General and Abdominal Obesity and Incident Distal Sensorimotor

240 Diabetes Care Volume 42, February 2019

SabrinaSchlesinger,1,2 ChristianHerder,2,3,4 General and Abdominal Obesity Julia M. Kannenberg,2,3 Cornelia Huth,2,5 Maren Carstensen-Kirberg,2,3 and Incident Distal Sensorimotor Wolfgang Rathmann,1,2,4 Gidon J. Bonhof,¨ 3 Wolfgang Koenig,6,7,8 Margit Heier,5 Polyneuropathy: Insights Into Annette Peters,2,5 Christa Meisinger,5,9 ﬂ Michael Roden,2,3,10 Barbara Thorand,2,5 In ammatory Biomarkers as and Dan Ziegler2,3,10 Potential Mediators in the KORA F4/FF4 Cohort Diabetes Care 2019;42:240–247 | https://doi.org/10.2337/dc18-1842

1Institute for Biometrics and Epidemiology, German Diabetes Center, Leibniz Center for Di- abetes Research at Heinrich Heine University OBJECTIVE Dusseldorf,¨ Dusseldorf,¨ Germany 2 To investigate the associations between different anthropometric measurements German Center for Diabetes Research, Munchen-Neuherberg,¨ Germany and development of distal sensorimotor polyneuropathy (DSPN) considering 3Institute for Clinical Diabetology, German Di- fl EPIDEMIOLOGY/HEALTH SERVICES RESEARCH interaction effects with prediabetes/diabetes and to evaluate subclinical in am- abetes Center, Leibniz Center for Diabetes Re- mation as a potential mediator. search at Heinrich Heine University Dusseldorf,¨ Dusseldorf,¨ Germany RESEARCH DESIGN AND METHODS 4Medical Faculty, Heinrich Heine University Dusseldorf,¨ Dusseldorf,¨ Germany This study was conducted among 513 participants from the Cooperative Health 5Institute of Epidemiology, Helmholtz Zentrum Research in the Region of Augsburg (KORA) F4/FF4 cohort (aged 62–81 years). Munchen,¨ German Research Center for Environ- Anthropometry was measured at baseline. Incident DSPN was defined by neu- mental Health, Neuherberg, Germany 6 ropathic impairments using the Michigan Neuropathy Screening Instrument at Deutsches Herzzentrum Munchen,¨ Technische Universitat¨ Munchen,¨ Munich, Germany baseline and follow-up. Associations between anthropometric measurements and 7German Center for Cardiovascular Research, DSPN were estimated by multivariable logistic regression. Potential differences by Partner Site Munich Heart Alliance, Munich, diabetes status were assessed using interaction terms. Mediation analysis was Germany 8 conducted to determine the mediation effect of subclinical inflammation in these Department of Internal Medicine II-Cardiology, University of Ulm Medical Center, Ulm, Germany associations. 9Chair of Epidemiology, Ludwig-Maximilians- Universitat¨ Munchen¨ am UNIKA-T Augsburg, RESULTS Augsburg, Germany After a mean follow-up of 6.5 years, 127 cases with incident DSPN were detected. 10Division of Endocrinology and Diabetology, Both general and abdominal obesity were associated with development of DSPN. Medical Faculty, Heinrich Heine University Dusseldorf,¨ Dusseldorf,¨ Germany The odds ratios (95% CI) of DSPN were 3.06 (1.57; 5.97) for overweight, 3.47 (1.72; Corresponding author: Sabrina Schlesinger, 7.00) for obesity (reference: normal BMI), and 1.22 (1.07; 1.38) for 5-cm differences [email protected] in waist circumference, respectively. Interaction analyses did not indicate any Received 31 August 2018 and accepted 4 differences by diabetes status. Two chemokines (C-C motif chemokine ligand 7 November 2018 [CCL7] and C-X-C motif chemokine ligand 10 [CXCL10]) and one neuron-specific This article contains Supplementary Data online marker (Delta/Notch-like epidermal growth factor–related receptor [DNER]) were at http://care.diabetesjournals.org/lookup/suppl/ identified as potential mediators, which explained a proportion of the total effect doi:10.2337/dc18-1842/-/DC1. up to 11% per biomarker. S.S. and C.He. contributed equally to this work. B.T. and D.Z. contributed equally to this work. CONCLUSIONS © 2018 by the American Diabetes Association. General and abdominal obesity were associated with incident DSPN among in- Readers may use this article as long as the work is properly cited, the use is educational and not dividuals with and without diabetes, and this association was partly mediated by fi fl for pro t, and the work is not altered. More infor- in ammatory markers. However, further mechanisms and biomarkers should be mation is available at http://www.diabetesjournals investigated as additional mediators to explain the remainder of this association. .org/content/license. care.diabetesjournals.org Schlesinger and Associates 241

Distal sensorimotor polyneuropathy (DSPN) cohort (9,10). It remains to be answered foot ulceration, ankle reflexes, and vi- is associated with several adverse health whether and to what extent these bio- bration perception threshold using the outcomes, including increased morbidity markers may explain the link between C 64 Hz Rydel-Seiffer tuning fork at such as foot ulcers and lower-extremity obesity and DSPN. the great toes. Age-dependent lower amputation, occurrence of chronic pain, The aims of this study were threefold. limits of normal for vibration perception depression, impaired quality of life, and First, we investigated the association be- threshold at the fifth percentile were risk of premature death (1,2). The prev- tween general and abdominal obesity and computed according to Martina et al. alence and incidence vary according to incident DSPN. Second, potential differen- (14) using the equation y = 5.75 2 assessment method of DSPN, prediabe- ces in these associations between individ- 0.026 3 age. The neuropathy assessment tes/diabetes status, and diabetes dura- uals with and without prediabetes/diabetes also included a bilateral examination of tion. The prevalence of DSPN has been were assessed. Third, inflammatory bio- sensory perception with a 10-g mono- estimated to be ;30% for hospital-based markers were considered as potential filament (Neuropen) (9). Thus, the total populations, 20% for the general pop- mediators of the observed associations. MNSI score ranged from 0 (all aspects ulation, and 10–15% for newly diagnosed normal) to a maximum of 10 points. patients with type 2 diabetes, whereas RESEARCH DESIGN AND METHODS Prevalent or incident DSPN were defined with longer duration of diabetes (.10 Study Design and Participants basedonacutoffvalueof.3pointsas years), the prevalence increases up to The study design has been described described in our previous study (9). 