Association between physical activity and insulin resistance using the homeostatic model assessment for insulin resistance independent of waist circumference

Only a few studies have evaluated the relationship between physical activity (PA) and Homeostatic model assessment for insulin resistance (HOMA-IR). Therefore, we aimed to analyze the association between HOMA-IR and PA. We included 280,194 Korean without diabetes who underwent health examinations. The short form of the International Physical Activity Questionnaire was completed. PA level was divided into sedentary, mild PA, and health-enhancing PA (HEPA). The HOMA-IR levels were calculated. Confounding factors including waist circumference were adjusted. The median follow-up duration was 4.13 years. A significant inverse relationship was observed between PA level and HOMA-IR (p < 0.001). Compared with the sedentary group, HOMA-IR was lower in the HEPA group (p < 0.001), even when HEPA group decreased PA level over time (p < 0.001). Mild PA (p < 0.001) or HEPA showed a lower risk of HOMA-IR progression (p < 0.001). Increasing PA or maintaining HEPA was significantly associated with a lower HOMA-IR (p < 0.001), HOMA-IR improvement (p < 0.001), and a lower risk of HOMA-IR progression (p < 0.001). Our findings support the inverse relationship between PA and HOMA-IR in a population without diabetes. PA might improve IR and prevent its progression among populations without diabetes, independent of the waist circumference.

Sex-stratified associations of PA with HOMA-IR according to the changes in PA level. Table 3 shows the sex-stratified associations of PA with HOMA-IR according to the changes in PA levels. The participants were divided into four groups based on the changes in the PA level: sedentary and mild PA level (SM) to SM (reference), HEPA to SM, SM to HEPA, and HEPA to HEPA. Compared with the SM to SM group, the other three groups showed significantly decreased HOMA-IR levels in the multivariable model (HEPA to SM group = estimate: 0.94, CI 0.93-0.94, p < 0.001; SM to HEPA group = estimate: 0.93, CI 0.93-0.94, p < 0.001; HEPA to HEPA group = estimate: 0.86, CI 0.85-0.87, p < 0.001).
Associations of PA with the change of HOMA-IR level. As shown in Table 4, the participants were divided into two groups; baseline HOMA-IR ≥ 2.2 group (n = 38,950) and baseline HOMA-IR < 2.2 group Table 1. Baseline characteristics. Numbers in the table are mean ± standard deviation, median (interquartile range), or (percentages). High alcohol intake defined as > 30 g/day for men and > 20 g/day for women; higher education defined as college graduate or higher. IPQ, International Physical Activity Questionnaire; BMI, body mass index; SBP, systolic blood pressure; HOMA-IR, homeostasis model assessment of insulin resistance.   Associations of change in the PA with change in HOMA-IR. Table 5 shows the associations between changes in PA level and changes in HOMA-IR level. The participants were divided into two groups; baseline HOMA-IR ≥ 2.2 group (n = 38,950) and baseline HOMA-IR < 2.2 group (n = 241,244). In each groups, the association between changes in PA level (SM to SM, HEPA to SM, SM to HEPA, HEPA to HEPA) and changes in HOMA-IR level (improvement, progression) were investigated.

