The obesity epidemic has gained increasing recognition as a major public health concern in the USA. Despite being the focus of numerous public health efforts, prevalence of overweight and obesity has increased steadily^{(Reference Ruhm1,Reference Sturm and Hattori2)} . Recent estimates indicate that more than two-thirds of the US adult population is overweight^{(Reference Ogden, Carroll and Kit3,Reference Cecchini4)} with up to 40 % classified as obese^{(Reference Hales, Fryar and Carroll5)}. The societal burden of obesity in terms of morbidity, mortality and healthcare expenditures is extensive. In the USA, 280 000 to 325 000 deaths per year are attributable to obesity^{(Reference Allison, Fontaine and Manson6)}, making it the second leading cause of preventable death behind tobacco use^{(Reference Frisco, Van Hook and Hummer7)}. Furthermore, estimates of total US medical costs attributed to obesity range from $147 billion in 2008^{(Reference Finkelstein, Trogdon and Cohen8)} to $342 billion in 2013^{(Reference Biener, Cawley and Meyerhoefer9)}.

Elevated BMI is an established risk factor for a series of chronic conditions including hyperlipidemia, diabetes mellitus and hypertension^{(Reference Sharma, Lee and Youssef10)}. These are concordant conditions^{(Reference Piette and Kerr11)}, meaning they share similar pathophysiologic risk profiles or management processes. On the other hand, patients with elevated BMI may also suffer from discordant comorbidities^{(Reference Piette and Kerr11)}, or conditions not directly related to the pathophysiology of high BMI measures, such as mental-health related illness^{(Reference Dickerson, Brown and Kreyenbuhl12)}.

In addition to increasing risk of multiple chronic conditions, elevated BMI is independently associated with reduced health-related quality of life (HRQoL). Measures of HRQoL have become widely used in clinical studies and routine outcome assessment to provide information on the effectiveness of healthcare intervention from the patient’s perspective^{(Reference Asadi-Lari, Tamburini and Gray13)}. Improving BMI measure could further improve well-being, reduce bodily pain, depression, or anxiety and increase physical activity and social interaction^{(Reference Ackermann, Edelstein and Narayan14)}.

The five-item instrument EuroQol-5 Dimensions (EQ-5D) and Short Form-6 Dimensions (SF-6D), which is derived directly from the 12-Item Short Form Health Survey (SF-12), are the most common preference-based HRQoL or health utility measures. They are increasingly used in health economic evaluations to aid resource allocation decisions^{(Reference Sach, Barton and Doherty15)}. The necessity for such health utility measures has been demonstrated as both the US Public Health Service Panel on Cost-Effectiveness in Health and Medicine and the UK National Institute for Health and Clinical Excellence (NICE) prefer health utility measures to be used when assessing cost-effectiveness of new and current health interventions^{(Reference Sach, Barton and Doherty15)}. The SF-6D outcome is an important preference-based health utility measure used to estimate quality-adjusted life years, a universal metric to measure health outcomes for health technology assessment. An increase in BMI is consistently associated with decrements in HRQoL outcomes across multiple measures including the EQ-5D, SF-12 and SF-6D^{(Reference Sach, Barton and Doherty15,Reference Fontaine and Barofsky16)} .

Previous studies have adjusted for comorbidities when evaluating the association between overweight and HRQoL^{(Reference Jia and Lubetkin17,Reference Hunger, Thorand and Schunk18)} . Alternatively, the effect of elevated BMI on HRQoL has been investigated, independently of comorbidities by including only otherwise healthy patients^{(Reference Korhonen, Seppala and Jarvenpaa19)}. Additional studies have investigated association of HRQoL in patients with obesity and select conditions, for example, type 2 diabetes, and inflammatory conditions^{(Reference Slagter, van Vliet-Ostaptchouk and van Beek20)}. Studies of multimorbidity often focused on the number of conditions, rather than interactions between different types of conditions^{(Reference Hunger, Thorand and Schunk18)}. In the current study, we examined the impact of elevated BMI on HRQoL outcome considering both the summative impact of multimorbidity, while separately assessing the type of chronic comorbid conditions (concordant v. discordant) in a nationally representative US sample.

Methods

Data source

The current study analysed pooled data from the Medical Expenditure Panel Survey (MEPS) Full Year Consolidated Data and Medical Condition Files from 2013 to 2015⁽²¹⁾. The MEPS, conducted by the Agency for Healthcare Research and Quality (AHRQ), is a nationally representative survey of the US civilian non-institutionalised population⁽²¹⁾. The MEPS collects individual-level data regarding healthcare expenditure, medical conditions, use of medical care services, health status, quality of life and demographic characteristics. Medical conditions were recorded verbatim by interviewers and subsequently converted to International Classification of Diseases, Ninth Revision (ICD-9) and Clinical Classification System (CCS) codes by professional coders⁽²¹⁾. MEPS data files are publicly available and de-identified; thus, this study was exempt from Institutional Review Board review.