50% (3,4). Evidence has emerged sug- previously in detail (9,11). Briefly, this gesting that the prevalence of DSPN is prospective cohort study is based on the Assessment of Anthropometry increased even in individuals with pre- KORA F4 (2006–2008) and KORA FF4 Body weight, height, and waist and hip diabetes compared to those with normal studies (2013 to 2014), which are both circumferenceweremeasuredbytrained glucose tolerance (1). Major risk factors follow-up examinations of the KORA S4 personnel based on standard protocols of DSPN include advanced age, smoking, study (1999–2001), a population-based (12). Body weight was assessed in light and metabolic factors, such as diabetes, cohort study conducted in the region of clothing to the nearest 0.1 kg. BMI hyperglycemia, and obesity (5). Augsburg, Germany. (kg/m2) was calculated as a measure Previous studies that reported asso- The study sample is identical to the of general body fatness, and individ- ciations between obesity, measured by sample used for a comprehensive anal- uals with a BMI of 18.5 to ,25 kg/m2 BMI or waist circumference, and DSPN ysis of associations between biomarkers were defined as normal weight, 25 to were mostly cross-sectional (2–6). Only of inflammation and incident DSPN (10). ,30 kg/m2 as overweight, and $30 one recent prospective study reported Briefly, out of the 1,161 KORA F4 study kg/m2 as obese (15). Waist circumfer- a positive association between obesity participants aged 62–81 years, we ex- ence was measured at the minimum and increased risk of diabetic polyneu- cluded participants with missing in- abdominal girth, and hip circumference ropathy (7,8). Further data on this relation- formation of exposure variables and was assessed at the maximum protrusion ship are lacking, and the previous study covariates (n = 113), missing follow-up of the hips at the level of the symphysis focused only on DSPN in patients with information (n = 452), and prevalent pubis to the nearest 0.1 cm (16). Ab- type 2 diabetes. It remains unclear DSPN (n = 83), resulting in an analysis dominal obesity was defined by the waist whether diabetes or prediabetes, de- sample of 513 participants. circumference according to the classifi- fined as impaired fasting glucose (IFG) Characteristics of the cohort, including cation of the International Diabetes and/or impaired glucose tolerance (IGT), age, sex, socioeconomic status, anthro- Federation (IDF) with sex-specific cutoff are effect modifiers of the relationship pometry, metabolic variables, lifestyle points (,94 vs. $94 cm for men and ,80 between obesity and DSPN. Moreover, factors, and glucose tolerance status vs. $80 cm for women) (17) and accord- the underlying mechanisms explaining using standard 75-g oral glucose toler- ing to the classification of the World this association are not well understood. ance tests were assessed as previously Health Organization (WHO) (,102 vs. A possible mediator for this relationship reported. Prediabetes was defined as the $102 cm for men and ,88 vs. $88 might be a low-grade inflammatory state. presence of IFG and/or IGT (12,13). DSPN cm for women) (15). Further measures In two recent reports, two proinflamma- was assessed at baseline and follow- for abdominal obesity were defined by tory cytokines (interleukin-6 [IL-6] and up examinations. All examinations were the waist-to-hip ratio (WHR), calculated tumor necrosis factor-a [TNF-a]), three carried out in accordance with the Dec- as waist circumference divided by hip cir- chemokines (C-C motif chemokine ligand laration of Helsinki, including written cumference (cutoffs ,0.95 vs. $0.95 for 7 [CCL7], C-X-C motif chemokine ligand informed consent from all participants. men and ,0.80 vs. $0.80 for women 9 [CXCL9], and CXCL10), as well as three The study was approved by the ethics (15), and the waist-to-height ratio (WHtR), soluble forms of inflammation-related board of the Bavarian Chamber of Physi- calculated as waist circumference di- membrane proteins (Delta/Notch-like cians (Munich, Germany). vided by height. epidermal growth factor–related receptor [DNER], CD40L receptor [CD40], and Assessment of DSPN Measurement of Biomarkers of TNF receptor superfamily member 9 DSPN was assessed using the Michigan Subclinical Inflammation [TNFRSF9]) have been identified as in- Neuropathy Screening Instrument (MNSI) Fasting serum levels of IL-6 and TNF-a dependent risk markers for incident as described (9). Briefly, the clinical ex- were measured by ELISA as described DSPN in the Cooperative Health Research amination part of the MNSI comprised previously (9,18). Fasting serum levels of in the Region of Augsburg (KORA) F4/FF4 the following items: appearance of feet, CCL7, CXCL9, CXCL10, DNER, CD40, and 242 Obesity, Inflammation, and Incidence of DSPN Diabetes Care Volume 42, February 2019

TNFRSF9) were measured using the models for BMI and WHtR). In addition, mediation effect was interpreted as sta- OLINK Inflammation multiplex immuno- we investigated in a third model whether tistically significant. assay (OLINK Proteomics, Uppsala, Swe- cardiometabolic factors, which might lie The statistical analyses were con- den) with intra-assay coefficients of on the pathway from obesity to incident ducted with R version 3.3.3 (www variation ranging from 2.3% to 6.5% DSPN, influence the association under .R-project.org/). Restricted cubic spline and interassay coefficients of variation investigation. Model 3 was based on regression models were analyzed using ranging from 5.4% to 10.3% as described model 2 with additional adjustment the “rms” package in R. We used the R (10). The assay uses proximity extension for cardiometabolic factors, including package “mediation” performing a quasi- assay technology, which combines a de- HbA1c, hypertension (defined as blood Bayesian approximation with 1,000 tection step using oligonucleotide- pressure $140/90 mmHg or use of an- Monte Carlo draws for CI estimation. To