Strengths and limitations of this study

This study used a large and nationally representative sample of the American adult population.

This study adjusted for multiple confounding factors and conducted sensitivity analysis to validate our conclusions.

This study employed a mediation model to explore the indirect effect of Systemic Immune-Inflammatory Index on the physical activity-diabetes mellitus (PA-DM) relationship.

The cross-sectional nature of National Health and Nutrition Examination Survey restricts causal inference, and the self-reported PA data may introduce recall bias.

The characteristics of other national populations may not be comprehensively represented, and we cannot entirely rule out the presence of unmeasured confounding variables.

Introduction

Diabetes mellitus (DM) stands as a formidable global health challenge, with its prevalence steadily rising in recent years. In 2021, the global DM population reached 529 million, and it is estimated to exceed 1.31 billion by the year 2050.1 2 This disorder, characterised by chronic hyperglycaemia, is associated with various complications such as cardiovascular issues, renal dysfunction and retinal lesions, diminishing overall quality of life.3–5 The occurrence of DM involves complex interactions among genetic, environmental and lifestyle factors.6 Among the intricate factors contributing to DM, lifestyle choices play a pivotal role, with physical activity (PA) emerging as a modifiable and impactful determinant.7

Numerous epidemiological studies and diabetes prevention trials have recognised the importance of lifestyle in DM risk, and physical activity is one of the effective lifestyle behaviors.8 9 Engaging in regular PA not only enhances insulin sensitivity but also has positive effects on weight management, a critical factor in DM prevention.10 11 Despite the robustness of these associations, the underlying mechanisms linking PA to DM risk remain intricate and warrant further exploration.

On the other hand, inflammation is also closely linked to DM and plays a pivotal role in the occurrence and development of DM.12 13 Research has demonstrated that pro-inflammatory cytokines, including tumour necrosis factor-alpha and interleukin-6, can disrupt insulin signalling pathways and induce beta cell stress and apoptosis, consequently impairing glucose uptake and utilisation, as well as decreasing insulin production and secretion, ultimately elevating blood glucose levels.14 15 The Systemic Immune-Inflammatory Index (SII) is a novel indicator that reflects the systemic inflammatory status by combining various components of peripheral blood, offering a comprehensive assessment of systemic inflammation.16 Elevated SII levels are indicative of a pro-inflammatory state.17 18 Therefore, the exploration of SII can offer deeper insights into the complex association linking PA, inflammation and DM.

Generally, moderate PA is associated with lower levels of inflammation, while prolonged periods of inactivity or sedentary behaviour may contribute to increased inflammation.19 PA can promote blood circulation and weight management, thereby aiding in the prevention or amelioration of inflammatory conditions.20 Conversely, prolonged sedentary behaviour and a lack of PA can disrupt physiological processes, leading to the release of inflammatory markers and adversely affecting health.21 Thus, PA plays an important role in maintaining immune system balance and reducing chronic inflammation.

PA and inflammation are inextricably linked to DM respectively, and PA and inflammation are closely related processes. However, the relationship between these three is complex and not fully understood, clarifying this intricate association will help to determine the prevention and treatment strategies of DM. To further explore the relationship between PA, inflammatory status and DM, we evaluated the association between PA and DM in adult population based on the National Health and Nutrition Examination Survey (NHANES) and conducted mediation analyses to examine the possible mediating role of SII in the link between PA and DM.

MethodsParticipants and study design

NHANES, carried out by the National Centre for Health Statistics, is a cross-sectional study aimed at assessing the health and nutritional status of both adults and children in the USA. Employing a sophisticated, stratified multistage probability survey design, the survey gathers data through standardised interviews, physical examinations and biological sample testing. NHANES annually recruits a demographically representative sample and obtains informed consent from participants. This study uses the publicly accessible NHANES dataset, which is available at https://www.cdc.gov/nchs/index.htm22. Our analysis delved into NHANES data from seven survey cycles spanning 2005 to 2018, encompassing a substantial cohort of 70 190 participants. Among 39 749 adult participants (aged ≥20 years), exclusions were made: 566 participants who were pregnant and lactated, 145 with missing DM data, 10 734 with missing PA data, 2291 with missing SII data, 2174 with missing demographics data, 4 with missing medical history data, 2348 with missing smoking and drinking data, 113 with missing Body Mass Index (BMI) data, 11 331 with missing laboratory test data, 13 participants with sleep duration data and 537 participants with a weight of 0 were excluded. Ultimately, we compiled a comprehensive dataset comprising 9493 participants for our thorough statistical analysis (figure 1).