Weight category

BMI in MEPS data files is calculated using height and weight as reported by household respondents. The study population was classified into six weight categories as defined by the WHO: underweight (BMI < 18·5 kg/m²), normal weight (BMI between 18·5 and 24·9 kg/m²), overweight (BMI between 25 and 29·9 kg/m²), Class-I obesity (BMI between 30·0 and 34·9 kg/m²), Class-II obesity (BMI between 35·0 and 39·9 kg/m²) and Class-III obesity (BMI ≥ 40 kg/m²)⁽²²⁾.

Chronic conditions

Chronic conditions of interest were categorised as concordant (elevated BMI-related) or discordant (elevated BMI-unrelated) using the US Department of Health and Human Services (HHS) conceptual model for defining and identifying chronic conditions^{(Reference Goodman, Posner and Huang23)}. Concordant conditions included the following: hypertension, hyperlipidemia, diabetes, congestive heart failure, coronary artery disease, cardiac arrhythmia, stroke, chronic kidney disease, arthritis, osteoporosis, asthma and chronic obstructive pulmonary disease. Conversely, discordant conditions of obesity included autism, cancer, dementia, depression, hepatitis, HIV, schizophrenia and substance abuse disorder. The full list of concordant and discordant comorbid conditions related to obesity is reported in Table 1.

Table 1

Concordant and discordant conditions

Health-related quality of life measures

The HRQoL SF-12 instrument consists of twelve survey questions (items) measuring eight health dimensions (general health, physical functioning, role limitations – physical, and bodily pain measure physical health component and vitality, social functioning, role limitations – emotional, and mental health measure mental health component)^{(Reference Brazier and Roberts24)}. Responses are summarised using two aggregate scores: the physical component summary and mental component summary^{(Reference Brazier and Roberts24)}. HRQoL SF-6D measures were directly derived from responses of the twelve items of the SF-12 used in the MEPS datasets. The SF-6D condenses the eight health dimensions in the SF-12 into six: (1) physical functioning; (2) role limitations; (3) social functioning; (4) pain; (5) mental health and (6) vitality^{(Reference Ackermann, Edelstein and Narayan14)}, hence called SF-6D. The SF-6D instrument has proven to be sensitive to the expected loss of HRQoL associated with obesity^{(Reference Sach, Barton and Doherty15)}. SF-6D scores range from 0·30 to 1·00, with 0·30 indicating worst possible health condition and 1·00 for perfect health^{(Reference Brazier and Roberts24,Reference Marra, Woolcott and Kopec25)} . To determine the clinical relevance of reduced HRQoL in terms of SF-6D measures, we compared the observed decline in SF-6D scores in our study to published estimates of minimal clinically important difference (MCID) for SF-6D scores. Prior studies reported mean estimates for SF-6D MCID ranging from 0·027 to 0·041^{(Reference Walters and Brazier26–Reference Luo, Johnson and Coons28)}.

Statistical analysis

Generalised linear models with Gaussian distribution and identity link were used to compare SF-6D scores across BMI categories, type (concordant v. discordant) of comorbidities and numbers of each type of comorbid conditions. All analyses controlled for baseline characteristics including age, race, gender, education, income, insurance coverage, marital status and tobacco use. The impact of concordant and discordant comorbid conditions and the incremental burden of each type of comorbidities on SF-6D scores was examined in the population with elevated BMI (BMI ≥ 25 kg/m²). Separate generalised linear models were used to evaluate the impact of different elevated BMI categories, type (concordant v. discordant) of comorbidities and the incremental burden of each type of comorbidities on the SF-6D HRQoL outcome. P-value < 0·05 were considered statistically significant for a single hypothesis testing. When combining hypotheses for multiple tests, P-values were adjusted using the conservative Bonferroni adjustment^{(Reference Dunn29)}. All analyses were performed using the R survey package version 3.36^{(Reference Lumley30)}, accounting for sampling strata and survey weights used in the MEPS survey design to control for survey non-response and adjustment to population control totals.