Discussion
Our results showed that there was a significant inverse relationship between PA level and HOMA-IR, a marker of IR. Second, compared with the sedentary group, HOMA-IR was lower even if the PA level in the HEPA group was decreased over time. Third, mild PA and HEPA showed a lower risk of HOMA-IR progression. Fourth, increasing the PA level or maintaining HEPA levels was significantly associated with lower HOMA-IR level. Lastly, the increasing PA or maintaining HEPA level was associated with HOMA-IR improvement and a lower risk of HOMA-IR progression. As a well-known fact, type 2 DM develops as a result of IR and is associated with metabolic abnormalities 4 . In addition, diabetes medications including metformin, glimepiride, and SGLT2 inhibitors can affect the HOMA-IR levels 21,22 . Previous studies that assessed the relationship between PA and HOMA-IR were limited due to their Table 4. Associations of PA with the change of HOMA-IR level. Sedentary, less than 600MET-minutes per week of physical activity; Mild physical activity, 600 MET-minutes per week; HEPA, Health-enhancing physical activity : 3000 MET-minutes per week; CI, confidence interval; HOMA-IR, homeostasis model assessment-estimated IR; HR, hazard ratio; PA, physical activity. Multivariable model : Adjusted for age, sex, systolic blood pressure, smoking, level of education, waist circumference, change of waist circumference (difference between waist circumference in last follow up and baseline), alcohol intake, baseline HOMA-IR. Time dependent model: Adjusted for age, sex, systolic blood pressure, smoking, level of education, waist circumference, baseline HOMA-IR, alcohol intake (waist circumference as time-varying covariates). a Incidence of improvement; Analyzed among the participants who had HOMA-IR ≥ 2.2 at baseline (n = 38,950). b Incidence of progression; Analyzed among the participants who had HOMA-IR < 2.2 at baseline (n = 241,244). www.nature.com/scientificreports/ small sample sizes 23,24 , were not adjusted for waist circumference as a confounding factor 25,26 , were conducted in pregnant women 24 , did not exclude the diabetes population 26 , or were conducted in type 2 DM patients 23 . Owing to these limitations, the quality of evidence is relatively low 20 . By excluding participants with DM, incorporating a large number of cohorts, and conducting extensive adjustment for confounding factors, our study provided more reliable results than the previous studies.

Person year Incident cases
Our study suggested the possible lingering effect of increased PA on IR, even after the individual's PA level was decreased. This finding can be explained by the cumulative effect of exercise on IR and insulin sensitivity 27 . A previous study including 346 men and 455 women from the RISC study showed that the total amount of and accumulated number of PAs performed were the determinants of insulin sensitivity 27 . Even when the physically active participants' level of activity decreases, they still have a higher amount of total accumulated PA than the continuously sedentary population. This higher accumulated PA time in participants with decreased PA level from HEPA to SM might have led to the reduction in the HOMA-IR level.
In addition, our findings suggest that PA might have a greater impact on attenuating HOMA-IR progression than resolving the pre-existing IR. This finding supports the pre-existing notion of performing PAs as a method to prevent or delay the development of type 2 DM, which results from IR and loss of insulin secretion 28,29 . Furthermore, increasing the PA level or maintaining a high level of PA is associated with HOMA-IR improvement and prevention of HOMA-IR progression, while decreasing the PA level makes individuals more susceptible to HOMA-IR progression and decreases the likelihood of HOMA-IR improvement. Overall, our findings consistently support the beneficial effects of PA on IR 11,29 .
PA has diverse influences on IR and glucose metabolism through acute changes that cause contractionmediated glucose uptake through glucose transporter 4, and chronic adaptations causing insulin-stimulated glucose uptake 6,30,31 . Although numerous studies support the beneficial effect of PA on IR, it remains unclear whether the effect of exercise is due to the decrease in waist circumference or whether it is the effect of exercise itself 32,33 . A cross-sectional study conducted in 6,500 adults in the United States showed that PA is associated with IR 33 . However, this relationship disappeared after adjusting for differences in waist circumference, suggesting that visceral fat, expressed as waist circumference 34 , mediates the relationship between PA level and HOMA-IR 33 . Meanwhile, another study conducted in a Canadian population showed an independent association between PA and insulin sensitivity in men after adjusting for waist circumference 35 . Our study results support the finding that PA per se has a direct association with IR. However, further prospective studies are warranted to verify the relationship between PA, visceral fat, and IR.
Our study is unique as it was conducted in a large number of participants (n = 280,194), including both men (n = 156,036) and women (n = 124,158). Participants who were newly diagnosed with diabetes during the health examination and those previously diagnosed with diabetes with or without medical treatment were excluded, which made our results more reliable. A robust adjustment for confounding factors was performed, and a timedependent analysis of waist circumference, a strong independent risk factor for IR, was carried out to verify the Table 5. Associations of change in the PA with change in HOMA-IR. Multivariable model : Adjusted for age, sex, systolic blood pressure, smoking, level of education, waist circumference, change of waist circumference (difference between waist circumference in last follow up and baseline), alcohol intake, baseline HOMA-IR. Time dependent model: Adjusted for age, sex, systolic blood pressure, smoking, level of education, waist circumference, baseline HOMA-IR, alcohol intake (waist circumference as time-varying covariate). S, Sedentary, less than 600MET-minutes per week of physical activity; M, Mild physical activity, 600 MET-minutes per week; HEPA, health-enhancing physical activity : 3000 MET-minutes per week; CI, confidence interval; HOMA-IR, homeostasis model assessment-estimated IR; HR, hazard ratio; PA, physical activity. a Incidence of improvement; Analyzed among the participants who had HOMA-IR ≥ 2.2 at baseline (n = 38,950). b Incidence of progression; Analyzed among the participants who had HOMA-IR < 2.2 at baseline (n = 241,244). www.nature.com/scientificreports/ independent association between PA and IR, expressed as HOMA-IR 36 . Moreover, the dynamic relationship between the change in PA level over time and HOMA-IR was assessed. To the best of our knowledge, this was the first study to assess such associations. In addition, this was the first study to assess the association between changes in PA level and HOMA-IR trend over time.