Figure 1

Flowchart of participants selection.

Exposure variable

Each NHANES survey recall records detailed information on the specific PA type, frequency, intensity and duration of PA events reported through self-administered questionnaire by participants. This information includes two different time periods: the last 30 days (2005–2006) and the most recent week (2007–2018). Using this information, combined with the type and intensity of activity, we calculated a metabolic equivalent (MET) score for the specific activity.23 These MET scores were subsequently multiplied by the average engagement frequency and duration over the preceding 30 days or week to calculate the MET minutes for each activity within the respective 30-day period (MET min/30d) or week (MET min/week). For those participants who reported PA on a monthly basis, we divided the total MET minutes by 30 and multiplied the result by 7 to calculate the total MET minutes per week (MET min/week) for each participant. Finally, participants were divided into three levels according to the standard scoring criteria of the International Physical Activity Questionnaire: low (<600 MET min/week), moderate (600–3000 MET min/week) and high (≥3000 MET min/week).24

Mediator variable

The SII is derived from results obtained in a complete blood cell count, with the laboratory methodology for this test comprehensively outlined on the NHANES website (https://www.cdc.gov/nchs/nhanes/index.htm). The calculation of SII follows a standard formula: SII=P × N/L, where P, N and L represent the counts of platelets, neutrophils and lymphocytes in peripheral blood, respectively.25

Outcome variable

Combining questionnaire data and examination data, the DM diagnostic criteria used in this study encompass the following: (1) physician-diagnosed DM; (2) glycated haemoglobin (HbA1c) >6.5%; (3) fasting blood glucose (FBG) ≥7.0 mmol/L; (4) random blood glucose ≥11.1 mmol/L; (5) 2-hour Oral Glucose Tolerance Test blood glucose ≥11.1 mmol/L; and (6) use of diabetes medication or insulin.

Definition of other variables

Drawing on previous research, our multivariate model incorporates potential confounding variables associated with the relationship between PA and DM. These variables encompass age, gender (female or male), race (non-Hispanic black, Hispanic white, Mexican American, other Hispanic, other race), education level (less than high school, completed high school and more than high school), marital status (never married, partner or married, separated and divorced, widowed), income status based on poverty index ratio (PIR) (<1, 1–4, >4), smoking status (never, former smoker, current smoker), alcohol use (never, former drinker, current drinker), hypertension (no or yes), hyperlipidemia (no or yes), cardiovascular diseases (CVD) (no or yes), estimated glomerular filtration rate (eGFR), HbA1c and FBG. Alternatively, we evaluate multicollinearity among multivariate variables using the Variance Inflation Factor, and it is observed that there is multicollinearity between insulin and homeostatic model assessment-insulin resistance (HOMA-IR).

Statistical analysis

In accordance with the Centres for Disease Control and Prevention recommendations, the description and statistical analysis of the study data used appropriate sample weights for weighted analysis. Baseline characteristics of the population were presented as mean±SD for continuous variables and as numbers (ratios) for categorical variables. Feature differences between the DM and non-DM groups were assessed using weighted Student’s t-test (for continuous variables) or weighted χ² test (for categorical variables). Weighted linear regression (for continuous variables) or weighted χ² tests (for categorical variables) were employed to evaluate inter-group differences in the distribution of three categories of PA.

Employing three distinct models, we used multivariate logistic regression to estimate the OR and 95% CI for the association between PA, SII and DM. Three models were proposed: Model 1, which did not adjust for any confounding factors; Model 2, which adjusted for age, gender, race, education level, marital status, PIR, BMI, sleep duration, smoking status, drinking status, hypertension, hyperlipidemia, CVD, eGFR, HbA1c and FBG; Model 3, adjusted for all covariates incorporated in Model 2 as well as SII. Notably, Models 1, 2 and 3 were used to assess the association between PA and DM; whereas, only Models 1 and 2 were used for the association between SII and DM. Moreover, we strategically positioned three nodes at the 25th, 50th and 75th percentiles of the distribution of PA. This facilitated the construction of a restricted cubic spline regression model to examine the non-linear relationship between PA and DM. If the relationship exhibits non-linearity, we estimate the most probable threshold and apply a two-segment logistic regression risk model to examine the correlation between PA and DM on both sides of the inflection point. It is noteworthy that during the process of statistical analysis, we observed an uneven distribution and a pronounced left skewness in the PA data, leading to the application of a natural logarithm transformation to make it more suitable for our statistical analysis.

