This study was a retrospective, cross-sectional analysis of responses obtained from adults who participated in the National Health and Wellness Survey (NHWS) in 2016 and 2020. The NHWS is a self-administered internet-based survey conducted annually among a nationally representative sample of adults in the USA based on age, sex, and race. Data from the 2016 and 2020 NHWS were selected for analysis because respondents provided information about bipolar disorder subtype (e.g., BD-I vs bipolar II disorder) in those years. The survey includes questions on patient-reported health outcomes, HCRU, and medical conditions.

This study was conducted in accordance with the ethical principles derived from the Declaration of Helsinki. All study procedures were reviewed and approved by the Pearl Institutional Review Board (Indianapolis, IN); an exempt status was obtained for this study per Federal Regulation 45 CFR Subtitle A, Subchapter A, Part 46, Subpart A, §46.104(d)(2), which governs deidentified survey data [19].

Respondents aged 18 to 64 years residing in the USA and completing either the 2016 or 2020 NHWS were eligible. Those who self-reported a physician diagnosis of BD-I were included in the BD-I cohort, and those who did not were included in the unmatched sample cohort (see Supplementary Fig. S1 for inclusion criteria). Respondents with BD-I were matched to controls using 1:2 greedy propensity-score matching based on age, sex, race, region, alcohol use, smoking status, exercise frequency, and modified Charlson Comorbidity Index (CCI). Matching was stratified by year of NHWS completion (2016 or 2020), age group (18–30, 31–45, or 46–64 years), and BMI (underweight/normal weight, < 25 kg/m2; overweight, 25 to < 30 kg/m2; or obese, ≥ 30 kg/m2). BMI was calculated from respondents’ self-reported height and weight. A total of 47 respondents with BD-I had missing variables and were excluded from propensity-score matching.

After matching was conducted, respondents were categorized and outcomes were assessed according to the BMI categories underweight/normal weight (BMI < 25 kg/m2), overweight (BMI 25 to < 30 kg/m2), and obese (BMI ≥ 30 kg/m2). Clinical outcomes included the prevalences of self-reported obesity-related medical comorbidities, including the presence of high blood pressure, hypercholesterolemia, type 2 diabetes, cardiovascular events (ministroke, stroke, heart attack, or congestive heart failure), asthma, liver disease, cancer, osteoarthritis, and sleep apnea. Healthcare resource utilization, including hospitalizations, emergency department (ED) visits, and visits to healthcare professionals in the past 6 months, was also assessed.

Humanistic effects (i.e., health-related quality of life [HRQoL]) were evaluated using the 36-item Short Form Version 2 (SF-36v2) and EuroQol EQ-5D health surveys. Scores on the Mental Component Summary and Physical Component Summary of the SF-36v2 were calculated using a norm-based scoring algorithm with linear T-score transformation and then scaled to 100. On both the SF-36v2 and the EQ-5D, lower scores represent worse HRQoL. For the SF-36v2, a clinically meaningful difference in Mental Component Summary score is 3 points [20], and estimates of the clinically meaningful difference for the EQ-5D range from 0.03 to 0.52 points among different patient populations [21,22,23].

From an economic standpoint, work productivity was evaluated via the Work Productivity and Activity Impairment (WPAI) questionnaire, which is composed of four assessments: absenteeism (missed work in the past 7 days), presenteeism (lost productivity at work in the past 7 days), overall work impairment (absenteeism and presenteeism in the past 7 days combined), and activity impairment (health-related impairment in daily activities in the past 7 days). Only respondents who reported being employed at the time of the survey provided responses to questions about absenteeism, presenteeism, and overall work impairment; all respondents answered questions about activity impairment. Estimates of total indirect costs, including indirect costs associated with absenteeism and presenteeism, were derived from 2019 hourly wage data published by the US Bureau of Labor Statistics [24]. Direct medical costs were estimated using data from the 2018 Medical Expenditure Panel Survey and based on national cost averages for each type of resource utilized [25].

Multivariable regression models were used to assess differences in outcomes between subgroups. In addition, NHWS year (2016 vs 2020) was included as a covariate in regression models to control for the potential effect of the COVID-19 pandemic on outcomes. Frequencies and percentages were reported for categorical outcomes, whereas SDs/SEMs were reported for continuous outcomes. Study outcomes were summarized descriptively, with no reported formal hypothesis testing between subgroups.

A sensitivity analysis was conducted with obesity further stratified into the following subgroups: underweight/normal weight, overweight, obese class 1 (BMI 30 to < 35 kg/m2), obese class 2 (BMI 35 to < 40 kg/m2), and obese class 3 (BMI ≥ 40 kg/m2).