Person year Incident cases
Despite these strengths, our study has several limitations. First, this study only included Korean individuals. Second, a self-report form (IPAQ) was used to assess the PA level since this tool is useful for evaluating a large cohort 37 . Although the IPAQ is a valid form to assess the PA level of an individual 37,38 , self-reporting and recall bias can occur 30 . Third, our study participants were young (mean age: 38.2 ± 7.7) and highly educated population (higher education = 75.4%). Age and educational attainment were associated with IR 39,40 . To overcome these limitations, we adjusted for age and education as confounding factors. In addition, the relatively young age of our study participants can highlight the relationship between PA and IR in a relatively young population. However, future prospective studies incorporating diverse races and populations are warranted to verify our results.
In conclusion, our study showed that PA level has an inverse relationship with IR, expressed as HOMA-IR. The positive effect of a high level of PA lingered even when the level of activity decreased over time. In addition, PA level might slow the progression of IR among populations without underlying IR, independent of the waist circumference and BMI status. Increasing the level of PA or maintaining HEPA can slow the progression of IR and improve IR. Our findings support the beneficial effect of PA on IR, which is associated with type 2 DM, hypertension, and dyslipidemia 5 .

Methods
Study population. The Kangbuk Samsung Health Study (KSHS) data were used in the study. The KSHS is an ongoing cohort study conducted in a Korean population aged 18 years and older who underwent comprehensive health examinations at one of the two total healthcare centers of Kangbuk Samsung Hospital in Seoul and Suwon, South Korea. In South Korea, all employees are required to undergo annual or biennial health screening examinations in accordance with the Industrial Safety and Health Law. More than 80% of the participants in the current study were either employees or spouses of employees of various companies and local government organizations. The remaining participants underwent medical checkups of their own accord.
In the KSHS, 300,187 individuals who underwent a comprehensive health examination at least twice between 2011 and 2018 were initially included. Those who met the following criteria were excluded from the analysis: participants with DM at baseline (determined based on the following factors: self-reported diabetes, use of antiglycemic medications, or previously diagnosed with DM, as indicated in the medical records), a fasting plasma glucose level of ≥ 126 mg/dL, and a hemoglobin A1c (HbA1c) level of ≥ 6.5%) (n = 10,615); individuals with missing covariates (systolic blood pressure [SBP], n = 574; alcohol intake, n = 17,209) were excluded. Overall, 280,194 participants were included in the final analysis (Fig. 1). This study was approved by the Institutional Review Board (IRB) of Kangbuk Samsung Hospital (IRB no: 2015-12-004-017). Informed consent was waived by the IRB of Kangbuk Samsung Hospital because anonymized and de-identified data were used in the analysis. All study methods were conducted in accordance with relevant guidelines and regulations.
Measurement. During health screening, the self-administered questionnaires were used to collect the demographic data, medical history, socioeconomic history including smoking status and alcohol intake, educational background, and level of PA. Alcohol intakes of > 30 g/day for men and > 20 g/day for women were defined as high alcohol intake 41,42 ; higher education was defined as college graduate or a higher level of education. The National Health Interview Survey criteria were used to define the smoking status. Current smokers were defined as those who smoked more than 5 packs (more than 100 cigarettes) in their lifetime and currently smoking at the time of the interview. A former smoker was defined as a person who had smoked more than 100 cigarettes in their lifetime but who had quit smoking at the time of the interview 43 .