labeled antibodies, a proximity-dependent tihypertensive medication), HDL choles- test for exposure–mediator interaction, DNA polymerization event, and a real- terol, LDL cholesterol, and triglycerides. we used the test.TMint function (26). time quantitative PCR amplification. Rela- The main analysis (model 2) was per- The P values presented were two tailed, tive protein values are given as normalized formed in the total cohort and stratified and P , 0.05 was considered statistically protein expression (NPX) values (19,20), by prediabetes/diabetes status (predia- significant. which have similar distributions as log2- betes and diabetes vs. normal glucose transformed protein concentrations. Us- tolerance). Prediabetes and diabetes RESULTS ing this technology, novel associations were combined to increase statistical Characteristics between biomarkers and cardiometabolic power in these analyses. Potential differ- Out of 513 participants, 127 developed risk were identified in multiple cohorts ences between persons with and without DSPN during a mean follow-up time of (19–23). prediabetes/diabetes were assessed by 6.5 (SD 6 0.2) years. Table 1 shows including an interaction term between the baseline characteristics of study Statistical Analysis anthropometric measurement and pre- participants with and without DSPN at Descriptive statistics are reported for diabetes/diabetes status in the logistic follow-up as partly reported before (10). the participants stratified by incident regression models. We investigated po- Participants with incident DSPN were DSPN shown as mean 6 SD, median tential nonlinear associations between older, had higher BMI values, were (interquartile range), or percentages. anthropometric measurements and in- more often overweight and obese, The associations between anthropom- cident DSPN by using restricted cubic had a higher waist circumference, WHR, etry (BMI, waist circumference, WHR, spline regression models, with three and HbA1c, were more often smokers, and WHtR) and incident DSPN were equally spaced knots between the 0.05 and were physically less active com- analyzed by calculating odds ratios and 0.95 quantiles. pared with participants without DSPN. (ORs) with corresponding 95% CIs in In addition, we investigated whether Differences were also observed for bio- logistic regression models. The exact biomarkers of subclinical inflammation markers of subclinical inflammation date of the manifestation of DSPN was were mediators of the identified associ- with higher levels of IL-6, CCL7, CXCL9, not known, and thus, we conducted ation between general and abdominal CXCL10, CD40, and TNFRSF9 in individ- logistic regression analysis rather than obesity and DSPN. In total, eight bio- uals with incident DSPN compared with survival analysis. Anthropometric mea- markers of subclinical inflammation (IL-6, individuals without DSPN. sures were investigated as predefined TNF-a, CCL7, CXCL9, CXCL10, DNER, categories as described above and as CD40, and TNFRSF9) showing an associ- Association Between Anthropometric 2 continuous measures (per 5 kg/m for ation with DSPN (which is one criterion Measures and DSPN BMI, per 5 cm for waist circumference, that needs to be fulfilled for the medi- General obesity defined by the BMI was and per 0.1 units for WHR and WHtR). ation analysis [24]) were selected based positively associated with incident DSPN We fitted three different regression on previous reports (9,10). Another cri- (Table 2). The multivariable adjusted OR models. The first model was adjusted for terion that needs to be satisfied is the (95% CI) for DSPN was 3.06 (1.57; 5.97) age and sex and the second (our main relation between the exposure (anthro- for overweight and 3.47 (1.72; 7.00) for model) for age, sex, socioeconomic sta- pometric measures) and the mediator obesity (Table 2, model 2) compared tus (including information on education, (biomarkers of subclinical inflammation). with normal weight. The multivariable income, and occupation, combined as a BMI was used to assess general obesity adjusted OR (95% CI) of DSPN was 1.68 score [range: 1–27 points] and classified and waist circumference for abdominal (1.28; 2.22) for an increment of 5 kg/m2 as low [1–9 points], moderate low [10–12 obesity. Thus, we checked if the poten- of BMI. When additionally adjusted for points], moderate [13–15 points], mod- tial mediators were associated with BMI metabolic factors (HbA1c, hypertension, erate high [16–19 points], or high [.19 or waist circumference, respectively. Fi- HDL cholesterol, LDL cholesterol, and points]), smoking status (never, former, nally, each potential mediator was added triglycerides), there was no substantial or current), alcohol intake (abstainer [0 to the main model (model 2) to estimate change in the OR (Table 2, model 3). g/day], moderate [.0to,40 g/day for the proportion mediated as indirect nat- There was no indication for a nonlinear men and .0to,20 g/day for women], ural effect divided by the sum of the association between BMI and DSPN and high [$40 g/day for men and $20 direct and indirect natural effect by (Supplementary Fig. 1A)(P for nonline- g/day for women]), physical activity (in- applying causal mediation analysis (25). arity = 0.163). active [,1 h of sports/week] or active If the 95% CI of the percent mediated For abdominal obesity defined as [$1 h sports/week]), and height (not in effect did not include the null value, the waist circumference, the multivariable care.diabetesjournals.org Schlesinger and Associates 243