Results

Baseline characteristics

A sample of 58 960 subjects aged 18 years or older were identified from the pooled cross-sectional MEPS data between 2013 and 2015. Of which, 1·7 %, 32·9 %, 34·0 % and 31·4 % were classified as underweight, normal weight, overweight and obese, respectively. Among subjects with obesity, most had Class-I obesity (58·5 %), followed by those with Class-II and Class-III obesity (25·2 % and 16·3 %, respectively). Males were more likely to be overweight or Class-I obese (56·1 % and 51·2 %, respectively). Females were more likely to be Class-II or Class-III obese (53·8 % and 65 %, respectively), while also having a higher proportion of underweight subjects (74·8 %). In the youngest age group (18 to 24 years of age), the proportion of underweight respondents was disproportionately high (26·2 %), while the proportion of subjects with overweight and obesity was disproportionately low (8·1 % and 6·3 %, respectively). The highest rate of elevated BMI was seen in those 45 to 64 years of age. Overweight and all classes of obesity were more prevalent in black and Hispanic populations as well as in those with lower income. Hypertension, hyperlipidemia, diabetes mellitus and arthritis were the most common conditions observed in respondents with obesity, followed by respiratory conditions. Prevalence of all concordant conditions was higher in the overweight category compared to the normal weight category apart from osteoporosis. A similar prevalence trend was observed with discordant conditions, except for those with dementia or HIV (Table 2).

Table 2

Baseline characteristics

COPD, chronic obstructive pulmonary disease.

Weight categories (kg/m²): underweight (BMI < 18 5), normal weight (18 5 ≤ BMI < 25), overweight (25 ≤ BMI < 30), Class-I obesity (30 0 ≤ BMI < 35), Class-II obesity (35 0 ≤ BMI < 40) and Class-III obesity (BMI ≥ 40).

Health-related quality of life outcomes

Compared to subjects with normal weight, SF-6D scores were significantly decreased across all obesity classes, with the largest reduction in Class-III obesity (0·033; P < 0·001) (Table 3). Overweight subjects had reduced SF-6D scores compared to those with normal BMI; however, the reduction was not found statistically significant (P = 1·000). Similarly, there was no-significant reduction in SF-6D scores observed in underweight subjects compared to those with normal BMI (P = 0·830).

Table 3

SF-6D disutility by BMI category (n 58 960)

SF-6D, Short Form-6 Dimensions.

* P-values were adjusted for multiple tests in combining hypotheses using the conservative Bonferroni adjustment.

† Based on the minimal clinically important difference for SF-6D ranging from 0·027 to 0·041.

Multivariate linear regression model for SF-6D disutility adjusted for all comorbid conditions and baseline characteristics including age, race, gender, education, income, insurance coverage, marital status and tobacco use in the overall sample.

Further comparing to the preceding weight category, that is, overweight v. Class-I obesity, Class-I v. Class-II obesity and Class-I v. Class-II obesity, the SF-6D disutilities were all significant (P-values < 0 05).

Chronic conditions

Having chronic conditions resulted in further significant reduction in SF-6D scores, apart from hyperlipidemia, chronic kidney disease, osteoporosis, autism, cancer and HIV. The largest reductions in SF-6D scores were observed in subjects with depression (0·110; P < 0·001), dementia (0·090; P < 0·001) and arthritis (0·064; P < 0·001) (Table 4). Having one or more concordant or discordant conditions further reduced SF-6D scores from 0·031 to 0·148 or from 0·080 to 0·212, respectively. In addition, when comparing incremental burden of having discordant conditions v. concordant conditions, the SF-6D scores were significantly lower: one discordant comorbidity v. one concordant comorbidity (0·080 v. 0·031, P < 0·001), two discordant comorbidity v. two concordant comorbidity (0·125 v. 0·057, P < 0·001), three discordant comorbidity v. three concordant comorbidity (0·157 v. 0·077, P < 0·001) and four discordant comorbidity v. four concordant comorbidity (0·212 v. 0·108, P < 0·001) (Table 5). Associations between SF-6D scores and incremental burden of type of chronic conditions are reported in Table 5.

Table 4

SF-6D disutility by chronic comorbid conditions in elevated-BMI sample (n 39 727)

SF-6D, Short Form-6 Dimensions; COPD, chronic obstructive pulmonary disease.

* P-values were adjusted for multiple tests in combining hypotheses using the conservative Bonferroni adjustment.

† Based on the minimal clinically important difference for SF-6D ranging from 0·027 to 0·041.

Multivariate linear regression model for SF-6D disutility adjusted for all comorbid conditions and baseline characteristics including age, race, gender, education, income, insurance coverage, marital status and tobacco use in the elevated-BMI sample.

Table 5

SF-6D disutility by incremental burden of concordant and discordant conditions (n 39 727)

SF-6D, Short Form-6 Dimensions.

* P-values were adjusted for multiple tests in combining hypotheses using the conservative Bonferroni adjustment.

† Based on the minimal clinically important difference for SF-6D ranging from 0·027 to 0·041.

Multivariate linear regression model for SF-6D disutility adjusted for baseline characteristics including age, race, gender, education, income, insurance coverage, marital status and tobacco use in the elevated-BMI sample.