Additionally, to effectively understand the complex relationship between PA, SII and DM, we applied a mediation model based on weighted linear regression and logistic regression. This model was used to investigate both the direct impact of PA on DM risk and the indirect effect mediated by SII, treating PA as a continuous variable. The indirect effect was estimated using the product of coefficients method, by multiplying the coefficient of the PA-SII relationship with that of the SII-DM relationship. And three effect estimates from the mediation model were computed through bootstrap method, followed by significance testing. Finally, to affirm the robustness of the correlation between PA and DM, we performed a sensitivity analysis using stratified logistic regression and assessed interactions by introducing a multiplicative term between the two variables in the logistic regression. All statistical analyses were conducted with R 4.2.3 (R Foundation for Statistical Computing, Vienna, Austria), and p values <0.05 (two-tailed) were considered statistically significant.

ResultsCharacteristics of the study population

This study comprised a total of 9493 participants, with 1672 diagnosed with DM. The baseline characteristics of the participants are detailed in table 1. In comparison to the non-DM group, the DM group exhibited a higher proportion of non-Hispanic Black, education level less than high school, widowed, moderate-income level, former smokers, former drinkers, obesity, short sleep duration (≤6 hours) and individuals with underlying conditions (hypertension, hyperlipidemia and CVD) (p<0.05). Additionally, participants with DM tended to have higher levels of age, DM-related indicators (HbA1c, FBG, insulin and HOMA-IR) and SII (p<0.05). Conversely, we observed lower eGFR levels in the DM group, and they were more inclined to have lower levels of PA (p<0.05) (online supplemental table 1).

Table 1

Characteristics of the participants according to the classification of PA

Characteristics of the participants according to the classification of PA

Table 1 outlines the characteristics of participants based on the three-class classification of PA. With an increase in PA levels, participants tend to have lower age, HbA1c and SII levels, while eGFR levels are higher (p<0.05). And we observed no significant association between consistent engagement in PA and levels of FBG, insulin, HOMA-IR or obesity (p>0.05). Furthermore, compared with those with lower levels of PA, participants who adhere to PA are more likely to be male, never married, former smokers and current drinkers, with a lower prevalence of underlying conditions (hypertension, hyperlipidemia, CVD and DM) (p<0.05).

Risk of DM according to the classification of PA and SII

The logistic regression models for assessing the association of PA and SII with DM. Across all three models, high levels of PA were significantly negatively associated with the risk of DM compared with moderate levels of PA [OR: 0.62 (95% CI: 0.47 to 0.83)]. Likewise, the analyses with PA to increase by 1-SD and the natural logarithm-physical activity (Ln-PA) yielded similar results. However, only Model 1 exhibited a noticeable trend (p-trend <0.05) (online supplemental table 2).

In Model 1, compared with the first tertile, SII showed a positive correlation with the risk of DM in the second tertile [OR: 1.23 (95% CI: 1.02 to 1.49)] and the third tertile [OR: 1.69 (95% CI: 1.43 to 2.00)]. In Model 2, SII also exhibited a positive association with the risk of DM in the third tertile [OR: 1.46 (95% CI: 1.09 to 1.96)]. Furthermore, a significant dose-response relationship was observed in both models (p-trend <0.05). The analysis with SII increasing by 1-SD yielded similar results (online supplemental table 2).

As illustrated in figure 2, the restricted cubic spline curve confirms a consistent non-linear association between Ln-PA and DM risk across all three models (p-non-linear <0.05), featuring an inverted U-shaped dose-response curve. An inflection point was identified at 6.71. Using a two-segment logistic regression model, we observed that in all three models, when Ln-PA <6.71, PA showed no significant association with DM risk. In contrast, when PA >6.71, PA was negatively associated with the risk of DM in all three models [OR: 0.81 (95% CI: 0.71 to 0.93)], with a 0.74 difference in regression coefficients between the two-segment models (p=0.018). Importantly, the likelihood ratio tests between the segmented models and the original logistic regression model all demonstrated statistical significance (p values for the log-likelihood ratio=0.015). Details are shown in online supplemental table 3.

Figure 2

The restricted cubic spline curve was used to model the relationship between Ln-PA and the risk of DM (A–C). (A) did not adjust any covariates; (B) was adjusted for age, gender, race, education level, marital status, PIR, smoking status, alcohol use, hypertension, hyperlipidemia, CVD, eGFR, HbA1c and FBG; (C) was adjusted for age, gender, race, education level, marital status, PIR, BMI, sleep duration, smoking status, alcohol use, hypertension, hyperlipidemia, CVD, eGFR, HbA1c, FBG and SII. The vertical red dashed line represents the inflection point of the non-linear curve estimate. BMI, Body Mass Index; CVD, cardiovascular disease; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HbA1c, glycated haemoglobin; Ln-PA, natural logarithm-physical activity; PIR, poverty index ratio; SII, Systemic Immune-Inflammatory Index.