Table 1—Baseline characteristics of the study population in KORA F4 stratified by OR (95% CI) for DSPN was 2.06 (1.30; incidence of DSPN* 3.26) when using WHO criteria and 2.72 Variable No incident DSPN Incident DSPN P (1.42; 5.19) when using the stricter IDF classification (Table 2, model 2). A 5-cm N 386 127 increment of waist circumference was Age (years) 67.9 6 4.6 70.2 6 5.0 ,0.001 also associated with incident DSPN Men (%) 49.2 55.9 0.169 (multivariable OR [95% CI] 1.24 [1.11; † Socioeconomic status (%) 0.139 1.39]). The associations between waist Low 23.6 27.0 Moderate low 21.2 19.8 circumference and DSPN were robust Moderate 21.8 27.8 after additional adjustment for meta- Moderate high 19.2 11.1 bolic factors (model 3). The association High 14.2 14.3 between waist circumference and DSPN Height (cm) 166 6 9 167 6 9 0.016 was linear (Supplementary Fig. 1B)(P for BMI (kg/m2) 27.6 6 3.9 29.2 6 4 ,0.001 nonlinearity = 0.941). BMI classes (%) 0.001 For further measurements of abdom- Normal weight (18.5 to ,25 kg/m2) 27.5 11.0 inal obesity (WHR and WHtR), similar Overweight (25 to ,30 kg/m2) 46.1 52.8 associations were observed (Table 2). 2 Obese ($30 kg/m ) 26.4 36.2 Higher compared with lower WHR Waist circumference (cm) 94.7 6 11.3 100.1 6 11.4 ,0.001 indicated a positive, but not statistically Waist circumference classes (%) ,0.001 significant, association with DSPN (mul- fi WHO de nition tivariable adjusted OR [95% CI] 1.51 Low (M: ,102, W: ,88 cm) 53.1 34.6 [0.91; 2.50]). When we investigated High (M: $102, W: $88 cm) 46.9 65.4 IDF definition 0.001 WHR as a continuous trait, a positive Low (M: ,94, W: ,80 cm) 24.1 11.0 statistically significant association was ob- High (M: $94, W: $80 cm) 75.9 89.0 served for an increment of 0.1 units of WHR 0.90 6 0.08 0.92 6 0.08 0.004 WHR and incident DSPN (multivariable WHR classes (%) 0.071 adjusted OR [95% CI] 1.79 [1.20; 2.67]). Low (M: ,0.95, W: ,0.8) 35.8 27.6 After additional adjustment for cardio- High (M: $0.95, W: $0.8) 64.2 72.4 metabolic factors, the associations were 6 6 HbA1c (%) 5.66 0.48 5.83 0.69 0.014 slightly attenuated (model 3). The OR

HbA1c (mmol/mol) 39 6 6406 7 0.014 (95% CI) of DSPN was 1.68 (1.19; 2.35) for Diabetes status (%) 0.299 an increment of 0.1 units of WHtR, with No diabetes 45.9 38.6 slightly weaker associations after adjust- Prediabetes§ 40.2 40.9 ment for the cardiometabolic markers Diabetes 14.0 20.5 (model 3). ‡ Hypertension (%) 56.5 65.4 0.452 Interaction analyses did not point 6 6 Total cholesterol mmol/L (mmol/L)¶ 6.06 1.01 5.82 1.12 0.094 toward differences by prediabetes/ Fasting triglycerides (mmol/L)¶ 1.27 (0.97; 1.77) 1.29 (1.05; 1.67) 0.745 diabetes status regarding associations Use of lipid-lowering drugs (%) 23.1 26.8 0.643 between anthropometric measures and eGFR (mL/min/1.73 m2) 80.2 6 13.1 76.9 6 14.3 0.498 DSPN (Supplementary Table 1). Smoking status (%) Never 51.6 54.3 0.002 Mediation Analysis Former 42.7 33.9 Table 3 shows the associations between Current 5.7 11.8 the potential mediators (IL-6, TNF-a, Alcohol intake (%)# Abstainer 31.1 30.7 0.584 CCL7, CXCL9, CXCL10, DNER, CD40, Moderate intake 52.3 51.2 and TNFRSF9) and BMI or waist circum- High intake 16.6 18.1 ference. CCL7, CXCL10, and DNER were Physically active (%)| 62.2 41.7 ,0.001 associated with both anthropometric Myocardial infarction (%) 5.2 7.1 0.850 measurements. None of the other bio- Neurological conditions that might cause markers showed an association with the nerve damage (%) 15.1 20.5 0.084 anthropometric measures. In line with Use of nonsteroidal anti-inflammatory this observation, statistically significant drugs (%)†† 1.0 2.4 0.375 mediation effects for the associations of IL-6 (pg/mL) 1.33 (0.94; 1.96) 1.71 (1.22; 2.52) 0.005 BMI and waist circumference with DSPN TNF-a (pg/mL) 1.90 (1.39; 2.73) 2.09 (1.60; 3.43) 0.062 were observed for these three markers CCL7 (NPX) 1.83 6 0.50 2.09 6 0.64 ,0.001 (CCL7, CXCL10, and DNER) (Table 4). The CXCL9 (NPX) 7.27 6 0.75 7.68 6 0.89 ,0.001 proportion of the total effect for the CXCL10 (NPX) 9.32 6 0.84 9.74 6 1.09 ,0.001 association between BMI and DSPN explained by the mediator was 6.8% (0%; Continued on p. 244 22.2%) for CCL7, 10.5% (1.3%; 32.8%) for 244 Obesity, Inflammation, and Incidence of DSPN Diabetes Care Volume 42, February 2019

Table 1—Continued CONCLUSIONS This prospective study shows a positive Variable No incident DSPN Incident DSPN P association of general and abdominal 6 6 DNER (NPX) 8.32 0.26 8.34 0.26 0.084 obesity with the development of DSPN CD40 (NPX) 10.19 6 0.32 10.31 6 0.34 0.001 in a population-based cohort of older TNFRSF9 (NPX) 6.19 6 0.44 6.41 6 0.47 ,0.001 individuals. The associations were lin- Data are shown as mean 6 SD, median (interquartile range), or percentages. The P values are ear and no interaction by prediabetes/ derived from logistic regression analysis (likelihood ratio tests comparing models with the diabetes status was observed. Three respective variable and age and sex as independent variables to models with age and sex only). All fl analyses were adjusted for age and sex except associations with age (sex-adjusted only) or sex biomarkers of subclinical in ammation (age-adjusted only). Data from this table have been published in a previous report from the KORA (CCL7, CXCL10, and DNER) partially me- study (10). CD, cluster of differentiation; eGFR, estimated glomerular filtration rate; HbA1c, diated the association between anthro- glycated hemoglobin; M, men; W, women. *DSPN defined by MNSI .3. †Socioeconomic status fi – pometric measures and DSPN. based on a score including education, income, and occupation and classi ed as low (1 9 points), fi moderate low (10–12 points), moderate (13–15 points), moderate high (16–19 points), and high This study is the rst to investigate (.19 points). §Prediabetes was defined as IFG and/or IGT. ‡Blood pressure of $140/90 mmHg the association of general and abdominal or antihypertensive medication given that the subjects were aware of as hypertensive. obesity with incident DSPN using a pro- fi ¶Individuals using lipid-lowering drugs excluded (n = 123). #Classi ed as none (0 g/day), moderate spective population-based design. One ($0to,20 g/day for women and $0to,40 g/day for men), or high ($20 g/day for women and $40 g/day for men). |Defined as .1 h sports/week in summer and winter. ††Nonsteroidal of our main findings is that both general anti-inflammatory drugs except acetylsalicylic acid used as platelet aggregation inhibitor. and abdominal obesity were associated with the development of DSPN. Another finding of our study is that we did not observe an interaction effect by CXCL10, and 7.5% (0.4%; 25.6%) for between waist circumference and DSPN diabetes or prediabetes in the associa- DNER. Similar mediation effects