The self-administered form of the Korean version of the International Physical Activity Questionnaire (IPAQ) was used to validate the PA levels 38 . In the questionnaire, participants were instructed to record the frequency and duration of PA over the past 7 days. All participants indicated the frequency (0-7 days/week) of every moderate or vigorous PA performed. PAs that lasted more than 10 min were included in the count. The duration of PA was recorded on a daily basis (min/day). In the same way, the time that the participants performed walking and other physical movements, including transportation, house chores, and working and leisure activities, were recorded (0-7 days/week and minutes/day). The total physical inactivity time was assessed using the following question: "During the last 7 days, how much time did you spend sitting or lying per each day?" Physical inactivity was defined as all activities performed while sitting or lying down. Strength exercises such as push-ups were counted separately based on the number of times per week. The participants were classified into three categories: sedentary, mild PA (600 metabolic equivalent of task [MET]-minutes per week), and health-enhancing PA (3,000 MET-minutes per week) 38 .
Anthropometric measurements (height, weight, systolic blood pressure, and diastolic blood pressure) were performed by trained medical staff. During the measurements, the participants wore a lightweight hospital gown (< 0.1 kg) without shoes. Body mass index (BMI) was calculated as weight divided by height in meters squared (kg/m 2 ). Blood pressure (BP) was measured after a period of rest in a sitting position. During the BP measurement, the arm was positioned at the heart level, and an automated oscillometric device (53,000, Welch Allyn, New York, USA) was used. Blood biochemical samples were collected after fasting for > 10 h. The blood samples were analyzed by the Laboratory Medicine Department at Kangbuk Samsung Hospital, which has been accredited by the Korean Association of Quality Assurance for Clinical Laboratories and the Korean Society of Laboratory Medicine. Student's t-test or the Mann-Whitney test was used to compare continuous variables between the two groups. Analysis of variance or Kruskal-Wallis test was used to compare multiple groups. The HOMA-IR with a rightskewed distribution was logarithmically transformed. A generalized mixed model with random effects (of individual and error) was performed to assess the longitudinal associations between HOMA-IR and PA category. The slope was estimated using the exponential coefficients and 95% CIs in the model. The HRs and 95% CIs for each improvement and progression of HOMA-IR changes according to the PA category were estimated using the Cox proportional hazards model. The multivariable model was adjusted for age, sex, SBP, smoking status (never, past, or current), educational level (< college education or ≥ college education), waist circumference, baseline HOMA-IR, and waist circumference change. A parametric proportional hazard model, including waist circumference as a time-varying covariate, was additionally implemented as a time-dependent model. For the time-varying covariate (waist circumference) and HOMA-IR level, all the data during the follow-up period were used for the analysis. For the PA, the data at baseline and the data at the end of the follow-up period (last followup) were used for the analysis. For all other variables, the data at baseline was used for the analysis. Statistical significance was defined as a two-tailed p value of < 0.05.
Ethics approval and consent to participate. This study was approved by the Institutional Review Board (IRB) of Kangbuk Samsung Hospital (IRB no: 2015-12-004-017). Informed consent was waived by the IRB of Kangbuk Samsung Hospital because anonymized and de-identified data were used in the analysis. Consent for publication. All authors gave full consent for publication. www.nature.com/scientificreports/

Data availability
All data generated or analyzed during this study are included in this published article.