Mediating role of SII

The results of the mediation analysis are shown in figure 3. After adjusting for covariates, an increase in PA remains associated with a reduced risk of DM. Both the direct effect (β: −1.92×10-6, 95% CI: −2.07×10-6, −4.50×10-7) and the indirect effect (β: −6.18×10-8, 95% CI: −5.89×10-8, −9.30×10-9) through SII were statistically significant (p=0.004). The proportion of the effect explained by the indirect impact was 3.12% (figure 3B). Additionally, the analysis with Ln-PA produced comparable results, showing that the proportion of the effect attributed to SII was 10.46% after adjusting for covariates (figure 3D). The directed acyclic graph of the mediation model is shown in figure 4.

Figure 3

Mediating effects of SII on the association between PA or Ln-PA and DM (A–D). The 95% CI of these estimates was computed using the bootstrap method (1000 samples). (A and C) did not adjust any covariates; (B and D) were adjusted for age, gender, race, education level, marital status, PIR, BMI, sleep duration, smoking status, alcohol use, hypertension, hyperlipidemia, CVD, eGFR, HbA1c and FBG. BMI, Body Mass Index; CVD, cardiovascular disease; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HbA1c, glycated haemoglobin; Ln-PA, natural logarithm-physical activity; PA, physical activity; PIR, poverty index ratio; SII, Systemic Immune-Inflammatory Index.

Figure 4

The directed acyclic graph of the mediation model. The total natural indirect effect (TNIE) is represented by the grey and yellow lines through the mediator Systemic Immune-Inflammation Index (a combination of paths a and b); the pure natural direct effect (PNDE) is indicated by the blue line; the total effect (TE) is a combination of the grey, yellow and blue lines. The influence of potential confounders is represented by the green line.

Sensitivity analysis

As shown in figure 5, our subgroup analyses showed that overall, the trend in the association between PA and DM remained consistent. High PA levels were inversely linked to DM, but the association varied among subgroups. For participants in the second tertile of SII, aged 40–59, other Hispanic and other races, education level less than high school, completed high school, separated or divorced, PIR <1, PIR ≥4, former smoker, never drinkers, normal and underweight, overweight, longer sleep duration, without hyperlipidemia, and individuals with CVD, high PA levels showed a negative association with DM, but this association lacked statistical significance. In the case of never married and current smoker participants, high PA levels indicated a positive correlation with DM, yet this association did not exhibit statistical differences. Besides, high PA levels in the remaining subgroups exhibited statistical differences concerning DM. Interaction tests indicated no significant alterations in the association between PA and DM across all variables, suggesting these factors do not significantly impact this relationship (p-interaction <0.05).

Figure 5

Forest plot of stratified analyses of the associations between PA on DM. Adjusted for age, gender, race, education level, marital status, PIR, BMI, sleep duration, smoking status, alcohol use, hypertension, hyperlipidemia, CVD, eGFR, HbA1c, FBG and SII, but except for the one defining the subgroup. BMI, Body Mass Index; CVD, cardiovascular disease; DM, diabetes mellitus; eGFR, estimated glomerular filtration rate; FBG, fasting blood glucose; HbA1c, glycated haemoglobin; PA, physical activity; PIR, poverty index ratio; SII, Systemic Immune-Inflammatory Index.

Discussion

This study finally included 9493 participants from the NHANES 2005–2018 cohort for analysis, including 4458 females and 5035 males. Of these, 1672 patients were diagnosed with DM. We found that high levels of PA were negatively correlated with the risk of DM, while SII showed a positive correlation with the risk of DM. To the best of our knowledge, our study is the first to report the mediation effect of SII between PA and the risk of DM.