were of 6.9% (0.7%; 21.3%) for CCL7, 11.2% tion between obesity and DSPN. There is observed for waist circumference, with (2.8%; 30.2%) for CXCL10, and 9.2% one prospective cohort study that in- proportions explaining the associations (1.3%; 32.4%) for DNER. vestigated the association between BMI and waist circumference and risk of diabetic polyneuropathy among individ- Table 2—ORs and 95% CIs of incident DSPN according to anthropometric uals with type 2 diabetes (7,8). In line measures in KORA F4 (n =513) with our findings, the ADDITION Denmark N Cases Model 1* Model 2† Model 3‡ study also reported a positive association BMI (kg/m2) between both anthropometric measures Normal weight (,25) 120 14 1 1 1 and risk of diabetic polyneuropathy. In Overweight (25 to that cohort including solely subjects ,30) 245 67 2.75 (1.45; 5.22) 3.06 (1.57; 5.97) 3.20 (1.57; 6.53) with type 2 diabetes, the relative risks Obese ($30) 148 46 3.31 (1.70; 6.47) 3.47 (1.72; 7.00) 3.21 (1.50; 6.87) (RRs) of diabetic polyneuropathy for P trend 0.001 0.002 0.017 anthropometric measures were slightly 2 BMI per 5 kg/m 513 127 1.66 (1.27; 2.16) 1.68 (1.28; 2.22) 1.58 (1.17; 2.14) lower (RR [95% CI] 1.14 [1.05; 1.25] for Waist circumference (cm) BMI per 2 kg/m2 and 1.14 [1.06; 1.23] for WHO definition , 5-cm differences in waist circumference, Low (M: 102, fi W: ,88) 249 44 1 1 1 respectively) compared with our ndings High (M: $102, (OR [95% CI] 1.68 [1.28; 2.22] for BMI 2 W: $88) 264 83 2.17 (1.40; 3.36) 2.06 (1.30; 3.26) 1.87 (1.13; 3.08) per 5 kg/m and 1.24 [1.11; 1.39] for IDF definition waist circumference per 5 cm, respec- Low (M: ,94, tively) (7). However, the ADDITION Den- , W: 80) 107 14 1 1 1 mark study used the MNSI questionnaire High (M: $94, to assess neuropathic symptoms rather W: $80) 406 113 2.60 (1.40; 4.84) 2.72 (1.42; 5.19) 2.55 (1.28; 5.05) than the examination portion of the Waist circumference per 5 cm 513 127 1.25 (1.12; 1.39) 1.24 (1.11; 1.39) 1.21 (1.07; 1.36) MNSI to assess neuropathic signs as WHR used in the current study. It is well known fi Low (M: ,0.95, that neuropathic signs (de cits and im- W: ,0.8) 173 35 1 1 1 pairments) rather than symptoms are High (M: $0.95, associated with neuropathy confirmed W: $0.8) 340 92 1.62 (1.01; 2.61) 1.51 (0.91; 2.50) 1.30 (0.76; 2.23) by nerve conduction studies (27). Thus, WHR per 0.1 units 513 127 1.83 (1.24; 2.68) 1.79 (1.20; 2.67) 1.60 (1.03; 2.49) the sensitivity to detect cases with in- WHtR cident DSPN was higher in the KORA WHtR per 0.1 units 513 127 1.73 (1.25; 2.38) 1.68 (1.19; 2.35) 1.51 (1.04; 2.18) F4/FF4 study than in the ADDITION Den- M, men; W, women. *Model 1: adjusted for age and sex. †Model 2: model 1 plus mark study. Investigating patients with adjustment for socioeconomic status, smoking status, alcohol intake, physical activity, and type 1 diabetes, the Diabetes Control and ‡ height (not in models for BMI and WHtR). Model 3: model 2 plus adjustment for HbA1c, Complications Trial/Epidemiology of Dia- hypertension, HDL cholesterol, LDL cholesterol, and triglycerides. betes Interventions and Complications care.diabetesjournals.org Schlesinger and Associates 245

Table 3—Association between potential mediators and anthropometric measures risk factors. In addition, several cross- in KORA F4 (n = 513)* sectional studies reported associations BMI (kg/m2) Waist circumference (cm) between obesity and polyneuropathy in individuals with and without diabetes Potential mediator b 95% CI b 95% CI (2–6); however, findings of these studies IL-6 (pg/mL) 0.086 20.158; 0.329 0.030 20.062; 0.122 are more difficult to interpret because TNF-a (pg/mL) 0.161 20.174; 0.496 0.069 20.059; 0.196 cross-sectional studies are more prone CCL7 (NPX) 0.063 0.001; 0.125 0.025 0.001; 0.048 to recall bias and reverse causality. CXCL9 (NPX) 0.008 20.080; 0.958 0.012 20.021; 0.046 Importantly, general and abdominal CXCL10 (NPX) 0.135 0.032; 0.239 0.059 0.020; 0.098 obesity are modifiable risk factors, and DNER (NPX) 20.043 20.072; 20.014 20.018 20.029; 20.007 thus, the question arises whether weight CD40 (NPX) 0.008 20.029; 0.045 0.003 20.011; 0.017 reduction in individuals with overweight TNFRSF9 (NPX) 0.046 20.005; 0.098 0.014 20.006; 0.034 or obesity by interventions such as life- fi Boldface type indicates statistically significant findings. The association was interpreted as style modi cation or bariatric surgery statistically significant if the 95% CI of the b-estimate did not include the null value. have an influence on the prevention *Associations are adjusted for age, sex, socioeconomic status, smoking status, alcohol intake, of DSPN. So far, there is hardly any ev- physical activity, and height (not in models for BMI). idence available in this context. In the Diabetes Prevention Program (DPP), a post hoc analysis was performed after (DCCT/EDIC) study reported that com- study only adjusted for few confound- a mean follow-up of 15 years. Among the bining the MNSI questionnaire and ex- ers, namely sex, age, diabetes duration, subgroup of participants whose most amination scores did not substantially and trial randomization, and it remains recent HbA1c was $6.5%, representing improve the prediction of clinical poly- unclear whether other factors might in- ;26% of the cohort, the lifestyle inter- neuropathy compared with the separate fluence this association (8). In our re- vention group showed reductions in the use of both tools (28). Moreover, in a port, we investigated several potential onset of neuropathy assessed by the 10-g more recent report from the ADDITION confounders for this association, includ- monofilament compared with the pla- Denmark study, neuropathy was con- ing socioeconomic status, smoking sta- cebo (RR [95% CI] 0.38 [0.19; 0.75]) and firmed by a point-of-care nerve conduc- tus, alcohol intake, physical activity, and metformin (RR [95% CI] 0.39 [0.19; 0.79]) tion study device (DPNCheck). This height (if appropriate) and also car- groups, suggesting that people who con- information was only available at follow- diometabolic factors, such as HbA1c, vert to diabetes lifestyle intervention up after 13 years, but not at baseline (8). hypertension, HDL cholesterol, LDL mayreducetheincidenceofneuropathy Thus, that study is limited in its ability cholesterol, and triglycerides. Our find- (29). A recent retrospective cohort study to assess the predictive value of risk ings show that the association between among patients with type 2 diabetes factors for the development of diabetic general and abdominal obesity still per- demonstrated that bariatric surgery polyneuropathy. Furthermore, that sists even after controlling for important was associated with a 63% lower risk

