutical expenditure is larger than with OW/OB, alone. These results justify in-depth and comprehensive public efforts to control for those modifiable lifestyle-related unhealthy behaviors. Acknowledgment Research support from the Korea National Research Foundation (2014R1A1A3A04049984) is gratefully acknowledged. The content is solely the responsibility of the authors and does not necessarily represent the official view of the Korea National Research Foundation. The Korea

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14

National Research Foundation had no involvement in preparation and submission of this manuscript. All authors have no conflict of interest to disclose.

17.

References

18.

1. World Health Organization. http://www.who.int/ gho/ncd/risk_factors/obesity_text/en/. 2. Kim TH, Han E. The impact of body mass on job quality. Econ Hum Biol 2015;17:75–85. 3. Andreyeva T, Sturm R, Ringel JS. Moderate and severe obesity have large differences in health care costs. Obes Res 2004;12:1936–1943. 4. Sturm R. The effects of obesity, smoking, and drinking on medical problems and costs. Health Aff (Millwood) 2002;21:245–253. 5. Daviglus ML. Health care costs in old age are related to overweight and obesity earlier in life. Health Aff (Millwood) 2005;24. W5R97–W5R100. 6. Goetzel RZ, Pei X, Tabrizi MJ, et al. Ten modifiable health risk factors are linked to more than one-fifth of employer-employee health care spending. Health Aff (Millwood) 2012 Nov;31:2474–2484. 7. Withrow D, Alter DA. The economic burden of obesity worldwide: a systematic review of the direct costs of obesity. Obes Rev 2011;12:131–141. 8. Tsai AG, Williamson DF, Glick HA. Direct medical cost of overweight and obesity in the United States: a quantitative systematic review. Obes Rev 2011;12: 1–18. 9. Bell JF, Zimmerman FJ, Arterburn DE, Maciejewski ML. Healthcare expenditures of overweight and obese males and females in the medical expenditures panel survey by age cohort. Obes (Silver Spring) 2011;19:228–232. 10. Finkelstein EA, Trogodan JG, Cohen JW, Dietz W. Annual medical spending attributable to obesity: payer- and service-specific estimates. Health Aff 2008;28:w822–w831. 11. Kang JH, Jeong BG, Cho YG, Song HR, Kim KA. Socioeconomic costs of overweight and obesity in Korean adults. J Korean Med Sci 2011;26:1533–1540. 12. Yoon Nan-He, Kwon Soonman. Impact of obesity on health care utilization and expenditure. Korean J Health Econ Pol 2013;19:61–80. 13. Balkrishnan R. The importance of medication adherence in improving chronic-disease related outcomes: what we know and what we need to further know. Med Care 2005;43:517–520. 14. Ministry of Health and Welfare. 2011. White Paper on Health and Welfare; 2012. 15. Cawley JH, Meyerhoefer C. The medical care costs of obesity: an instrumental variables approach. J Health Econ 2012;31:219–230. 16. Shen C, Sambamoorthi U, Rust G. Co-occurring mental illness and health care utilization and

19.

20.

21.

22.

23. 24.

25.

26.

27.

28.

29.

30.

31.

32.