Many studies have highlighted the importance of PA for health, and considerable evidence points to the positive improvements associated with PA in DM. Our findings are broadly consistent with existing research illustrating that participating in PA was related to a decreased prevalence of DM.26 According to the Diabetes Prevention Programme study (n=3234), both self-reported and accelerometer-measured PA demonstrated an inverse association with the risk of DM.27 In a randomised controlled experiment conducted in Finland, moderate-to-vigorous leisure-time PA demonstrated a substantial risk reduction of 63%–65% for DM.28In a prospective cohort study involving Chinese individuals with impaired fasting glucose, those engaging in low, moderate and high volumes of PA demonstrated respective reductions of 12%, 20% and 25% in the risk of DM compared with inactive individuals.29 In addition, previous evidence indicates that elevated PA levels can significantly improve HbA1c in individuals with DM.30 Similarly, in our study, with an increase in PA levels, participants tend to have lower HbA1c. Besides, we also observed that participants consistently engaging in PA may display lower levels of FBG, insulin, HOMA-IR and obesity, although the clinical relevance is yet to be definitively established. The non-linear association between Ln-PA and DM risk, characterised by an inverted U-shaped dose-response curve with a turning point at an Ln-PA value of 6.71, suggests a potential threshold effect. In the two-segment logistic regression model, Ln-PA <6.71 showed no significant association with an increase in DM risk. In contrast, when Ln-PA >6.71, the DM risk was significantly reduced. This finding emphasises that only a certain level of physical activity (Ln-PA>6.71) can effectively reduce diabetes risk, highlighting the importance of personalised recommendations for optimal PA levels to mitigate DM risk.

Moreover, it is well known that inflammation plays a crucial role in the occurrence and progression of DM.12 13 Kashima et al reported that elevated white blood cell counts and C-reactive protein levels independently predicted the risk of DM, and their combined presence demonstrated a higher predictive value for DM compared with either alone.31 Another study found that the erythrocyte sedimentation rate was significantly increased in patients with DM compared with the healthy population.32 As a novel inflammatory marker, SII has significant value in both medical research and clinical practice, which can better grasp the extensive immune response and inflammatory state of the body.33 34 To the best of our knowledge, our study is the first to report that the SII level of patients with DM was higher than that of healthy controls by using data from the NHANES database. However, several research studies have demonstrated that SII is associated with diverse conditions related to DM, including diabetic kidney disease, diabetic macular oedema, diabetic foot and others.35–37 These findings underscore the pivotal research significance of SII in DM and its related conditions. In our study, SII exhibited a positive association with the risk of DM, and a dose-response relationship was observed. Therefore, managing systemic inflammation may be an effective avenue for preventing DM, aligning with the current understanding of the association between inflammation and DM.12

Existing studies have revealed that PA was strongly associated with the levels of cytokines, which were also involved in various regulatory and inflammatory processes.38–40 These findings led us to propose whether there was a mediating role of inflammatory marker in the relationship between PA and DM. Strikingly, we found that SII significantly regulated this process by conducting the mediation analysis. An increase in SII was consistently associated with an increased risk of DM, and the link between PA and DM was mediated by SII, with mediation proportions of 4.32%, and 3.12% for covariates adjusted. Moreover, SII also acted as a mediator in the link between Ln-PA and DM, with mediation proportions of 12.11%, and 10.46% for covariates adjusted. This suggests that the impact of PA on DM risk may, in part, be mediated through its influence on systemic inflammation. Certainly, SII provided us a new clue to explore the underlying mechanism between the relationship of PA and DM.

The greatest strength of this study was the utilisation of NHANES data, providing a large and nationally representative sample, thereby enhancing the generalisability of our findings. Additionally, we adjusted for multiple confounding factors to produce more reliable results and conducted sensitivity analysis to validate our conclusions. Our findings underscore the importance of promoting regular PA as a modifiable lifestyle factor in DM prevention and reveal the mediating role of SII, providing insights into the complex link PA to DM.

However, our study has certain limitations. First, the cross-sectional nature of NHANES restricts causal inference, and the self-reported PA data may introduce recall bias. Second, the findings of this research were derived from American adults, and therefore, the characteristics of other national populations may not be comprehensively represented. Third, despite adjusting for numerous potential confounders, we cannot entirely rule out the presence of unmeasured confounding variables. Additionally, the NHANES database does not specify diabetes type, so our analysis assumed the majority of cases are type 2 diabetes, which may introduce limitations in our findings. Lastly, a one-off measurement of physical activity and SII may not accurately represent long-term levels of these variables.

Conclusion

In conclusion, this study investigated the relationship among PA, SII and DM. We provide robust evidence supporting the inverse association between PA and DM risk while highlighting the mediating role of inflammation, as reflected by SII. These findings contribute valuable insights to inform public health strategies and clinical interventions aimed at reducing the global burden of DM.

Data availability statement

Data are available in a public, open access repository.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.

Acknowledgments

We thank all participants in the NHANES study. We thank the NHANES research team for providing the data.