Table 4—Effect estimates (95% CI) for incident DSPN for 5-unit increase in anthropometric measure and estimated percent mediated by inflammatory biomarker* Direct effect Indirect effect Total effect Percent mediated† BMI (kg/m2) TNF-a 0.013 (0.005; 0.021) 0.000 (0.000; 0.002) 0.014 (0.005; 0.021) 1.6 (22.0; 9.7) IL-6 0.014 (0.005; 0.024) 0.000 (20.004; 0.003) 0.014 (0.005; 0.022) 1.3 (218.6; 20.3) CCL7 0.013 (0.005; 0.020) 0.001 (0.000; 0.004) 0.014 (0.005; 0.022) 6.8 (0.0; 22.2) CXCL9 0.014 (0.005; 0.023) 0.000 (20.003; 0.002) 0.014 (0.006; 0.021) 7.0 (216.4; 14.2) CXCL10 0.012 (0.005; 0.019) 0.002 (0.000; 0.006) 0.014 (0.005; 0.021) 10.5 (1.3; 32.8) DNER 0.015 (0.006; 0.026) 20.001 (20.005; 0.000) 0.014 (0.005; 0.021) 27.5 (20.4; 225.6) CD40 0.014 (0.005; 0.022) 0.000 (20.002; 0.002) 0.014 (0.005; 0.021) 1.4 (29.6; 11.3) TNFRSF9 0.012 (0.005; 0.020) 0.001 (0.000; 0.004) 0.014 (0.006; 0.021) 7.8 (21.7; 21.6) Waist circumference (cm) TNF-a 0.002 (0.000; 0.005) 0.000 (0.000; 0.000) 0.002 (0.000; 0.005) 1.9 (21.1; 8.8) IL-6 0.002 (0.000; 0.006) 0.000 (20.001; 0.000) 0.002 (0.000; 0.005) 9.0 (21.7; 15.2) CCL7 0.002 (0.000; 0.004) 0.000 (0.000; 0.001) 0.002 (0.000; 0.005) 6.9 (0.7; 21.3) CXCL9 0.002 (0.000; 0.005) 0.000 (0.000; 0.001) 0.002 (0.000; 0.005) 2.8 (210.5; 15.6) CXCL10 0.002 (0.000; 0.004) 0.000 (0.000; 0.001) 0.002 (0.000; 0.005) 11.2 (2.8; 30.2) DNER 0.002 (0.000; 0.007) 0.000 (20.001; 0.000) 0.002 (0.000; 0.005) 29.2 (21.3; 232.4) CD40 0.002 (0.000; 0.005) 0.000 (0.000; 0.000) 0.002 (0.000; 0.005) 1.5 (29.7; 13.0) TNFRSF9 0.002 (0.000; 0.005) 0.000 (0.000; 0.001) 0.002 (0.000; 0.005) 5.4 (25.4; 18.3) Data are b (95% CI) unless otherwise indicated. *Models adjusted for age, sex, socioeconomic status, smoking status, alcohol intake, physical activity, and height (not in models for BMI). †Boldface type indicates statistically significant findings. The mediation effect was interpreted as statistically significant if the 95% CI of the percent mediated effect did not include the null value. 246 Obesity, Inflammation, and Incidence of DSPN Diabetes Care Volume 42, February 2019

of DSPN (hazard ratio [95% CI] 0.37 [0.30; effect ranging from 7% to 11%. The effect circumference, whereas data on fat 0.47]) compared with patients who did size of the mediation indicates that mass (absolute or as percentages) not have surgery (30). However, more factors other than subclinical inflam- were only available for KORA FF4 but well-designed clinical and observational mation explain part of the relationship not for KORA F4 (baseline). Fourth, studies are needed to investigate the between obesity and DSPN. First, it is follow-up information was missing for potential of weight loss regarding risk conceivable that a larger cohort would a large number of study participants, reduction of DSPN. have allowed us to identify additional and it has been shown previously that Furthermore, our study extends the inflammation-related biomarkers asso- participants with missing follow-up in- current literature, because we investi- ciated with anthropometric measures formation were older, had a higher BMI gated potential mediators in the relation- (and thus more potential mediators), so and waist circumference, and were ship between obesity and incident DSPN that the current analysis based on three in general less healthy (higher HbA1c, and found three biomarkers of inflam- biomarkers most likely underestimates lower estimated glomerular filtration mation, which partly explained the as- the proportion of the association be- rate, and higher levels of inflammatory sociation. Obesity is a proinflammatory tween obesity and incident DSPN that is biomarkers) compared with participants state (31,32), and we previously iden- explained by subclinical inflammation. with follow-up information (9). Thus, tified multiple proinflammatory cyto- Second, biomarkers of oxidative stress it is likely that a relevant proportion kines, chemokines, and soluble forms of appear as promising candidates be- of those lost to follow-up had developed membrane-bound inflammation-related cause oxidative stress and inflammation DSPN. Finally, our study sample was at an proteins as novel risk factors for DSPN have been described as pathomechan- older age and from one area in South (9,10). In the current study, we identified isms that may integrate metabolic de- Germany, and thus, the generalizability two chemokines (CCL7 and CXCL10) and rangements such as hyperglycemia and of the data to populations at a younger the soluble form of the transmembrane dyslipidemia and link them to the de- age or from different geographic loca- receptor DNER as proteins, which may velopment of cardiometabolic diseases tions or ethnic backgrounds is unknown. explain the mechanism linking obesity (38,39). In line with this, higher levels In conclusion, this study demonstrates andincidentDSPN.CCL7andCXCL10are of methylglyoxal, myeloperoxidase, that general obesity and abdominal both expressed in subcutaneous and and extracellular superoxide dismutase obesity are prospectively associated visceral adipose tissue (33). CCL7 expres- 3 have recently been found associated with the development of DSPN in a sion has primarily been found in adipo- with prevalent and/or incident DSPN population-based cohort. We identified cytes compared with stromal vascular (7,40). a linear relationship between anthro- cells, and its expression is upregulated in The current study has a number of pometric measures and DSPN, and no obesity, potentially due to activation of strengths, including the prospective differences by prediabetes/diabetes sta- the