13

expenditures in adults with obesity and chronic physical illness. Dis Manag 2008;11:153–160. Abdullah A, Peeters A, Court d, Stoelwinder J. The magnitude of association between overweight and obesity and the risk of diabetes: a meta-analysis of prospective cohort studies. Diabetes Res Clin Pract 2010;89:309–319. Wang F, McDonald T, Champagne LJ, Edington DW. Relationship of body mass index and physical activity to health care costs among employees. J Occup Environ Med 2004;46:428–436. Wei M, Kampert JB, Barlow CE, et al. Relationship between low cardiorespiratory fitness and mortality in normal-weight, overweight, and obese men. J Am Med Assoc 1999;282:1547–1553. Brown WJ, Hockey R, Dobson AJ. Physical activity, Body Mass Index and health care costs in mid-age Australian women. Aust N Z J Public Health 2008; 32:150–155. Wang F, McDonald T, Reffitt B, Edington DW. BMI, physical activity, and health care utilization/ costs among Medicare retirees. Obes Res 2005;13: 1450–1457. Alter DA, Wijeysundera HC, Franklin B, et al. Obesity, lifestyle risk-factors, and health service outcomes among healthy middle-aged adults in Canada. BMC Health Serv Res 2012;12:238. Greene WH. Econometric Analysis. 4th ed. Prentice Hall, Inc; 2000. National Health Insurance Service, Korea Institute for Health & Social Affairs. User’s Guide for Korea Medical Panel Survey 2008-2011; 2014. World Health Organization. Overweight and obesity Factsheet N 311. http://www.who.int/mediacentre/ factsheets/fs311/en/. World Health Organization Regional Office for the Western Pacific, the International Association for the Study of Obesity (IASO) and the International Obesity Task Force(IOTF). The AsiaPacificperspective: Redefining Obesity and its Treatment; 2000. Charlson M, Szatrowski TP, Peterson J, Gold J. Validation of a combined comorbidity index. J Clin Epiedemiol 1994;47:1245–1251. Manning WG, Duan N, Rogers WH. Monte Carlo evidence on the choice between sample selection and two-part models. J Econom 1987;35:59–82. Woodridge JM. Introductory Econometrics: A Modern Approach. 4th ed. South-Western College Publishing; 2012. Stock JH, Wright JH, Yogo M. A survey of weak instruments and weak identification in generalized method of moments. J Bus Econ Stat 2002;20:518– 529. Hewitt JK. The genetics of obesity: what have genetic studies told us about the environment? Behav Gen 1997;27:353–358. Grilo CM, Pogue-Geile MF. The nature of environmental influences on weight and obesity: a behavioral

14

33.

34.

35.

36. 37.

38. 39.

Kim et al. / Research in Social and Administrative Pharmacy j (2016) 1–14 genetic analysis. Psychopharmacol Bull 1991;110: 520–537. Price RA, Gottesman II. Body fat in identical twins reared apart: roles for genes and environment. Behav Gen 1991;21:1–7. Maes HHM, Neale MC, Eaves LJ. Genetic and environmental factors in relative body weight and human adiposity. Behav Gen 1997;27:325–351. Wardle J, Carnell S, Haworth CMA, Plomin R. Evidence for a strong genetic influence on childhood adiposity despite the force of the obesogenic environment. Am J Clin Nutr 2008;87:398–404. Cawley J. The impact of obesity on wages. J Hum Res 2004;39:451–474. Kline B, Tobias JL. The wages of BMI: Bayesian analysis of a skewed treatment–response model with nonparametric endogeneity. J Appl Econ 2008; 23:767–793. Frolich M, Melly B. Estimation of quantile treatment effects with Stata. Stata J 2010;10:423–457. Wee CC, Phillips RS, Legedza AR, Davis RB, Soukup JR, Colditz GA. Health care expenditures

40.

41.

42.

43.

44.

associated with overweight and obesity among US adults: Importance of age and race. Am J Public Health 2005;95:159–165. Buchmueller TC, Johar M. Obesity and health expenditures: evidence from Australia. Econ Hum Biol 2015;17C:42–58. Ellis RP, Fiebig DG, Johar M, Jones G, Savage E. Explaining health care expenditure variation: largesample evidence using linked survey and health administrative data. Health Econ 2013;22:1093–1110. Nelson MC, Gordon-Larsen P, North KE, Adair LS. Body mass index gain, fast food and physical activity: effects of shared environment over time. Obesity 2006;14:701–709. Haberstick BC, Lessem JM, McQueen MB, et al. Stable genes and changing environments: body mass index across adolescence and young adulthood. Behav Genet 2010;40:495–504. Chu NF, Wang SC, Chang HY, Wu DM. Medical services utilization and expenditure of obesityrelated disorders in Taiwanese adults. Value Health 2010;13:829–836.