inhibitor of kB kinase-b and c-Jun study design and the population-based tus were observed. Biomarkers of sub- N-terminal kinase pathways (34). CXCL10 sample, which reduced recall bias, se- clinical inflammation partly mediated expression has been reported for mature lection bias, and risk of reverse cau- the association between obesity and human adipocytes (35), and this chemo- sality. The use of different statistical DSPN. Importantly, general and ab- kine can be induced by factors derived approaches enabled us to investigate dominal obesity are modifiable risk from macrophages for which numbers in multivariable adjusted associations be- factors for DSPN, and thus, recommen- adipose tissue are positively correlated tween general and abdominal obesity dations regarding the prevention of with the degree of obesity (36). Inter- with the development of DSPN by ex- DSPN should include the adherence estingly, CCR2 and CXCR3, the receptors ploring linear and nonlinear relations, to a healthy body weight with a normal for CCL7 and CXCL10, respectively, are interactions with prediabetes/diabetes, BMI and waist circumference. expressed on nerves (33), suggesting and mediation effects of biomarkers that a direct effect on neuronal cells of inflammation for these associations. and DSPN risk is biologically plausible. This study has also some limitations. Acknowledgments. The authors thank Ulrike In line with this, we recently found that First, the MNSI allowed the diagnosis of Partke (German Diabetes Center, Dusseldorf,¨ both chemokines had neurotoxic effects possibleDSPN,butfurtherexaminations Germany) for excellent technical assistance. on the SH-SY5Y neuroblastoma cell line, were not possible in this epidemiolog- Funding. The study was funded in part by which is commonly used to assess neu- ical setting. Second,wehadtocombine grants from the German Center for Diabetes Research (DZD; to C.He. and B.T.) and the rotoxicity (10). DNER is a transmembrane participants with prediabetes and di- German Diabetes Association (Deutsche Diabetes- protein that is mainly expressed in brain, abetes for a meaningful interaction Gesellschaft; to C.He.). This work was also sup- adrenal gland, and pituitary tissue. How- analysis. To confirm our results, more ported by the Ministry of Culture and Science ever, it is also found in human adipose studies are needed that investigate the of the State of North Rhine-Westphalia and the tissue–derived mesenchymal stem cells, interaction effect between prediabetes German Federal Ministry of Health as well as by a grant from the German Federal Ministry and its knockdown has been shown to or diabetes and obesity related to risk of of Education and Research to the DZD. The promote human adipose tissue–derived DSPN. Generally, findings from obser- KORA study was initiated and financed by the mesenchymal stem cell adipogenesis vational studies do not necessarily serve Helmholtz Zentrum Munchen,¨ German Re- (37), which may link the quality of ad- as a proof of mechanistic or causal search Center for Environmental Health, which is funded by the German Federal Ministry ipose tissue expansion with DSPN. interactions. Third, we had to rely on of Education and Research and the State of Our mediation analysis for these three anthropometric measurements based Bavaria. Furthermore, KORA research was biomarkers revealed a potential mediation on height, weight, and waist and hip supported within the Munich Center of Health care.diabetesjournals.org Schlesinger and Associates 247

Sciences, Ludwig-Maximilians-Universitat,¨ as part adaptive immunity is associated with incident 26. Tingley D, Yamamoto T, Hirose K, Keele L, of LMUinnovativ. polyneuropathy: KORA F4/FF4 study. Diabetes Imai K. Mediation: R package for causal medi- Duality of Interest. No potential conflicts of 2018;67:2434–2442 ation analysis. J Stat Softw 2014;59:1–38 interest relevant to this article were reported. 11. Rathmann W, Haastert B, Icks A, et al. High 27. Feldman EL, Stevens MJ, Thomas PK, Brown Author Contributions. S.S. designed the study, prevalence of undiagnosed diabetes mellitus in MB, Canal N, Greene DA. A practical two-step contributed data, drafted the analysis plan, in- Southern Germany: target populations for effi- quantitative clinical and electrophysiological as- terpreted data, and wrote the manuscript. cient screening. The KORA survey 2000. Diabe- sessment for the diagnosis and staging of diabetic C.He. designed the study, contributed data, drafted tologia 2003;46:182–189 neuropathy. Diabetes Care 1994;17:1281–1289 the analysis plan, interpreted data, and wrote 12. Rathmann W, Strassburger K, Heier M, et al. 28. Herman WH, Pop-Busui R, Braffett BH, et al.; the manuscript. J.M.K. performed the statistical Incidence of type 2 diabetes in the elderly DCCT/EDIC Research Group. Use of the Michigan analysis and contributed to data interpretation. German population and the effect of clinical neuropathy screening instrument as a measure C.Hu., M.C.-K., W.R., G.J.B., and M.R. contributed and lifestyle risk factors: KORA S4/F4 cohort of distal symmetrical peripheral neuropathy in data and contributed data interpretation. W.K., study. Diabet Med 2009;26:1212–1219 type 1 diabetes: results from the Diabetes Con- M.H., A.P., and C.M. contributed data. B.T. and 13. Herder C, Ouwens DM, Carstensen M, et al. trol and Complications Trial/Epidemiology of D.Z. designed the study, contributed to the Adiponectin may mediate the association be- Diabetes Interventions and Complications. Dia- analysis plan, and contributed and interpreted tween omentin, circulating lipids and insulin bet Med 2012;29:937–944 data. S.S., C.He., J.M.K., C.Hu., M.C.-K., W.R., sensitivity: results from the KORA F4 study. 29. Diabetes Prevention Program Research G.J.B., W.K., M.H., A.P., C.M., M.R., B.T., and Eur J Endocrinol 2015;172:423–432 Group. Long-term effects of lifestyle interven- D.Z. reviewed and edited the manuscript and 14. Martina IS, van Koningsveld R, Schmitz PI, tion or metformin on diabetes development approved of its submission. S.S. is the guarantor van der Meche´ FG, van Doorn PA; European and microvascular complications over 15-year of this work and, as such, had full access to all Inflammatory Neuropathy Cause and Treatment follow-up: the Diabetes Prevention Program of the data in the study and takes responsibility (INCAT) group. Measuring vibration threshold Outcomes Study. Lancet Diabetes Endocrinol for the integrity of the data and the accuracy with a graduated tuning fork in normal aging 2015;3:866–875 of the data analysis. and in patients with polyneuropathy. J Neurol 30. O’Brien R, Johnson E, Haneuse S, et al. Neurosurg Psychiatry 1998;65:743–747 Microvascular outcomes in patients with dia- References 15. World Health Organization. Obesity: pre- betes after bariatric surgery versus usual care: 1. Ziegler D, Papanas N, Vinik AI, Shaw JE. venting and managing the global epidemic. Re- a matched cohort study. Ann Intern Med 2018; Epidemiology of polyneuropathy in diabetes port of a WHO consultation. World Health Organ 169:300–310 and prediabetes. Handb Clin Neurol 2014;126: Tech Rep Ser 2000;894:i–xii, 1–253 31. Skurk T, Alberti-Huber C, Herder C, Hauner 3–22 16. Thorand B, Baumert J, Doring¨ A, et al.; KORA H. Relationship between adipocyte size and 2. ZieglerD,RathmannW,DickhausT,MeisingerC, Group. Sex differences in the relation of body adipokine expression and secretion. J Clin En- Mielck A; KORA Study Group. Prevalence of composition to markers of inflammation. Ath- docrinol Metab 2007;92:1023–1033 polyneuropathy in pre-diabetes and diabetes erosclerosis 2006;184:216–224 32. Hotamisligil GS. Inflammation and metabolic is associated with abdominal obesity and macro- 17. Alberti KG, Zimmet P, Shaw J. Metabolic disorders. Nature 2006;444:860–867 angiopathy: the MONICA/KORA Augsburg Sur- syndrome–a new world-wide definition. A con- 33. The Genotype-Tissue Expression (GTEx) veys S2 and S3. Diabetes Care 2008;31:464–469 sensus statement from the International Di- project [Internet], 2017. Available from https:// 3. Tesfaye S, Chaturvedi N, Eaton SE, et al.; abetes Federation. Diabet Med 2006;23:469– www.gtexportal.org/home/. Accessed 20 No- EURODIAB Prospective Complications Study 480 vember 2018 Group. Vascular risk factors and diabetic neu- 18. Herder C, Bongaerts BW, Rathmann W, et al. 34. Jiao P, Chen Q, Shah S, et al. Obesity-related ropathy. N Engl J Med 2005;352:341–350 Association of subclinical inflammation with upregulation of monocyte chemotactic factors 4. Cheng YJ, Gregg EW, Kahn HS, Williams DE, polyneuropathy in the older population: KORA in adipocytes: involvement of nuclear factor- De Rekeneire N, Narayan KM. Peripheral insen- F4 study. Diabetes Care 2013;36:3663–3670 kappaB and c-Jun NH2-terminal kinase path- sate neuropathy–a tall problem for US adults? 19. Enroth S, Johansson A, Enroth SB, Gyllensten ways. Diabetes 2009;58:104–115 Am J Epidemiol 2006;164:873–880 U. Strong effects of genetic and lifestyle factors 35. Herder C, Hauner H, Kempf K, Kolb H, Skurk T: 5. Ylitalo KR, Sowers M, Heeringa S. Peripheral on biomarker variation and use of personalized Constitutive and regulated expression and se- vascular disease and peripheral neuropathy in cutoffs. Nat Commun 2014;5:4684 cretion of interferon-gamma-inducible protein individuals with cardiometabolic clustering and 20. Nowak C, Sundstrom¨ J, Gustafsson S, et al. 10 (IP-10/CXCL10) in human adipocytes. Int J obesity: National Health and Nutrition Examina- Protein biomarkers for insulin resistance and Obes (Lond) 2007;31:403–410 tion Survey 2001–2004. Diabetes Care 2011;34: type 2 diabetes risk in two large community 36. Bassols J, Ortega FJ, Moreno-Navarrete JM, 1642–1647 cohorts. Diabetes 2016;65:276–284 Peral B, Ricart W, Fernandez-Real JM. Study of 6. Callaghan BC, Xia R, Reynolds E, et al. Associ- 21. Enroth S, Enroth SB, Johansson A,˚ Gyllensten the proinflammatory role of human differentiation between metabolic syndrome components U. Protein profiling reveals consequences of ated omental adipocytes. J Cell Biochem 2009; and polyneuropathy in an obese population. lifestyle choices on predicted biological aging. 107:1107–1117 JAMA Neurol 2016;73:1468–1476 Sci Rep 2015;5:17282 37. Park JR, Jung JW, Seo MS, Kang SK, Lee YS, 7. Andersen ST, Witte DR, Dalsgaard EM, et al. 22. Lind L, Arnl¨ ov¨ J, Lindahl B, Siegbahn A, Kang KS. DNER modulates adipogenesis of hu- Risk factors for incident diabetic polyneuropathy Sundstrom¨ J, Ingelsson E. Use of a proximity man adipose tissue-derived mesenchymal stem in a cohort with screen-detected type 2 diabetes extension assay proteomics chip to discover new cells via regulation of cell proliferation. Cell Prolif followed for 13 years: ADDITION-Denmark. Di- biomarkers for human atherosclerosis. Athero- 2010;43:19–28 abetes Care 2018;41:1068–1075 sclerosis 2015;242:205–210 38. Brownlee M. The pathobiology of diabetic 8. Andersen ST, Witte DR, Andersen H, et al. 23. Nowak C, Carlsson AC, Ostgren¨ CJ, et al. complications: a unifying mechanism. Diabetes Risk-factor trajectories preceding diabetic poly- Multiplex proteomics for prediction of major 2005;54:1615–1625 neuropathy: ADDITION-Denmark. Diabetes Care cardiovascular events in type 2 diabetes. Dia- 39. Shah MS, Brownlee M. Molecular and 2018;41:1955–1962 betologia 2018;61:1748–1757 cellular mechanisms of cardiovascular disor- 9. Herder C, Kannenberg JM, Huth C, et al. 24. Valeri L, Vanderweele TJ. Mediation analysis ders in diabetes. Circ Res 2016;118:1808– Proinflammatory cytokines predict the inci- allowing for exposure-mediator interactions 1829 dence and progression of distal sensorimotor and causal interpretation: theoretical assump- 40. Herder C, Kannenberg JM, Huth C, et al. polyneuropathy: KORA F4/FF4 Study. Diabetes tions and implementation with SAS and SPSS Myeloperoxidase, superoxide dismutase-3, Care 2017;40:569–576 macros. Psychol Methods 2013;18:137–150 cardiometabolic risk factors, and distal sen- 10. Herder C, Kannenberg JM, Carstensen- 25. VanderWeele TJ. Mediation analysis: a prac- sorimotor polyneuropathy: the KORA F4/FF4 Kirberg M, et al. A systemic inflammatory sig- titioner’s guide. Annu Rev Public Health 2016;37: study. Diabetes Metab Res Rev 2018;34: nature reflecting cross talk between innate and 17–32 e3000